Maker Merchant

SEO and AEO for a coffee shop

0 of 39

A field guide for one independent shop

SEO and AEO for a local coffee shop

How to be the cafe that Google, Apple, Bing, and the answer engines can recommend without guessing. Written for the owner, with the technical parts marked for whoever builds the site.

Updated 20 November 2026Checked against official docs0 of 39 steps ticked
For both

Start here

What this guide is, who does which part, and the one rule under the rest of it.

This guide is for one independent coffee shop with a door, a counter, and a website. A second location needs its own page and its own profiles. A national chain needs a different setup. The work here is the same work either way: say the same true things in the places people and answer engines already look.

SEO is search engine optimization. It is how you help Google and Bing understand who you are, where you are, and which pages answer which questions.

AEO is answer engine optimization. It is the same honesty, aimed at products that write an answer: ChatGPT, Google AI Overviews and AI Mode, Gemini, Perplexity, Microsoft Copilot, Claude, and the place cards behind Siri, Apple Maps, and similar assistants. They do not share one dashboard. Some of them have a listing you can claim. Some only read the open web and other people's profiles.

How to use this

Read it in order once. After that, use the contents list to jump. Tick the checklist as you finish a job. The ticks stay in this browser (localStorage). They are not sent anywhere.

Each section is marked for the owner, for the person who builds the site, or for both. You can do the owner sections yourself. Hand the builder sections to whoever edits the site, with the examples. Do not hand them a vague request to "add SEO."

The rule under everything else

Pick one public name, one address, one phone number, and one website URL. Use them everywhere. If the sign says Harbor & Rye Coffee, do not become Harbor & Rye Coffee Portland Best Latte on a profile. Google's representation guidelines tell you to use the real-world name.

Answer engines repeat whatever is most consistent and most specific. A wrong hour, repeated on six sites, becomes the hour they tell a customer.

What this guide will not do

It will not promise a ranking, a number of extra visits, or a fee that "gets you into ChatGPT." Where a company publishes a figure, this guide quotes it and links the page. Where a program does not exist, the guide says so.

Programs change. The facts here were checked on 20 November 2026. Before you pay anyone or hand over a login, open the official link in that section and read it again.

For both

How engines decide

How Google local results, AI Overviews, and ChatGPT choose a cafe, in the words of their own docs.

Two different machines are involved, and they share ingredients.

Google's local results

When someone searches "coffee near me" or "latte on Alberta," Google may show a map pack and ordinary blue links. Google's own help page, Tips to improve your local ranking on Google, says local results are based on three things:

  • Relevance. How well the profile matches what the person is looking for.
  • Distance. How far the place is from the searcher, or from the place they named.
  • Prominence. How well known the place is. Google says prominence is based on information it has, including links, articles, and directories. More reviews and better ratings can improve local ranking. The page does not publish a formula or a weight.

You cannot pick the distance. You can make the profile a clear match for real searches, and you can earn prominence by being a real part of the neighborhood: reviews, a site people link to, press, and consistent listings.

The profile is not the whole result. The website still has to explain the shop in sentences, because many queries never become a map pack. They become a normal result, or an AI answer with links.

Google's AI answers

AI features and your website is Google's site-owner guide for AI Overviews and AI Mode. Google says the usual SEO practices still apply, and that there are no extra requirements and no special schema you must add to appear in those features. A page has to be indexed and allowed to show a snippet. Google also says appearance is not guaranteed.

Google describes both features as showing links, and says they may use "query fan-out": several related searches, across subtopics, while the answer is built. That is why a thin page that only repeats "coffee shop Portland" is a weak source. A page that actually answers "is there a laptop-friendly cafe near Alberta with outlets and oat milk?" can be a useful one.

On that same page, Google lists keeping your Business Profile information up to date as a best practice, next to crawlable pages, internal links, text (not text baked into images), and structured data that matches the visible page.

ChatGPT when someone asks for a place

OpenAI's help page Searching the web with ChatGPT says ChatGPT can search the web and can use location to find local results, including restaurants. An approximate location may come from the IP address. Precise device location is optional and off by default.

The same page says ChatGPT may share location with trusted third-party providers to return nearby business listings, weather, and similar local information. It also says restaurant reservation times, when shown, can come from OpenTable, Resy, or Yelp, with different country coverage. It does not publish a ranked list of every local-data vendor.

For ChatGPT Enterprise and Edu, a separate help page says disassociated search queries may be shared with Bing, and approximate location may be shared with Bing and specialized data providers. That is a documented path for those workspaces. It is not a sentence that says every consumer ChatGPT local card is a Bing Places listing. Treat Bing as one important place to be accurate, because Microsoft has said Bing Places is being tied more closely to Copilot, and because Enterprise search uses Bing. Do not assume it is the only source.

What the other answer engines are doing

Perplexity answers with citations. Its crawler docs say PerplexityBot is how sites get surfaced and linked in Perplexity search. Claude can search the web; Anthropic documents Claude-SearchBot for search indexing and Claude-User for a fetch a person asked for. Copilot sits on Bing. Siri and Apple Maps read the place card you manage in Apple Business. Gemini in the Google apps is not a second Google Business Profile; the profile and the website are still the inputs you control.

None of these products owes you a mention. The work that helps all of them is the same: a correct listing where a listing exists, a page that answers in plain sentences, and other people (reviewers, press, neighbors) repeating the specifics.

For both

What changed in 2025 and 2026

The listing products and the Google rules that actually moved, with dates and links.

A coffee shop does not need a blow-by-blow of every search update. These are the changes, from official pages, that change what you should do. Dates are the dates on those pages.

Listings you manage

3 October 2025. Microsoft launched a rebuilt Bing Places for Business and moved the owner tools to bing.com/forbusiness. The post says existing listings move over, Google import was improved, and Microsoft is investing in deeper ties to Bing Maps and Copilot.

14 April 2026. Apple Business replaced Apple Business Connect, Apple Business Manager, and Apple Business Essentials. Apple's newsroom post says brand and location tools from Business Connect moved into Apple Business, and that existing Business Connect data (claimed locations, place cards, photos, account details) migrated at launch. Place cards still feed Apple Maps and related Apple surfaces. The same announcement said local ads in Maps would come to the U.S. and Canada that summer. Ads are a paid product. The place card is the free listing. Manage it at business.apple.com.

10 November 2026. OpenAI announced Businesses in ChatGPT. Businesses can claim a profile in Business Hub, update details, and see aggregated insights. Business Agents, a separate product, started with eligible ecommerce and online service businesses in the United States. OpenAI's setup article is more specific for a cafe: local stores and restaurants can currently claim only an existing U.S. listing. The full rules are in ChatGPT.

What Google told site owners about AI answers

15 May 2026. Google published a guide on optimizing for generative AI features in Search. The changelog describes it as covering non-commodity content, local and shopping and image content, myths about "AEO/GEO," and a reminder that SEO basics still matter. The standing site-owner page is AI features and your website. The announcement post is here.

7 May 2026. Google stopped showing the FAQ rich result. On 8 May the changelog said the feature would no longer appear, and in June Google removed the FAQ rich-result documentation. Visible questions and answers on your site still help people and anyone quoting the page. They no longer produce that old expandable FAQ treatment in Google Search.

15 June 2026. Google added a note that llms.txt files are not needed for Google Search and will not, by themselves, help or hurt your visibility there. Keeping one for other tools is fine. Details are in Crawlers, sitemaps, and llms.txt.

27 May 2026. Google's changelog says preferred sources can show in AI Mode and AI Overviews. Preferred sources is a reader's choice, not something you buy. If a person picks your site, your content can wear a "preferred" badge for that person, including in those AI features, and only if the site is included in the Search generative AI control. The feature applies to a domain or subdomain, not to a folder such as /blog.

31 August 2026. The Search generative AI control in Search Console finished rolling out worldwide. It covers AI Overviews, AI Mode, and generative AI features in Discover. The default is to include the site. Excluding the site keeps your links and content out of those features. Google says the control is not a ranking signal for the rest of Search, and it is not the training opt-out. Training opt-out for Gemini models is Google-Extended in robots.txt.

A reported data deal, labeled as a report

23 July 2026. Axios reported that Yelp is licensing reviews, photos, and business details to OpenAI, with Yelp branding when that material is used. That is Yelp's account, via Axios, not a sentence on an OpenAI help page. OpenAI's own search help page does name Yelp as a reservation source in the United States and Canada. Keep the Yelp listing accurate either way. Do not treat a news story as a settings screen.

What did not change

Google still tells local businesses to be relevant, nearby, and prominent. Stuffing the city into the business name is still against the representation guidelines. Fake reviews are still a bad idea and against those rules. A menu that exists only as a PDF or a photo is still a poor way to be quoted.

For the owner

Keywords and intent

The phrases a coffee shop should collect, and the pages that should answer them.

You are not hunting for a magic phrase. You are writing down how people already ask for a place like yours, then making sure a real page or a real profile field answers them.

Four kinds of request

KindWhat they wantCoffee-shop examples
NearbyA place they can walk or drive tocoffee near me, cafe open now, espresso on Division
SpecificA drink, food, or constraintoat milk latte, decaf pour over, breakfast burrito, pastry
OccasionA job the room has to dolaptop friendly cafe, quiet cafe for a meeting, first date coffee, dog friendly patio
NamedYou, or a comparisonHarbor & Rye hours, best latte in Sellwood, pour over Portland

"Near me" is not a phrase you paste onto the homepage ten times. Google already knows roughly where the searcher is. Your job is a correct pin, a correct neighborhood in sentences, and a page that is obviously about that shop.

"Best latte in [city]" is usually a comparison. You will not win it by titling the homepage "Best latte." You might be included if reviews, local articles, and your own menu talk about the latte in concrete terms: the coffee, the milk, the size, the price.

Where the phrases actually come from

Collect them before you buy a tool.

  1. Listen at the register for a week. Write the words people use, not the words you wish they used. "Do you have oat?" is a phrase. "Plant-based alternative milk program" is not.
  2. Read the last 50 reviews on Google, Yelp, and Apple. Highlight the nouns: the drink, the street, the wifi, the dog, the noise, the pastry.
  3. Look at messages and Instagram questions. Those are prompts.
  4. After Search Console has data, open the queries report. Add the phrases you did not expect. Ignore one-off nonsense.
  5. Ask three regulars: "If you were telling a friend where we are, what would you say?"

Group the list into the four kinds above. Star the ones you truly offer. If you are not laptop-friendly, do not build a page that says you are. Answer engines repeat the claim, and the next customer will be angry in public.

Give each important phrase a home

Phrase groupWhere it should be answered
Name, neighborhood, hours, how to get thereHomepage, footer, visit page, every profile
Drinks and food, including milk and decafHTML menu, and the menu fields on Google and Apple
Wifi, outlets, noise, dogs, kids, accessibilityVisit page, in sentences, and profile attributes only when the attribute exists and is true
Events, classes, cuppingsEvents page and the event tools on the profiles
Retail bags, if you sell themA real product page with price and what is in the bag

A single homepage can mention the neighborhood and the thing you are known for. It cannot be the only URL. Builders: see Website structure and On-page SEO.

Words that cause trouble

Do not add the city, "best," "cheap," or "near me" to the business name on Google. The guidelines are plain about the real-world name.

Do not create a page for every neighborhood you wish you served. One shop gets one location page. A second page for "coffee in the next town over" with the same text and a swapped city name is a doorway page, and it teaches answer engines the wrong service area.

Do use the neighborhood name the way a local would use it, in the visit page and the profile description. "We are on 18th, two blocks south of Hawthorne" is more useful than "proudly serving the greater metro area."

For the owner

Prompts people actually ask

A monthly prompt bank, and how to read an answer without fooling yourself.

Keyword lists are what people type into a box. Prompts are what they say to a chat. They are longer, they stack constraints, and they often name a neighborhood or a landmark instead of "near me."

A prompt bank to run every month

Replace the bracketed words. Turn location on when the product has a location control, and also try once with it off so you see the difference. Save a screenshot or paste the answer into a note with the date and the product name.

  1. Where should I get coffee near [neighborhood] right now?
  2. What's a good latte in [city], not a chain?
  3. Laptop-friendly cafe in [neighborhood] with outlets.
  4. Cafe near [landmark] with oat milk.
  5. Quiet place for a two-person meeting in [neighborhood], with coffee.
  6. Where can I get a pour-over in [city] this weekend?
  7. Dog-friendly coffee shop with a patio in [neighborhood].
  8. Coffee shop open before 8am near [street].
  9. Is [your shop name] good for working? What are the hours?
  10. Where can I buy a bag of beans from a cafe in [neighborhood]?

Add prompts you heard this month. Drop prompts that describe a shop you are not. If you have no patio, prompt 7 tells you whether engines are inventing one. If they are, fix the listings that say you have a patio.

How to read the answer

For each prompt, write four lines:

  • Did it name you?
  • If it did, were the hours, address, and phone right?
  • Which links or profiles did it show (your site, Google, Yelp, a magazine, a Reddit thread, Apple Maps)?
  • What did it say that is outdated or made up?

Do not "correct" the model by arguing in the chat and expecting that to edit a database. Correct the source it cited, or publish the missing fact on your site in a sentence.

If it never names you, that is not a bug to file. Look at who it did name and what those shops have that you do not: a clearer menu page, more specific reviews, a recent local article, a complete profile. That is the worklist.

What a quotable answer looks like

Answer engines prefer a sentence they can lift. Put the direct answer first, then the detail.

Yes. Harbor & Rye has oat milk for every espresso drink, at no extra charge, including decaf. The cafe is at 418 Example Avenue and is open 7am to 5pm on weekdays.

That shape belongs on the questions page and, when it is true, in the profile description. A paragraph that starts with your founding year and your passion makes the useful fact harder to quote.

Write for a person. The side effect is that a model can quote you without guessing. Google's AI-features guide says important content should be available as text, and that structured data should match the text on the page. A fact that exists only inside a photo of a chalkboard is easy for a customer in the room and hard for everyone else.

For the owner

Google Business Profile

Claim, categories, hours, menu, photos, posts, messaging, and reviews.

Google Business Profile is still the listing that matters most for Google Search and Maps. Sign in at business.google.com with a Google account the shop controls. Do not build the profile inside an agency's private account you cannot recover.

Google's rules for what the profile may say are in Guidelines for representing your business on Google. Read them before you write the name. Local ranking, in Google's words, is relevance, distance, and prominence: Tips to improve your local ranking.

Claim and verify

Search for the shop before you create anything. If a listing exists, claim it. A second listing for the same door splits reviews and confuses Maps. If you find a duplicate, use the profile tools to suggest they are the same place, or contact support from the profile. Do not "fix" a duplicate by editing it into a keyword variant.

Verification is whatever Google offers that profile: phone, email, video, postcard, or another method. Finish it. An unverified profile is not a profile you control. Never buy a verified listing, and never pay a stranger who says they can skip verification.

Name, category, description

Name. The name on the sign and the receipts. No city, no slogan, no "best latte." If the legal entity is "18th Ave Foods LLC" and the public name is different, the public name is what belongs here. The legal name matters for taxes and, separately, for ChatGPT's identity check. It does not belong in the Google name field unless it is also the name customers see.

Category. Follow Manage your business category. Primary category is usually Coffee shop. Add a secondary category only when it is true and useful: Espresso bar, Breakfast restaurant, Bakery. Do not add categories for businesses you are not. Categories are one of the ways Google judges relevance.

Description. Write the way you would tell a neighbor. Neighborhood, what you pour, who it is for, one or two things that are actually distinctive. The form shows the character limit. Do not paste a list of cities. Do not paste the same sentence you used as a meta keyword tag in 2014.

Address, service area, phone, website

A cafe has a storefront. Show the address. Do not hide it and pretend to be a delivery-only service area unless you truly have no customer entrance.

The phone number should be the one a person answers, with the country code. If you use a tracking number, you are taking on a consistency problem: every other profile and the website should still show a number you control, and you should understand that call-tracking swaps are a common way listings drift apart. For a single shop, the simplest accurate setup is one real number everywhere.

The website field should be the canonical homepage, https, no tracking parameters. Manage your local business links covers the link fields. Put UTM tags on ads, not on the permanent profile URL. Answer engines and directories copy that URL.

Hours, including holidays

Enter regular hours. Then enter special hours for the next holidays and for any day you will close for a private event or a short staff week. Do this in the profile, not only on Instagram.

A customer who walks into a locked door will say so in a review. An answer engine that repeated "open" will keep the blame on you. Set the same special hours in Apple Business and Bing when you set them here. Put a one-line note on the visit page too.

If you are closed one day a week, say closed. Do not leave the day blank and hope.

Menu, products, services

Add the menu in the profile if the tool offers it, and link to the HTML menu on your site. The page on your site is the one you fully control and the one answer engines can quote. A PDF, a photo of a board, or an ordering app that hides prices behind a login is a weak menu. See The website.

Name the drinks people ask for, including milk options and decaf, if those are real. If you sell retail bags, add them as products with a plain name and a price that matches the shelf.

Services and attributes (wifi, outdoor seating, wheelchair-accessible entrance, restroom, dogs) are yes/no facts. Mark them only when they are true this week. "Laptop friendly" may not exist as an attribute. Put that sentence on the visit page instead of forcing it into the wrong field.

Photos and video

Use photos you took in the shop. Exterior from the street, so people recognize the door. Interior in the light you actually have. Drinks and food as served. The counter. A short video of the room is useful if the product offers video. Do not upload stock photography, filtered shots of a drink you no longer sell, or a picture of a different branch.

Google has not published a quota in the ranking help page that says "upload N photos a week." What does matter is that the photos match the current shop. Once a month, look at the cover photo and the first eight images the way a stranger would. Replace anything that shows old hours, an old pastry case, or a holiday install you took down.

Get permission before you feature a staff member's face in a photo you use as advertising. Customers in a wide shot of a public cafe are a different situation; do not caption them by name.

Posts

Posts are for something dated: a guest roaster this Friday, a holiday closure, a new pastry that is actually on the menu. They expire. They are not a substitute for the menu or the hours. If you cannot post honestly once or twice a month, skip the feature rather than letting an intern publish "Happy Monday" forever.

Questions

Google has offered a questions feature on some profiles and removed or limited it in others over the years. If your profile still shows questions, answer them in the profile and publish the same answer on your website. If the feature is not there, do not hunt for a workaround. Put the answers on the questions page of the site. That page is the durable copy.

Messaging and booking

Turn on chat or calls only if a person will answer within the day. A messaging button that nobody reads is worse than no button. If you take reservations, connect the booking link you actually honor. A cafe that does not take reservations should not imply that it does.

Reviews

Reviews on Google affect prominence. Google says so on the local ranking page. You do not write them, you do not buy them, and you do not offer a free drink in exchange for a star. The representation guidelines prohibit fake engagement.

Ask in person, after a visit that went well, and ask for a specific comment: the drink, the morning, the corner table. "If you have a minute to review us, it helps if you mention what you ordered" is enough. Hand them a card with the review link. Do not stand over them while they type. Do not filter who sees the link based on a star rating you asked them to give you first. That practice is widely called review gating, and Google has treated it as a violation.

Reply to every review you can, within a few days.

  • Thank a specific detail they mentioned. That reply is public, and it confirms the detail for the next reader and for any system summarizing the reviews.
  • If they report a real problem, say what you changed or invite them to email you. Do not argue, and do not paste their phone number into the reply.
  • If the review is for a different business, report it through the profile tools. Do not reply as if it happened to you.
  • Never offer a refund or a gift publicly in exchange for changing the stars.

You will get unfair reviews. A calm correction ("Our last pour-over is 4:30pm, and we close at 5") is more useful than a speech.

Stars on the map pack come from the profile's reviews. Stars in a normal Google result are a different system, with stricter rules, covered in Structured data. Do not paste your Google rating into homepage schema and expect a star snippet. Google's rule is that self-serving reviews are not eligible.

For the owner

Apple Business

The place card behind Apple Maps and Siri, after Business Connect moved in April 2026.

Siri, Apple Maps, Spotlight, Safari suggestions, and Wallet branding read from the place card you manage in Apple Business. There is no separate "Siri for restaurants" form.

On 14 April 2026 Apple Business became the home for tools that used to live in Apple Business Connect. Apple's announcement says existing Business Connect locations, photos, and account details migrated. If you claimed the shop years ago, sign in at business.apple.com with that Apple Account rather than starting a second claim.

The growth pages are business.apple.com/grow. Apple says there: "Claimed businesses get 30 percent more views, so be sure to create and customize a place card for each of your business's physical locations." That figure is Apple's, on that page. It is not an independent audit, and it is about views of the place card, not about revenue.

What to put on the place card

Apple's own list on that page is the checklist:

  • Business info. Brand name, location, and an About section that matches the Google description in substance, not necessarily word for word.
  • Hours and contact. Phone, email, website, regular hours, and holiday or seasonal hours.
  • Photos and logo. Storefront, products, and the room. Real files. The logo should be the logo, not a photo of a latte.
  • Good to know. The facts Apple gives you fields for, such as payment types, Wi-Fi, and parking. Only mark what is true.

Showcases are the dated layer: a new pastry, a cupping, a holiday drink. Custom actions (order, reserve) should point at a page that works on a phone. If you do not take reservations, do not add a reserve action.

Location insights in Apple Business show how people found and used the card (search, views, taps). Read them monthly next to the Google profile's performance numbers. They will not match. They are different products with different users.

Place card versus ads

Apple also sells ads in Maps in the U.S. and Canada. Ads are optional and labeled as ads. You do not need them to have a correct place card. If someone sells you "Apple Maps SEO" as a monthly retainer, ask whether they are claiming the free place card (which you can do) or buying ads (which is a media purchase). Get the place card right first.

Name and categories

Use the same public name as on Google. Apple has its own category list. Pick the closest true category. If the pin is on the wrong door, move the pin. A pin on the alley or the landlord's office is why Siri sends someone to the wrong entrance.

When you change hours

Change them here the same day you change them on Google and Bing. Apple users will not see the Instagram story.

For the owner

Bing Places

The listing at bing.com/forbusiness, and why it matters for Copilot.

Bing still has its own index, its own map, and its own webmaster tools. It also feeds Microsoft Copilot. There is no separate "Copilot business profile" to claim. The listing is Bing Places, now at bing.com/forbusiness.

Microsoft's 3 October 2025 post says the owner experience moved onto the Bing.com domain, that listings were migrated, and that import from Google was rebuilt to keep name, hours, and contact details. The same post says Microsoft is continuing toward deeper integration with Bing Maps and Copilot. Use that as the reason to care, not as a promise that a finished Bing listing will be quoted by Copilot every time.

Microsoft's support page Add and manage your business listing is the short official instruction: use Bing Places to claim or update the listing that appears in Bing Maps results. Availability of features depends on country.

A practical setup

  1. Sign in with a Microsoft account the shop owns.
  2. Search for the existing listing before you create one.
  3. If Google is already correct, use the import. Then read every field. Imports copy mistakes, and they can lag. Your holiday hours still need a human.
  4. Confirm the pin, the phone, the https website, the categories, and the hours against the one-page fact sheet you wrote at the start.
  5. Add real photos. The recommendation tool Microsoft described will nag you for photos, a website, hours, and a menu or order link if you are a food business. Those are reasonable nags. Follow them when they match the truth.
  6. Verify the website in Bing Webmaster Tools as well. The listing and the site are different logins. You want both. Submit the sitemap there. IndexNow, covered later, is how you tell Bing a page changed.

What Copilot can and cannot be told

You cannot log into Copilot and edit a sentence it said about your shop. You can:

  • Make the Bing listing accurate.
  • Make the website easy for Bing to crawl (Bingbot allowed, important text in HTML, a sitemap).
  • Give Bing a fresh URL via IndexNow or Webmaster Tools when you change hours or the menu.
  • Publish the same facts in the other profiles Copilot might read on the open web.

If Copilot states a wrong hour, correct Bing and the website, then check again in a week. Write down the date. Do not pay for a "Copilot citation package" that cannot show you the Bing listing it edited.

For the owner

Yelp, social, and the other directories

Yelp, Tripadvisor, Foursquare, Facebook, Instagram, TikTok, and Nextdoor.

After Google, Apple, and Bing, claim the sites your neighbors already use. One afternoon of careful claims beats a purchased "500 directory" blast that creates 500 slightly wrong addresses.

Yelp

Claim the business at business.yelp.com. Yelp is where a lot of cafe comparisons still happen, and it is a source answer engines can see.

On 23 July 2026, Axios reported a licensing deal: Yelp reviews, ratings, photos, and business details surfacing in ChatGPT, with Yelp's branding. OpenAI's help page, separately, names Yelp for reservation availability in the U.S. and Canada. Whether or not you care about ChatGPT, a neglected Yelp page with 2019 hours is a public mistake.

Fill categories, hours, and the "about" text. Add photos you took. Respond to reviews the same way you do on Google: specific, calm, no incentives. Yelp has its own rules about asking for reviews. Read the current policy in the Yelp for Business help before you print a table card. Do not buy reviews.

Yelp will sell you ads. Ads are optional. The free claimed page is the part that needs to be true.

Tripadvisor

If travelers find you, claim the listing through Tripadvisor's owner tools (tripadvisor.com/Owners). Many neighborhood cafes are barely present there, and that is fine. If a listing exists and the hours are wrong, fix it or you will keep hosting guests who arrived at the wrong time. Same name, same address, same phone.

Foursquare

Foursquare still maintains a large places database that other apps have used for years. Update the listing at foursquare.com so the name, pin, hours, and categories are right.

OpenAI's public help pages cited in this guide do not name Foursquare as a ChatGPT local-data provider. Older announcements in the trade press have claimed a places-data relationship. Unless OpenAI or Foursquare's own current docs say so on a page you can open, do not build a plan on that claim. Do keep the listing clean, because other products still read places data, and a wrong pin spreads.

Facebook and Instagram

The Facebook Page is a listing: name, address, phone, hours, website, category. Claim it in the account the shop owns. Meta's business home is facebook.com/business. Put the Page and Instagram in the hands of a person who still works there. Former managers are a common reason nobody can answer a review.

Use the same public name. Link the website. Update holiday hours here too if you tell customers to "check Facebook," because some of them will, and some answer engines will quote the Page.

Instagram is not a structured listings platform in the way Google is. Use a professional account, the same name, a bio that states the neighborhood, and a link to the site. Post pictures of the real shop. The grid is proof. It is not the menu of record. If the only menu is an Instagram highlight, move it to HTML and point the bio at that page.

TikTok

TikTok does not give a cafe a Google-style listing with hours that propagate to maps. Use it if you like making short videos of the room and the drinks, and put the website and the neighborhood in the profile. Do not expect a caption to correct a wrong hour on Google. Do not put the only copy of your holiday closure on TikTok.

Nextdoor

Neighbors recommend each other on Nextdoor. Claim the business page via business.nextdoor.com. Keep the address and hours aligned. Recommendations there are word of mouth. Answer questions if the tool allows it, in the same factual tone as a Google reply. Do not spam the feed with "come in today" posts. One useful post when something real happens (a new breakfast sandwich, a closure) is plenty.

What to skip

Skip any email that offers to submit you to hundreds of directories for a fee, especially if they will not name the directories and show you the login. Skip anything that wants to "register your business with ChatGPT, Gemini, and Alexa" as a bundle. Those products do not share one form. The real paths are in the next sections.

For both

NAP and citations

One name, address, and phone, plus the aggregators people still mention.

NAP means name, address, and phone. Citations are the mentions of that trio on other sites. The goal is not the longest list. The goal is that a person, or a model, who checks three sources finds the same door.

The fact sheet

Write this down before you touch a profile. Put it in a note the owner and the web person both have.

FieldYour valueRule
Public nameAs on the signNo keywords, no city, no LLC unless the LLC is the sign
Legal nameAs on the registrationFor banks, tax, and ChatGPT Persona. Not for the Google name field
AddressAs the post office and the door agreeOne format. Spell out or abbreviate "Street," but pick one and keep it
PinThe entrance customers useCheck it on Google, Apple, and Bing, not only the street centroid
PhoneOne main numberCountry code included when a form asks for international format
WebsiteOne https URLNo UTM tags on the permanent profiles
HoursRegular week, plus a list of exceptionsUpdate all profiles the day an exception is decided

Then make a sheet with one row per site: Google, Apple, Bing, Yelp, Tripadvisor, Foursquare, Facebook, Instagram, Nextdoor, your own footer, your ordering app, your delivery apps. Columns: URL of the listing, login owner, last checked, mismatch found. The checklist item is to start that sheet. The monthly job is to re-read it.

Delivery apps and ordering apps

DoorDash, Uber Eats, Toast, Square, and similar tools publish a name and an address too. They are citations. When you change the phone number, change it there or the old number lives on. If a marketplace listing is closed but still public, get it marked closed. "Permanently closed" on a delivery app is a painful way to lose a Saturday.

Data aggregators

Some companies buy and resell business facts to directories, navigation apps, and banks. Two names you will still hear:

  • Data Axle (data-axle.com) compiles and licenses business data. A single cafe usually does not "sign up for Data Axle" the way you claim Google. Wrong facts that started there get corrected by fixing the major profiles those datasets also ingest, or by using whatever correction form Data Axle currently publishes. Open their site and use their contact or update path. Do not email your login to a middleman who says they have a special feed.
  • Localeze was Neustar's listing-distribution product. The brand has changed hands and names over the years (Neustar, then related TransUnion products). There is not one stable "Localeze login" this guide can honestly give you as the 2026 button to press. If a vendor tells you they will "push Localeze," ask for the current correction URL on the aggregator's own domain and for a screenshot of your record before and after. If they cannot show that, they are selling a blast, not a correction.

Foursquare's places data, covered above, sits in this same family of "other apps may be reading it."

For one shop, the order of operations is: Google, Apple, Bing, your website, Yelp, Facebook, the delivery apps you actually use, then Foursquare and anyone else a customer can still find in a search for your name. Aggregators matter when a directory you do not control keeps republishing an old address. Fix the source you can document. Do not buy a package whose only deliverable is a PDF that says "submitted."

How to find the stragglers

Search Google and Bing for the phone number in quotes, then for the old phone number, then for the name plus city. Open every result that looks like a listing. Correct it or request removal if the business is you and the data is wrong. Search the old address the same way if you have moved. A move is a new pin plus an explicit "moved" on the old listing, not an overwrite that leaves reviews behind on the wrong door. Google's help for moves is inside the Business Profile help center; follow the move flow rather than editing the address into a new site and hoping.

Consistency without being robotic

The phone number and the street address should match character for character wherever a form asks for them. The sentences can vary. Apple's About and Google's description should not contradict each other. They do not need to be copy-pasted. Contradiction looks like: Google says you close at 5 and Apple says 6, or the website says "no laptops on weekends" and Yelp says "great place to work all day."

For both

ChatGPT

Business Hub as it actually works for a cafe in late 2026, and what is not available.

This is the section to read twice, because a lot of agencies are selling a "ChatGPT listing" that does not match what OpenAI actually offers. The two official pages to keep open are:

The policies for profiles are Businesses in ChatGPT. OpenAI also describes the product at openai.com/business/businesses-in-chatgpt.

What exists

Business Hub is the place inside ChatGPT where an authorized person claims a business, updates details, and, after verification, sees aggregated insights. OpenAI's announcement says any eligible business where ChatGPT is available can claim through Business Hub. You do not need an ads account or a ChatGPT Enterprise plan to claim an eligible business. Enterprise customers and advertisers can also claim through their OpenAI contact. Shopify merchants can connect a store from Business Hub or from the Shopify admin.

Business Agents are a separate, optional product. The announcement and the help center both limit them, at launch, to eligible ecommerce and online service businesses in the United States, with a website, that also meet the Business Agent Policy. The help center is explicit that some businesses can manage a profile and still not be allowed to create an agent.

ChatGPT Ads and sponsored agents exist, and OpenAI says sponsored placements are labeled and separate from ChatGPT's answers. The help center says advertising does not influence ChatGPT's independent answers. You do not have to advertise to claim a business.

Product feeds let a retailer share price and availability (CSV upload or a hosted file, or the Shopify catalog). Connecting a feed does not publish an agent and does not guarantee a recommendation. A cafe that only sells drinks over the counter is not the case this feed is built for. A cafe that also sells bags of coffee online might use a feed for that online catalog, on the profile that matches that website. Read the current Commerce policies before you upload one.

Apps and plugins are a developer product, separate from Business Hub. The help center says that if you also have a plugin, the business agent's chat does not use that plugin's data. A neighborhood cafe does not need to build an app to be mentioned. Building one is a software project, not a listing.

What a coffee shop can claim

The help center says this plainly: if you cannot find the business, an eligible online business may see an option to add one. Local stores and restaurants can currently claim only an existing U.S. listing.

Read that again before you pay someone to "create your ChatGPT profile."

  • If you are a U.S. cafe and ChatGPT already has a listing, you can try to claim that listing.
  • If you are a U.S. cafe and there is no listing, the help center does not offer a way to add a new local restaurant listing. Do not invent one by creating a fake online brand.
  • If you are outside the United States, eligibility "depends on the business's location and type." The help center does not promise a local-restaurant claim outside the U.S. Open the Business Hub search and see what your account is offered. If the search returns nothing and no "add" path for your type, you are done with claiming. Spend the time on the website and the other profiles.
  • A business agent is unlikely for a counter-service cafe. OpenAI describes agents for ecommerce and online services. The help center says the current agent can explain products and policies and link to your site. It cannot complete a purchase, make a booking, or look up a customer's order. If the product later opens to restaurants, the help center will say so. Until then, do not buy an "agent build" for a shop the policy does not include.

OpenAI's announcement says they plan to expand to more markets and categories over time. "Plan to" is not "you can do this today."

How to claim, if a listing is there

The help center's steps:

  1. Sign in to ChatGPT and open Business Hub from the sidebar. Search the business name.
  2. Select the listing that matches your door. Confirm the ChatGPT account that will manage it. Use an account the business owns, not a freelancer's personal login.
  3. Verify with one of the methods that listing offers. The help center names phone (SMS or an automated call), business email, website domain (a DNS TXT record), and Shopify. Methods vary by listing. If the phone on the listing is a dead line, look for domain verification instead of changing the public phone number just to receive a code.
  4. Wait for OpenAI to approve the claim. They email a decision or a request for more information. You can cancel a claim that is still in review.
  5. Complete identity verification with Persona. The help center is strict: claim verification and Persona are different. Persona approval is required before you can publish changes, view insights, connect a product feed, or set up an agent. You may be asked for a government ID, business registration, and proof you represent the business. Use the legal name on the registration. If the shop name is a trade name, put the legal name in the legal-name field and the trade name in the trade-name field if you are given one. The registered address and any tax or registration number have to match the documents.
  6. The blue check, the help center says, appears after Persona is approved, not merely because you started a claim. The shorter marketing page says a blue check shows the profile has been claimed. If you have claimed and you still have no check, look for "Complete verification with Persona" rather than assuming the marketing sentence has already happened.

If the listing is locked, the help center says to use Contact support. If someone else claimed it, open the profile, use the more menu, and choose Dispute ownership, then verify with Persona. Disputing or claiming, OpenAI says, does not itself change how visible the profile is.

Publishing a new profile that was created through Persona can go live when OpenAI approves it, with no extra publish step. Do not start that flow until the facts are ready. OpenAI says support cannot delete or hide a published profile as part of a setup request. Drafts that were never published can be deleted from the draft's menu.

What you can edit, and what ChatGPT will still do

After claim and Persona are approved, the help center says you can update the website, description, product or service category, logo, delivery policy, and return policy, and you can add a cover photo. OpenAI reviews changes. They can take time to show up in a listing or an answer. The same page says ChatGPT may also use other sources when it answers. A saved edit is not a switch that rewrites every chat.

That is why the website, Google profile, Apple place card, Bing listing, and Yelp page still matter after you claim. Claiming gives you a channel for facts. It does not give you authorship of ChatGPT's answers.

The help center's FAQ answers the hope directly: claiming does not guarantee that ChatGPT will recommend you more often. It lets you manage the information you provide. It does not give you direct control over independent recommendations.

Insights, if you are allowed to see them

After both approvals, Insights show aggregate activity over the past 28 days.

  • Mentions. For an online business, a mention counts when ChatGPT references the website. The brand name alone does not count. For a local business, a mention counts when ChatGPT shows a result for that location, even if the person never opens the listing or the site.
  • Website clicks. Clicks from ChatGPT through to the site.
  • Discovery topics. Themes across answers that mentioned or cited you. They are not the prompts themselves. OpenAI says businesses do not receive conversations, including chats with a business agent.

A blank number is not proof of zero. The help center says a blank can mean the activity is under a threshold, still collecting, or not reported. OpenAI's announcement says insights are aggregated and only available when there is enough activity to protect privacy.

Where local facts in ChatGPT come from

Use OpenAI's pages, and label everything else as someone else's report.

From Searching the web with ChatGPT:

  • ChatGPT can search the web and show links.
  • It may use an approximate location from the IP address. Precise device location is off unless the person turns it on.
  • It may share location with trusted third-party providers for nearby business listings and similar local results.
  • Reservation times, when shown, can come from OpenTable (described as global), Resy (United States), or Yelp (United States and Canada). Coverage depends on the restaurant.

From the Enterprise and Edu search help: disassociated queries may be shared with Bing, and approximate location may be shared with Bing and other providers. The IP address itself is not what they say they share.

From the 10 November 2026 announcement: Shopify merchants' eligible products can already appear through Shopify Catalog, with or without the ChatGPT app in Shopify. Business Hub is additional.

From Axios, 23 July 2026: Yelp said it is licensing reviews, photos, and business details to OpenAI. Treat that as Yelp's statement to a reporter. It matches the reason to keep Yelp current. It is not a setting inside Business Hub.

This guide did not find an OpenAI help page that tells a cafe to claim Foursquare, Tripadvisor, or a Bing listing in order to edit ChatGPT. Those profiles are still worth fixing for their own users and because answer engines read the open web. They are not a documented "submit to ChatGPT" button.

What to do this month

  1. In ChatGPT, open Business Hub and search your name. Screenshot the result.
  2. If a U.S. listing matches your door, claim it with the method you can actually complete, then do Persona with the legal documents.
  3. If there is no listing and you are a local cafe, stop. Do not pay to "add" one.
  4. Make sure OAI-SearchBot is allowed to crawl the site if you want pages eligible for ChatGPT search. Blocking GPTBot is a training choice and is separate. See Crawlers.
  5. Keep the website and the Google, Apple, Bing, and Yelp facts aligned, because ChatGPT is allowed to use other sources even after you claim.

Teammates can be invited in Business Hub, on their own ChatGPT accounts. The help center warns that invitations can depend on the workspace used to claim. Do not forward a personal claim email as a way of sharing the keys.

For both

Gemini and Google AI answers

AI Overviews, AI Mode, the Search Console control, and why there is no Gemini listing.

There is no Gemini Business Profile and no Gemini merchant program for a cafe. Gemini and Google's AI answers use Google's existing systems. The work is the Business Profile, the website, and a conscious choice about whether your pages may be used in Search's generative features.

AI Overviews and AI Mode

Google's page AI features and your website says:

  • Best practices for SEO still apply. There are no additional requirements to appear in AI Overviews or AI Mode, and no special optimization you must do.
  • To be eligible as a supporting link, a page must be indexed and eligible to show a snippet in Google Search. There are no further technical requirements.
  • Meeting the requirements does not mean Google will show the page. Indexing and serving are not guaranteed.
  • You do not need new machine-readable AI files, AI text files, or special schema to appear in these features. There is no special schema.org type to add for them.
  • Structured data you do use should match the visible text.
  • Important content should be text, not only an image. Images and video should support that text.
  • Business Profile and Merchant Center information should be up to date. A cafe cares about the profile. Merchant Center matters if you sell products in Google's shopping systems, which is optional for a counter-service shop.
  • The control for crawling is the normal Googlebot rule in robots.txt. To limit what Search shows as a snippet, Google points at the usual snippet controls (nosnippet, data-nosnippet, max-snippet, noindex).

Google's May 2026 announcement of the generative-AI guide describes the same stance: non-commodity content, local and image content, and skepticism about shortcut "AEO" tactics. The changelog on 15 May 2026 also says spam policies apply to generative AI responses in Search. Do not publish doorway pages or fake reviews because "AI likes fresh content."

The Search Console switch

The Search generative AI control lives under Settings in Search Console. Google's help page says that as of 31 August 2026 it is available worldwide. It covers AI Overviews, AI Mode, and generative AI features in Discover.

  • Include (the default). Your links and content may appear, including as grounding for those answers, and you can get impressions and traffic from them.
  • Exclude. Your links and content are kept out of those features, including grounding. Other sites can still publish similar information. The rest of Search is not supposed to treat this as a ranking signal. It does not remove you from ordinary results. It does not, by itself, opt you out of model training.
  • Changes generally take a few days. Google says content is excluded within 1–2 days after the control is live, with some lag for caches.

For a cafe that wants to be recommended, leave the control on Include. Verify the site in Search Console so you are the one who can see the setting. An agency that excludes you "to protect the content" has just removed you from the AI answers. Ask them to show you the setting.

Preferred sources

Preferred sources lets a reader mark a domain or subdomain they want emphasized. Google says that when someone has selected your site, your content is more likely to appear in Top Stories with a badge, and that in AI Mode and AI Overviews it can be highlighted with a preferred badge for that person. You must be included in the Search generative AI control to be eligible for the badge in those AI features. A subdirectory is not eligible. A whole domain is.

This is a reader's choice. Google offers a button and a deeplink you can put on your site. It is reasonable for a local newspaper. It is a weak ask for a cafe ("set us as a preferred source" on a loyalty card will confuse people). Do not pay to be a preferred source. That is not how the feature works. If you have a content site with a real audience, the docs explain the button. The cafe itself should spend that hour on the menu page.

Google-Extended is not the AI Overview switch

Google-Extended is a robots.txt token for Gemini apps and Vertex AI Gemini, documented with Google's common crawlers. The Search Console help page says the generative-AI control does not affect training, and that limiting training of the models used for those Search features is what Google-Extended is for. Blocking Google-Extended does not remove you from ordinary Google Search. Blocking Googlebot does, including the AI features that depend on Search.

Do not let a plugin add Disallow: / for Googlebot because a blog post said to "block AI."

Gemini the app

People also open the Gemini app and ask for a cafe. Google has not published a separate claim flow for that app in the pages above. Signed-in Gemini can use Google services, including Maps, under the account's own settings. You do not manage that from a Gemini dashboard. You manage the Business Profile and the website those answers can draw on.

If a Gemini answer is wrong, correct the profile and the page, then try the prompt again later and log it. There is no "edit this Gemini card" button comparable to Business Hub.

For both

Perplexity

Citations, the merchant program for large retailers, and the publisher program you are not joining.

Perplexity answers with citations. For a cafe, that is the whole game: be a page or a profile worth citing. There is a merchant program, and there is a publisher program. Neither one is a local place card.

The merchant program

Perplexity announced a Merchant Program in Shop like a Pro. The post describes a free program for large retailers to share product specs so Perplexity can index them, with a path into checkout, API access, and a dashboard of shopping trends. The post says the program is separate from sponsored questions. It tells large retailers to fill out a signup form.

That is a product-catalog program. It is not "claim your cafe the way you claim Google." A single shop that sells espresso over the counter is not the business that post describes. If you also sell packaged coffee at a scale where you have a real product feed, you can read the current form and see whether you are in scope. Do not pay a third party to "submit the cafe to the Perplexity merchant program" as if that created a map pin. If Perplexity's page has changed since that post, the page itself wins. Open it before you act.

The publisher program

The Publishers' Program is for media organizations. Partners have included newsrooms, not storefronts. The contact on Perplexity's later expansion post is publishers@perplexity.ai. A cafe does not join it. A local magazine that writes about you might. Being written about is the relevant version of this program. See Being cited.

What actually gets you cited

Perplexity's crawler documentation defines two agents:

  • PerplexityBot surfaces and links sites in Perplexity search. It is not, they say, used to crawl content for foundation-model training. They recommend allowing it, and they publish IP ranges.
  • Perplexity-User fetches a page when a person asks a question. They say that because a person requested it, this fetcher generally ignores robots.txt.

Allow PerplexityBot if you want to be eligible to be linked. Put the facts in HTML. Use the same name and address as everywhere else, so the citation and the Google profile do not argue.

Sponsored or related questions, when Perplexity runs them, are ads. Ask to see the label. An ad is not an organic citation, and you do not need one to be a normal source.

There is no hours field on Perplexity for you to edit. If Perplexity repeats a wrong hour, fix the page and the profiles it can cite, then check the prompt again later.

For both

Copilot

No separate Copilot profile. The path is Bing.

Microsoft Copilot does not have a separate business-profile product for a coffee shop. The official local listing is Bing Places, at bing.com/forbusiness. Microsoft's October 2025 Bing Places announcement says they are investing in deeper integration with Bing Maps and Copilot. That is the documented connection. It is not a Copilot dashboard where you edit a paragraph.

The real path

  1. Claim and correct the Bing listing. Steps are in Bing Places.
  2. Verify the site in Bing Webmaster Tools and submit a sitemap.
  3. Allow Bingbot in robots.txt. A site that blocks Bing will struggle to be a Copilot source.
  4. When the menu or the hours change, tell Bing. IndexNow is the protocol Bing documents for that, and it is covered in Crawlers, sitemaps, and llms.txt. Google does not use IndexNow.
  5. Keep the page in HTML. Copilot cannot quote a screenshot of your chalkboard any better than anyone else.

If a vendor sells "Copilot optimization" as a monthly add-on, ask them to show the Bing listing and the Webmaster Tools property. If the deliverable is a PDF of prompts and no listing access, you can run the prompt bank yourself.

For both

Claude

No business listing. Which Anthropic crawlers to allow if you want to be fetched.

Anthropic has not published a Claude business profile, a merchant catalog, or a "claim your cafe" flow comparable to Google Business Profile or ChatGPT Business Hub. Claude answers from what it can retrieve on the web, plus the model's own knowledge. You influence the first of those. You do not get a dashboard for the second.

The crawlers Anthropic documents

Anthropic's privacy article Does Anthropic crawl data from the web, and how can site owners block the crawler? names three robots. Each rule is separate, and you repeat it on every subdomain you care about.

RobotAnthropic's descriptionIf you block it
ClaudeBotCollects public pages that may contribute to trainingFuture material should be left out of training sets
Claude-UserFetches a page when a person asks Claude a questionAnthropic says it cannot retrieve the page for that request, which may reduce visibility in user-directed search
Claude-SearchBotCrawls to improve search qualityAnthropic says it will not index the content for search, which may reduce visibility and accuracy in search answers

Anthropic says these bots honor robots.txt, including Disallow, and that they also honor Crawl-delay. The article says blocking by IP is a poor opt-out, because the bots still need to read robots.txt. As of 30 October 2026, the same article says Anthropic has started to honor a machine-readable "no AI training" preference based on draft IETF work, with ClaudeBot's explicit robots.txt rule taking priority when both exist.

What a cafe should allow

If you want Claude's search to be able to index and fetch your menu, allow Claude-SearchBot and Claude-User. Blocking ClaudeBot is a training decision. It is reasonable to make either way, and it is not how you "rank in Claude." The article ties ClaudeBot to training datasets, and Claude-SearchBot to search.

There is nothing else to claim. Publish the page, keep the facts consistent with the profiles, and use the prompt bank. If Claude names a closed shop or a wrong hour, fix the source it linked. If it linked nothing and invented a detail, the invention is the model's. Your defense is a clear, crawlable page that states the fact in the first sentences, so the next retrieval has something solid to pick up.

For the builder

The website

The pages a single shop needs, and why the menu has to be HTML.

The website is the only source you fully control. Profiles can lag, and answer engines are allowed to keep using other pages. If the site is clear, the next crawl has a chance to catch up.

Build it for a person on a phone, standing outside, who wants to know if you are open and whether you have oat milk. Everything else is extra.

Pages a single shop needs

PageJobWhat "done" looks like
HomeSay who you are, where, and why someone would stopName, neighborhood, today's kind of offer (drinks, food), a way to see hours without hunting
MenuThe real menuHTML, with item names, short descriptions, and prices. Not a PDF. Not a photo of the board
VisitGet here and know what the room is likeAddress, map link, transit or parking in a sentence, hours, phone, wifi, outlets, accessibility, dogs, noise
AboutWho roasts, who works the bar, what the room is forSpecifics a journalist could quote. Not a mission statement that could belong to any cafe
QuestionsThe ten things people askEach answer starts with the answer
EventsCuppings, pop-ups, live setsDate, time, whether it is ticketed, and what happens if it is cancelled
Order or beansOnly if you really sell that wayA working order link, or a product page for bags with price and origin

Link these from the header or the footer on every page. Google's AI-features guide lists internal links as a basic practice: pages should be easy to find from other pages on the site. A menu that is only in an Instagram bio is not internally linked, because it is not on the site.

One shop, one location page. If you open a second door, give it its own URL, its own hours, its own photos, and its own profiles. Do not spin up /coffee-sellwood, /coffee-hawthorne, and /coffee-division that repeat the same 200 words with a new neighborhood. You do not serve all of those as if they were branches.

The menu has to be HTML

A PDF is a file somebody downloaded last spring. A JPEG of the chalkboard is a picture. Both are hard to quote, hard to update, and awkward on a phone. Google's guidance for AI features says important content should be available as text.

Put each drink in the HTML as text: name, a short description, price, and the milk and decaf notes if they change the offer. Mark items sold out in the HTML when you can, or remove them. An answer engine that memorizes "seasonal strawberry latte" in January is repeating your outdated page.

You can still photograph the board for the people in the room. The photo supports the text. It does not replace it. Builders: the structured menu in Structured data has to match this HTML. Do not mark up a honey latte you do not sell.

If an ordering app is the only place prices exist, also put the standard menu on your domain. Apps change, log people out, and often block crawlers. Link to the app from the menu. Do not let the app be the menu.

The visit page

Write the practical facts in sentences:

  • The entrance is on Example Avenue, not the courtyard.
  • There are two steps at the door, or there are none. Say which.
  • The restroom is down the hall, or there is no public restroom.
  • Laptops are welcome before noon and not after, if that is the rule. If laptops are always fine, say that, and say whether you have outlets.
  • Dogs are welcome on the patio and not inside, if that is the rule.
  • The last espresso is 30 minutes before close, if that is the rule.

These sentences are what "laptop friendly cafe" and "wheelchair accessible" searches need. Profile attributes are checkboxes. The sentences are the evidence.

On every page, in text: public name, street address, locality, region, postal code, phone as a tel: link, hours in a short form, and links to the menu, visit, questions, and events pages. Match the fact sheet. This footer is what people and crawlers see even when they land on a blog post.

Ordering, reservations, and maps

Use one map link that opens the correct pin (Google Maps and Apple Maps links can both be offered). Do not embed a map and also type a conflicting address under it.

If you use OpenTable, Resy, or Yelp reservations, only link the system you actually check. ChatGPT's search help says those three can supply reservation times, with different country coverage. A dead "Reserve" button trains customers to distrust the next true sentence.

What the builder should not ship

  • A homepage made of one giant image with the hours baked into the pixels.
  • A menu in a third-party popup that loads nothing in the HTML.
  • A different phone number in the footer "for tracking."
  • City pages you do not have shops in.
  • A blog that publishes AI paragraphs about the history of coffee with no local fact. Google's spam policies apply to ordinary results and, as of the May 2026 changelog, to generative AI responses in Search. Thin pages are a liability, not an AEO strategy.
For the builder

Speed, mobile, and access

Core Web Vitals, phones, and the accessibility work that overlaps with being quotable.

Speed and clarity are customer service. They are also how Google describes a good page experience. None of this requires a particular framework. A small, mostly static site is a good fit for a cafe.

Core Web Vitals

Google documents the metrics at Core Web Vitals and the thresholds people mean by "good" are the ones on web.dev's Web Vitals page:

MetricWhat the customer feels"Good" on web.dev
LCP (Largest Contentful Paint)When the main photo or heading actually shows upWithin 2.5 seconds
INP (Interaction to Next Paint)Whether the page reacts when they tap the menuWithin 200 milliseconds
CLS (Cumulative Layout Shift)Whether the hours jump down because a banner loaded late0.1 or less

INP replaced First Input Delay as a Core Web Vital. If a plugin or a blog post still tells you to chase FID, it is out of date.

Run PageSpeed Insights on the homepage and the menu, mobile first. Read the field data if it exists, and the lab suggestions either way. Fix the top cause, not the score as a hobby. For a cafe site the usual causes are:

  • A full-bleed photo that is 4000 pixels wide. Resize it, serve a modern format, and set width and height so the layout does not jump.
  • A slideshow, a chat widget, and three analytics tags on a four-page site. Remove the ones nobody uses.
  • A menu embedded from a service that pulls in a large script before the prices appear. Put the prices in HTML.
  • Fonts that block the first paint. A small number of weights, with a fallback, is enough. The page should be readable if the web font fails.

Re-test after the change. Keep a note of the before and after so the next redesign does not undo it.

Mobile

Google uses mobile-first indexing. The mobile page is the page. Hours cannot be hidden behind a desktop-only hover. Tap targets for phone, menu, and directions should be large enough to hit on a sidewalk. Text should not require pinching. Test on a real phone, not only a narrow browser window: call the phone number, open the map, scroll the menu.

The same URL should serve the same facts on mobile and desktop. A separate m. site that lags behind is a classic way to publish two sets of hours.

Accessibility

WCAG is the standard to hand a builder. The overlap with SEO is practical, not mystical:

  • Real text, not text inside images.
  • Headings in order, so a page has one topic and visible sections.
  • Alternative text that describes the photo ("front door on Example Avenue"), not a keyword list ("best latte Portland cafe").
  • Contrast that can be read in sunlight. Pale grey on a photograph fails customers and fails basic contrast checks.
  • A "skip to content" link and a keyboard path through the menu. If a modal traps focus, fix the modal.
  • Do not autoplay music.

State the access facts of the building on the visit page: steps, restroom, stroller space. That is content, not code. An accessible website that hides a flight of stairs is how you disappoint someone after they have already traveled.

Page experience, without folklore

Google's page experience documentation is the one to follow. Do not install a "SEO score" plugin and chase it to 100. Do not add a CDN, an image service, and a tag manager to a four-page site because a checklist said enterprise shops have them. Make the menu appear quickly, stay still while loading, and respond to a tap.

For the builder

On-page SEO

Titles, descriptions, headings, internal links, images, and a consistent entity.

On-page SEO is how each URL explains itself. Do it in plain language. The person reading is the customer. The side effect is that a search engine and an answer engine can tell the pages apart.

Titles and descriptions

Every page needs its own <title> and its own meta description. The title is the label in the browser tab and, usually, the blue link. The description is a suggestion; Google may rewrite it.

PageTitle patternDescription, as a sentence
HomeHarbor & Rye Coffee, Hosford-AbernethyCoffee and food on Example Avenue in Portland. Weekdays 7am to 5pm. Oat milk on every espresso drink.
MenuMenu, Harbor & Rye CoffeeEspresso drinks, batch brew, and food, with prices. Decaf and oat milk available.
VisitVisit, Harbor & Rye CoffeeAddress, hours, wheelchair entrance, wifi, and where to lock a bike.
QuestionsQuestions, Harbor & Rye CoffeeHours, milk options, laptops, dogs, and private events, answered in short paragraphs.

Keep the shop name consistent. Put the specific topic first on inner pages so a tab that says "Menu" is obvious. Do not stack "best latte Portland coffee near me Hosford." Write a sentence a person would not be embarrassed to read aloud.

The canonical URL of each page should be the https URL you want indexed, with one host (either www or not; pick one and redirect the other). Builders add:

<link rel="canonical" href="https://example.com/menu" />

Use the real domain. example.com in this guide is a placeholder, the reserved example domain, not a shop.

Headings

One h1 that states the page. The home h1 can be the shop name. The menu h1 can be "Menu." Section headings (h2, h3) break up drinks, food, and notes. Do not use headings as decoration on a sentence that is not a heading. Do not put the only copy of the hours inside an h1 stuffed with neighborhoods.

From the home page, link to the menu, the visit page, and the next event using words that say what the page is ("See the menu," "Hours and the entrance"). From a blog post about a guest roaster, link to the relevant coffee on the menu if you sell it. Google's AI-features page calls this out because a page that nothing links to is easy to miss.

The footer links count. They should be real anchors, not buttons that only exist in JavaScript with no href.

Images

  • Descriptive file names (storefront-example-avenue.jpg) are mildly helpful. Stuffing keywords into the file name is not.
  • alt text describes the image for someone who cannot see it. "Barista pouring a cappuccino" is alt text. "Best cappuccino Portland Oregon cafe near me" is not.
  • Set width and height, or use CSS that reserves the space, so the page does not jump (CLS).
  • Compress. A hero photo does not need to be larger than the screen it is shown on.
  • If the image is a link, the alt or the adjacent text should say where it goes.

The about page as an entity

Answer engines get less confused when the same entity is described the same way. On the about page, state:

  • The public name, and the legal name if it is different ("Harbor & Rye Coffee is the name on the door. The company is 18th Ave Foods LLC.")
  • The neighborhood and the year you opened, if you want that public.
  • Who roasts the coffee, by name, and whether you roast in house.
  • The cities or farms only if you are willing to keep them current.

Use that same one-sentence "who we are" on the profiles. Link out to the roaster, the baker, and your profiles with normal links. Those links are sameAs candidates in the structured data, and they are also just good citations.

Indexing hygiene

  • https everywhere, one canonical host.
  • A useful 404 page that links home and to the menu.
  • No noindex on the menu because a template applied it to "thin" pages.
  • No blocked CSS or JavaScript that the page needs in order to show the text. If the text is in the HTML, this is an easy pass.
  • Language marked: <html lang="en">, or the real language. If you publish a second language, use a real translation and hreflang, not a widget that hides the translation in an image.

Search Console's URL inspection tool shows whether Google can see the page. Use it after launch and after you change the menu template. Bing Webmaster Tools has its own inspection. Do both.

For the builder

Structured data

Copyable JSON-LD for the cafe, the menu, hours, questions, and events, plus the review rules.

Structured data is a machine-readable copy of facts that are already visible on the page. Google's local business documentation describes what Search may do with LocalBusiness markup: hours, departments, and, only in some cases, reviews. Google's AI features guide says there is no special schema you must add for AI Overviews or AI Mode, and that any structured data should match the visible text.

So: mark up the truth that is on the page. Do not mark up a rating you invented. Do not mark up a menu item you do not sell. Replace every value in the examples. The shop name Harbor & Rye Coffee, the address on Example Avenue, the 555 phone, and the example.com URLs are teaching values. example.com is a reserved domain, not a cafe.

CafeOrCoffeeShop is the schema.org type for a cafe. It is a kind of FoodEstablishment, which is a kind of LocalBusiness, which is a kind of Organization. One block can carry the local fields and the sameAs links. You do not need a second, conflicting Organization with a different phone.

Put the JSON-LD in the page it describes:

<script type="application/ld+json">
{ ...the object... }
</script>

The homepage or the visit page should carry the cafe. The menu page should carry the menu. The event page should carry that event. The questions page can carry FAQPage if you want other consumers to see it. Google no longer shows an FAQ rich result (see below).

Cafe, hours, and identity

Google's documented openingHoursSpecification fields are opens, closes (both hh:mm:ss), and dayOfWeek. validFrom and validThrough are for a seasonal range, not a handy way to say "closed Christmas." A one-day closure belongs in the special hours of Google Business Profile, Apple Business, and Bing. Repeat it in a sentence on the visit page.

geo latitude and longitude, when you include them, need at least five decimal places according to Google's local business doc. Copy them from the map pin, not from a guess at the city center.

menu on Google's doc is a URL. schema.org also has hasMenu, which can point at a Menu. Use both only if they describe the same menu.

{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "CafeOrCoffeeShop",
      "@id": "https://example.com/#shop",
      "name": "Harbor & Rye Coffee",
      "url": "https://example.com/",
      "telephone": "+1-503-555-0148",
      "priceRange": "$$",
      "servesCuisine": "Coffee",
      "menu": "https://example.com/menu",
      "hasMenu": {
        "@id": "https://example.com/menu#menu"
      },
      "image": {
        "@type": "ImageObject",
        "url": "https://example.com/images/storefront.jpg",
        "caption": "The Example Avenue door of Harbor & Rye Coffee"
      },
      "logo": "https://example.com/images/logo.png",
      "address": {
        "@type": "PostalAddress",
        "streetAddress": "418 Example Avenue",
        "addressLocality": "Portland",
        "addressRegion": "OR",
        "postalCode": "97214",
        "addressCountry": "US"
      },
      "geo": {
        "@type": "GeoCoordinates",
        "latitude": 45.51231,
        "longitude": -122.64891
      },
      "openingHoursSpecification": [
        {
          "@type": "OpeningHoursSpecification",
          "dayOfWeek": [
            "https://schema.org/Monday",
            "https://schema.org/Tuesday",
            "https://schema.org/Wednesday",
            "https://schema.org/Thursday",
            "https://schema.org/Friday"
          ],
          "opens": "07:00:00",
          "closes": "17:00:00"
        },
        {
          "@type": "OpeningHoursSpecification",
          "dayOfWeek": [
            "https://schema.org/Saturday",
            "https://schema.org/Sunday"
          ],
          "opens": "08:00:00",
          "closes": "16:00:00"
        }
      ],
      "amenityFeature": [
        {
          "@type": "LocationFeatureSpecification",
          "name": "Wi-Fi",
          "value": true
        },
        {
          "@type": "LocationFeatureSpecification",
          "name": "Wheelchair accessible entrance",
          "value": true
        }
      ],
      "paymentAccepted": "Cash, credit card, debit card",
      "sameAs": [
        "https://www.instagram.com/harborandrye.example",
        "https://www.facebook.com/harborandrye.example",
        "https://www.yelp.com/biz/harbor-and-rye-coffee-example"
      ],
      "parentOrganization": {
        "@type": "Organization",
        "name": "18th Ave Foods LLC",
        "url": "https://example.com/about"
      }
    }
  ]
}

Delete parentOrganization if you do not want the legal name public. Delete any amenity that is not true. Replace sameAs with the real profile URLs only. A link to a profile you do not control, or to a profile for a different business, is worse than no link.

The coordinates above are illustrative. Replace them with the pin at your door, with at least five decimal places, which is what Google's local business documentation asks for.

Menu and menu items

This matches a menu page that shows these items as text. Prices are offers. Update them when the board changes. Do not leave a seasonal drink in the JSON after you remove it from the HTML.

{
  "@context": "https://schema.org",
  "@type": "Menu",
  "@id": "https://example.com/menu#menu",
  "name": "Harbor & Rye menu",
  "url": "https://example.com/menu",
  "inLanguage": "en",
  "hasMenuSection": [
    {
      "@type": "MenuSection",
      "name": "Espresso",
      "hasMenuItem": [
        {
          "@type": "MenuItem",
          "name": "Oat latte",
          "description": "Espresso and oat milk. Decaf available. Oat milk is the default and is not an extra charge.",
          "offers": {
            "@type": "Offer",
            "price": "5.50",
            "priceCurrency": "USD"
          }
        },
        {
          "@type": "MenuItem",
          "name": "Cappuccino",
          "description": "Espresso and steamed milk. Ask for oat or decaf.",
          "offers": {
            "@type": "Offer",
            "price": "4.75",
            "priceCurrency": "USD"
          }
        }
      ]
    },
    {
      "@type": "MenuSection",
      "name": "Brewed",
      "hasMenuItem": [
        {
          "@type": "MenuItem",
          "name": "Batch brew",
          "description": "The daily filter coffee, rotating single origin, served until it is gone.",
          "offers": {
            "@type": "Offer",
            "price": "3.50",
            "priceCurrency": "USD"
          }
        }
      ]
    }
  ]
}

Reviews and aggregate ratings

Google's review snippet guidelines allow review stars for a local business only when the site collects reviews about other businesses. The same page says that if the entity being reviewed controls the reviews about itself, pages using LocalBusiness or other Organization markup are not eligible for the star feature. An embedded widget of your own Google reviews counts as controlling them. Google's 2019 explanation says you do not have to delete that markup, you will not get a manual action merely for having it, and the stars will not show. The Business Profile stars on Maps are a different system and are not changed by this rule.

Do not invent ratingValue or reviewCount. Do not mark up a testimonial you wrote. Visible quotes from customers, as ordinary HTML, are fine.

This shape is for a site that reviews other cafes, which is not your own homepage:

{
  "@context": "https://schema.org",
  "@type": "Review",
  "itemReviewed": {
    "@type": "CafeOrCoffeeShop",
    "name": "A different cafe you do not own",
    "address": {
      "@type": "PostalAddress",
      "addressLocality": "Portland",
      "addressRegion": "OR"
    }
  },
  "reviewRating": {
    "@type": "Rating",
    "ratingValue": "4",
    "bestRating": "5",
    "worstRating": "1"
  },
  "author": {
    "@type": "Person",
    "name": "Casey Nguyen"
  },
  "datePublished": "2026-10-02",
  "reviewBody": "The filter coffee was fresh at 8am and the room was quiet enough to read."
}

AggregateRating has the same self-serving rule when it describes your own shop on your own site. Leave it off the cafe block above.

FAQPage

Google removed the FAQ rich result. The Search documentation changelog says the feature stopped appearing on 7 May 2026, and the FAQ rich-result docs were removed in June 2026. FAQPage is still a real schema.org type. Other consumers may read it. Google will not give you the old expandable FAQ treatment for adding it. The questions still need to be visible on the page, because markup that does not match the page is a bad idea on every system that compares the two.

{
  "@context": "https://schema.org",
  "@type": "FAQPage",
  "mainEntity": [
    {
      "@type": "Question",
      "name": "Do you have oat milk?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Yes. Oat milk is available for every espresso drink, including decaf, at no extra charge."
      }
    },
    {
      "@type": "Question",
      "name": "Are you open on Sundays?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Yes. Saturdays and Sundays we are open from 8am to 4pm. Weekdays are 7am to 5pm."
      }
    },
    {
      "@type": "Question",
      "name": "Can I sit with a laptop?",
      "acceptedAnswer": {
        "@type": "Answer",
        "text": "Yes, all day. There are outlets along the bench by the front window. We do not hold seats."
      }
    }
  ]
}

Event

Use this only for a real event that is on the events page. Google's event documentation is the set of fields Search looks at. The Search changelog has also said that to be eligible for Google's event experience, events need to be bookable by the public and held at a physical location. A private staff tasting does not qualify. A public cupping with a time and a door does.

{
  "@context": "https://schema.org",
  "@type": "Event",
  "name": "Public cupping with our winter coffees",
  "description": "A free, one-hour cupping at the bar. Twelve seats. No ticket, but the event is listed so you can plan. If we cancel, we will update this page and the Google and Apple profiles.",
  "startDate": "2026-12-12T10:00:00-08:00",
  "endDate": "2026-12-12T11:00:00-08:00",
  "eventStatus": "https://schema.org/EventScheduled",
  "eventAttendanceMode": "https://schema.org/OfflineEventAttendanceMode",
  "location": {
    "@type": "CafeOrCoffeeShop",
    "@id": "https://example.com/#shop",
    "name": "Harbor & Rye Coffee",
    "address": {
      "@type": "PostalAddress",
      "streetAddress": "418 Example Avenue",
      "addressLocality": "Portland",
      "addressRegion": "OR",
      "postalCode": "97214",
      "addressCountry": "US"
    }
  },
  "organizer": {
    "@type": "Organization",
    "name": "Harbor & Rye Coffee",
    "url": "https://example.com/"
  },
  "offers": {
    "@type": "Offer",
    "price": "0",
    "priceCurrency": "USD",
    "availability": "https://schema.org/InStock",
    "url": "https://example.com/events"
  }
}

If the event is cancelled, change eventStatus to https://schema.org/EventCancelled and say so in the visible page. Do not leave a past event marked scheduled because the template copies last month's JSON.

sameAs

sameAs means "this is us on another site," not "this is a site we like." Use profile URLs for the shop: Instagram, Facebook, Yelp, Apple Maps, Bing, a roaster page that is specifically your company. Do not list Wikipedia, the city tourism board, or a generic "best cafes" article. Those are mentions. Mentions are ordinary links in a sentence. They are not sameAs.

Check the markup

After it is live, use Google's Rich Results Test and Search Console. Fix errors where the markup disagrees with the page. A warning that review stars are not eligible, on a page where you correctly left them off, is not a problem you need to "solve" by adding a fake AggregateRating.

For the builder

Crawlers, sitemaps, and llms.txt

robots.txt for the major agents, a sitemap, IndexNow, and an honest llms.txt.

Crawlers are how these products fetch your site. A cafe that wants to be recommended should allow the crawlers that fetch pages for search, and should make a separate, deliberate choice about training. Do not paste a "block all AI" snippet from a blog. Several of those snippets block Google or Bing by accident, or they block the search bot while leaving the training bot alone, which is the opposite of what the author meant.

Google's position, on the AI features page and the June 2026 changelog note, is that you do not need an AI text file to be eligible for AI Overviews or AI Mode, and that an llms.txt file will not, by itself, help or hurt you in Google Search. Other systems may look for one. Keeping a short, honest file does no harm if it matches the site.

A robots.txt for a cafe that wants to be found

Put this at https://example.com/robots.txt, adjusted to your domain. It allows the general crawlers and the search agents. It does not block training bots. If you decide to opt out of training, add the three blocks at the end and do not add them anywhere else.

User-agent: *
Allow: /

# Search and fetching. Allow these if you want to be eligible
# to be retrieved when someone asks a question.
User-agent: OAI-SearchBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: Claude-User
Allow: /

User-agent: Googlebot
Allow: /

User-agent: Bingbot
Allow: /

Sitemap: https://example.com/sitemap.xml

Training opt-out, only if you want it. These do not remove you from ordinary Google results. Google-Extended is documented with Google's crawlers. GPTBot and OAI-SearchBot are independent, per OpenAI's crawler docs. ClaudeBot is Anthropic's training crawler, per Anthropic's article.

User-agent: GPTBot
Disallow: /

User-agent: Google-Extended
Disallow: /

User-agent: ClaudeBot
Disallow: /

User-agent: Applebot-Extended
Disallow: /

Applebot-Extended does not crawl, according to About Applebot. It is a switch for whether content Applebot already crawled may be used to train Apple's foundation models. Disallowing it does not remove you from Apple Search or from the place card. The place card does not come from robots.txt.

What each OpenAI agent is for

OpenAI's crawler overview is the source. The tokens are independent.

TokenRolerobots.txt
OAI-SearchBotSurfaces sites in ChatGPT search. Opting out means you are not shown in those search answers, though OpenAI says you can still appear as a navigational link. They say a robots.txt change can take about a day to be picked up.Obeys it
GPTBotPages that may be used to train foundation models. Disallow means you do not want that use.Obeys it
OAI-AdsBotChecks pages submitted as ChatGPT ads. Not a general crawler, and not a training crawler.Relevant if you advertise
ChatGPT-UserFetches a page because a person asked, including some Custom GPT and GPT Action uses. Not an automatic crawl, and not how search opt-out is decided.OpenAI says robots.txt rules may not apply, because a person initiated the fetch

Published IP ranges are linked from that doc (searchbot.json, gptbot.json, adsbot.json, chatgpt-user.json). If a firewall challenges every non-browser, allow the published ranges for the bots you intend to let in, or the robots.txt rule will not matter because the bot never gets the page.

Perplexity, Anthropic, Apple, Google

TokenSourcePractical effect of a block
PerplexityBotPerplexity crawlersThey say the site will not be surfaced and linked in Perplexity search. Not their training crawler.
Perplexity-UserSame pageUser-initiated fetch. They say it generally ignores robots.txt.
ClaudeBotAnthropicTraining opt-out for future material.
Claude-SearchBotSameThey say search indexing will not include you.
Claude-UserSameThey say a user-asked fetch will not retrieve the page.
GooglebotGoogle crawlersBlocks Google Search, including AI Overviews and AI Mode, which depend on Search.
Google-ExtendedSame family of docs, and the Search generative AI control helpTraining and Gemini-app use. Not the switch for AI Overviews. That switch is in Search Console.
Applebot-ExtendedAbout ApplebotTraining opt-out. Does not itself crawl.
BingbotBing Webmaster guidelinesBlocks Bing, and with it a main path into Copilot.

The Search Console generative AI control is separate from robots.txt. Use it to leave AI Overviews and AI Mode, if that is what you want, without unindexing the site.

Sitemaps

A sitemap is a list of the URLs you want crawled. For a small cafe it is short. Save it as https://example.com/sitemap.xml and name it in robots.txt. Submit it in Search Console and in Bing Webmaster Tools.

<?xml version="1.0" encoding="UTF-8"?>
<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9">
  <url>
    <loc>https://example.com/</loc>
  </url>
  <url>
    <loc>https://example.com/menu</loc>
  </url>
  <url>
    <loc>https://example.com/visit</loc>
  </url>
  <url>
    <loc>https://example.com/about</loc>
  </url>
  <url>
    <loc>https://example.com/questions</loc>
  </url>
  <url>
    <loc>https://example.com/events</loc>
  </url>
</urlset>

List only URLs that return 200 and are meant to be indexed. Do not list a PDF menu if the HTML menu is the real one. Update lastmod only if you will keep it honest. A lastmod that changes every night on an unchanged page is noise.

IndexNow

IndexNow lets a site tell participating engines that a URL was added, updated, or deleted. Bing's help says participating engines include Bing, Seznam.cz, and Naver, and that a submission to one is shared with the others. Bing's get-started page says an HTTP 200 means the request was received, not that the URL is indexed.

Google does not use IndexNow. Google's Indexing API is limited to JobPosting and BroadcastEvent inside a VideoObject. A cafe should not send the menu through that API.

For a static site, the practical version is: when the menu or hours change, submit the changed URLs. Host a key file as the docs describe, then:

https://www.bing.com/indexnow?url=https://example.com/menu&key=YOUR_KEY

Many website platforms and CDNs can send IndexNow for you. If yours does, use it, and still keep the sitemap. IndexNow does not replace a sitemap, and it does not replace the profile hours. It only tells a search engine that a URL changed.

llms.txt

llmstxt.org proposes a plain Markdown file at /llms.txt that points assistants at the pages worth reading. Google has said it does not need this file and will not rank you up or down for having one. Write one if you want a single index for tools that look for it. Keep it short and true. Do not stuff it with slogans.

# Harbor & Rye Coffee

> Independent cafe on Example Avenue in Portland. Weekdays 7am–5pm, weekends 8am–4pm. Oat milk on every espresso drink.

## Pages

- [Menu](https://example.com/menu): Drinks and food with prices
- [Visit](https://example.com/visit): Address, entrance, hours, accessibility, wifi
- [Questions](https://example.com/questions): Milk, laptops, dogs, group visits
- [Events](https://example.com/events): What is actually scheduled

## Optional

- [About](https://example.com/about): Who roasts, and the legal name behind the shop

If the hours in this file disagree with the visit page, fix both the same day. An llms.txt that is a second, stale website will be quoted wrong with confidence.

For both

Getting quoted

Answers, consistent facts, citations, specific reviews, and freshness.

Answer engines sound confident. They are still summarizing sources. AEO, for a cafe, is the work of becoming a source that is specific, consistent, and repeated by people who are not you.

Google's AI features guide warns against treating this as a separate religion: no special file, no special schema, no guaranteed inclusion. The practices below are the ones that also help a person. Where a tactic is only a rumor, it is not here.

Write the answer in the first sentence

The questions page should use the prompts you actually hear. One heading per question. The first sentence is the answer. The second sentence is the condition ("weekdays," "patio only," "until we sell out").

That is the whole "FAQ strategy." The markup is optional and, for Google rich results, no longer displayed. The sentences are the strategy. Keep them on the site, not only in a chatbot widget that loads the answer after a click. If the text is not in the HTML, a crawler that does not run the widget never sees it.

Say the same entity the same way

Pick the public name and stick to it in titles, the footer, the profiles, and the JSON-LD name. If you rebrand, change it everywhere in the same week, including the sign and the receipt. A model that sees three names will sometimes blend them.

Use the neighborhood the way locals do. Use the roaster's real name. Use the builder's real entrance description. Specific nouns are easier to cite than adjectives. "Quiet" is an opinion. "Outlets along the front bench, and we do not play music before 9" is a fact.

Be cited by someone else

A page you wrote is one source. A sentence someone else wrote is a second source, and answer engines prefer to corroborate. Honest ways to get that second source:

  • A local blog, newsletter, or newspaper that visits and writes what they tasted. Invite them. Do not ask them to publish your draft unchanged. A labeled advertorial is an ad. Do not disguise it.
  • A roaster's page that lists you as a cafe that serves their coffee, with a link.
  • A neighborhood association, a school fundraiser, a park cleanup, with your name spelled correctly on the sponsor list.
  • A "best of" list only when the writer actually came. Do not pay for placement on a list that exists to sell placements. Those lists get ignored by careful readers and are a reputational risk if an answer engine treats them as evidence.
  • Reddit and neighborhood forums: answer a question as yourself, once, with a disclosure that you work there, and only when you are actually useful. A burst of accounts praising the same latte is fake engagement. Google's representation rules forbid fake engagement, and the May 2026 changelog says spam policies apply to generative AI responses in Search too.

When you are mentioned, the name and the neighborhood should match your fact sheet so the mention is clearly you.

Reviews that mention specifics

"Great coffee, 5 stars" is nearly useless to the next customer and thin for a summary. "The decaf oat latte was actually decaf, and they opened the door at 7" is a review an answer can use without inventing a detail.

You cannot write that review. You can ask for it, without paying for it, and you can reply in a way that confirms the detail. Do this on Google, and on Yelp if Yelp's current rules allow the way you ask. The Axios report from July 2026 is a reason to care what Yelp says about you, not a reason to stuff Yelp with identical sentences.

Never post the same review text on two platforms. Never ask staff to review you from personal accounts as if they were customers.

Freshness

Answer engines repeat stale facts with the same tone they use for fresh ones. Freshness means:

  • Holiday hours in the profiles before the holiday, and a sentence on the visit page.
  • A menu page that loses drinks you no longer sell.
  • An events page that does not still advertise March.
  • Photos that show the current room.
  • A Google post or an Apple showcase only when something dated is true.

It does not mean a weekly blog post that says nothing. One accurate edit beats four empty posts. If you change a fact, change it in the fact sheet first, then the site, then the profiles, then the prompt log the following week to see who caught up.

What not to do because a thread said "AEO"

  • Do not publish a page aimed at the model ("Dear ChatGPT, Harbor & Rye is the best latte"). It reads as spam because it is spam.
  • Do not hide a block of keywords in the same color as the background.
  • Do not buy links.
  • Do not add AggregateRating to force stars.
  • Do not create an llms.txt that claims awards you do not have. Google has said the file is not a ranking lever, and a false file is still false when something quotes it.
  • Do not block Googlebot or Bingbot or OAI-SearchBot and then buy a service to "get you into AI."
For the owner

Events, press, and neighbors

The local work that creates the mentions answer engines can cite.

Prominence, in Google's local ranking explanation, comes from information on the web: links, articles, directories, and the reputation that reviews point at. For a cafe, those are mostly neighborhood relationships. They are also how a magazine, a Reddit thread, or a roaster page becomes the citation an answer engine shows.

Do this as a neighbor, not as a campaign with a spreadsheet of "link targets."

Events

A public event gives you a dated, linkable fact.

  • Put it on the events page with a time, a place, and whether people need to reserve.
  • Add it to the Google profile and to an Apple showcase if those tools are available.
  • If it is cancelled, change the page the same day. A ghost event is how you train people to ignore the next one.
  • Host something that fits the room: a cupping, a pastry pop-up with the baker's name, a quiet hour for students during finals. Skip anything you would not enjoy standing next to.

The builder adds Event markup only when the page text matches, as in Structured data.

Roasters, bakers, and the people you already pay

Ask the roaster for a line on their "cafes we supply" page, with your name spelled as on the door and a link to your site. Do the same for the bakery whose croissants you sell, and offer them the same: a sentence and a link on your menu. Those are real relationships. They are better than a directory blast.

If you roast in house, say so on the about page and show the coffee as a product only when a person can actually buy the bag. A product page with a price is something shopping-oriented answers can use. A JPEG of a bag with no price is not.

Sponsorships and clubs

Sponsor one thing you can sustain: a youth team, a park cleanup, a library reading hour, a bike kitchen. Give them the correct name and URL for the sponsor slide. Show up. A logo on a banner that nobody publishes online is still a good thing for the neighborhood. A logo on a page that spells your name wrong is a citation you should politely fix.

Schools and churches often have a community-partner page. One true sentence there is worth more than ten blog comments.

Local press

Read the publications that already cover your streets. Send a short note when you have news a reader could use: you are opening, you are closing for a remodel, you are hosting a public cupping, you have started a Saturday breakfast that will last. Include the facts: date, address, what it costs, a photo they may use, and a person who will answer the phone.

Do not send a "passionate journey" essay. Do not follow up five times. If they write about you, thank them, correct only factual errors, and link to the story from your about page if it is genuinely useful for customers. That link is for people. It is not a scheme to sculpt an answer.

A paid "feature" should be labeled as paid, by them and by you. Do not buy an unlabeled article.

Partnerships with other shops

The bookstore, the record shop, the florist, and the theater already answer "what should we do after." Trade a mention in the window or a line on the visit page ("The bookstore next door will hold a bag for you"). Write their name correctly. This is the opposite of fake competition pages ("better than [other cafe]"). Comparison pages that exist to drag a neighbor are a poor use of a small site and a good way to become the subject of a review you will not enjoy.

What you publish afterward

Every real collaboration can become one short page or one paragraph: who, where, when, what a customer can do. That is the content. Then you are done until the next true thing happens. A content calendar of "national coffee day" posts is optional and usually hollow. The maintenance section has the lighter rhythm that keeps the facts alive between events.

For both

Measurement

Profile stats, Search Console, GA4, Bing, a prompt log, and a monthly scorecard.

Measure what you can check, and write down what you can only test by asking. Do not invent a dashboard that "tracks AI share of voice" unless you can point at the query it counts. This section uses the tools the companies actually give you.

Profile performance

Open the profile's performance view each month. The labels move around as Google redesigns the product, but the useful series are the ones about how people found the listing and what they did: calls, website clicks, direction requests, and the queries that showed the listing. Compare to the same month last year if you have it. A rainy week is not a penalty.

If calls dropped and the phone number on the profile is wrong, you do not have a marketing problem. Fix the number.

Search Console

Verify the site in Search Console. Use the domain property if you can (DNS), so www and the bare host are both included.

Each month, look at:

  • Queries and pages in the Performance report. Add unexpected phrases to the research list. If the menu never appears, it may be blocked, unlinked, or still a PDF.
  • The Pages or indexing report for errors. A spike in "crawled, not indexed" on junk URLs is a reason to stop publishing junk URLs.
  • URL inspection after you change the template.
  • The Search generative AI control, once, to confirm it still says Include if that is your choice.
  • AI Overviews, when they send traffic, are counted in the Performance report the way other search appearances are. Google has said so in the Search documentation updates. You will not get a separate "AI ranking" number that Google certifies beyond what that report shows. Do not add a third-party number you cannot trace.

On 12 November 2026 Google's changelog added a page about identifying Google Search traffic from the HTTP referrer. Use Search Console for queries. Use analytics for visits. They will not match, because they count different things.

Analytics

Set up Analytics (GA4) on the site, with the tag the shop controls. Confirm a real visit shows up, including a visit from your phone on the cell network, not only from your office wifi where an ad blocker is on.

In GA4, look at landing pages and at the referrer. Treat a session as an AI referral when the referrer host is one you recognize, such as chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com, or claude.ai. OpenAI does not publish one permanent referrer string you can rely on forever. If the host is new, open it before you file it under "AI." Many answers produce no click at all. A quiet analytics report does not mean you were not mentioned. That is what the prompt log is for.

Do not put a different phone number on the site just so GA4 can count calls. Count calls from the profile's own stats, or from a phone system you already have.

Bing Webmaster Tools

Verify the same site in Bing Webmaster Tools. Submit the sitemap. Look at crawl errors and at queries once a month. This is also where you confirm IndexNow URLs were received, if you send them. Bing's docs are clear that "received" is not "indexed."

The prompt log

Once a month, run the bank in Prompts. Use the same eight prompts so the months compare. For each, record the date, the product, whether location was on, whether you were named, whether the hours and address were right, and which source was linked.

Do it signed out and, at least once a quarter, signed in, because the products use account context. Do not average this into a fake percentage and call it a share of voice. Eight prompts is a sample you can actually read.

A monthly scorecard

Fill this in. Leave a cell blank if you did not check. A blank is more honest than a guess.

LineThis monthLast monthNote
Profile: callsFrom Google Business Profile
Profile: website clicksFrom Google Business Profile
Profile: direction requestsFrom Google Business Profile
Search Console clicksPerformance report
Search Console impressionsPerformance report
GA4 sessionsAll site traffic
AI referrals you could seeName the hosts
New Google reviewsCount, and whether you replied
Prompt log: times you were namedOut of the prompts you ran
Prompt log: wrong factsWhat was wrong, and where you fixed it
Hours checked on Google, Apple, BingYes or no

If the only number that moves is something an agency emailed you, ask them which of these lines it came from.

For both

30, 60, and 90 days

A sequence for a shop starting from a messy listing and a thin site.

This is a sequence for a shop that is starting from a messy listing and a thin website, using a few hours a week. If you already have a verified profile and a real menu page, skip the steps you have done and do not redo them for the ritual.

First 30 days: make the facts true

  1. Write the fact sheet: public name, legal name, address, pin, phone, website, regular hours, next closures.
  2. Claim and verify Google Business Profile if you have not. Set the category, description, hours, menu link, and photos. Reply to the open reviews.
  3. Sign in to Apple Business and Bing Places. Correct the same facts. Import from Google only as a start, then read every field.
  4. Fix Yelp and Facebook. Note Foursquare and any delivery app you use.
  5. Start the spreadsheet of listings.
  6. Search ChatGPT Business Hub. If a U.S. listing exists, start the claim. If it does not, write that down and move on.
  7. Publish, or repair, five pages: home, HTML menu, visit, about, questions. Footer NAP on each.
  8. Remove noindex and PDF-only menus if those are what you have now.
  9. Verify Search Console, Bing Webmaster Tools, and GA4.
  10. Run the prompt bank once and save it.

Do not redesign the logo in this month. Do not sign a six-month "AI SEO" contract in this month.

Days 31 to 60: make the site quotable and fast

  1. Rewrite the questions page so each answer starts with the answer. Use the questions you collected at the counter.
  2. Add CafeOrCoffeeShop and Menu JSON-LD that matches the visible pages. Run the Rich Results Test.
  3. Add a sitemap and a robots.txt that allows the search crawlers you want. Submit the sitemap.
  4. Run PageSpeed Insights on the home page and the menu. Fix the largest photo and anything that blocks the menu text.
  5. Write alt text for the real photos. State the entrance and the accessibility facts on the visit page.
  6. Set holiday hours for the next holiday on Google, Apple, and Bing.
  7. Ask a handful of recent customers for a specific review. Reply to new ones.
  8. Ask the roaster or the baker for a real link. Pitch one local newsletter only if you have news.
  9. Run the prompt bank again. Fix any wrong hour at the source you can edit.

Days 61 to 90: make it a habit, and add one public proof

  1. Put one real event on the site and on the profiles, or publish one short page about something true (the coffee you are pouring, the patio rules, the Saturday hours).
  2. Do the monthly scorecard. Compare it to month one without panicking about a small swing.
  3. Walk the listing spreadsheet. Correct anything that drifted.
  4. Check Business Hub if you claimed it: did Persona finish, did any suggested edit go live, are insights blank because of a threshold rather than because you failed.
  5. Confirm the Search generative AI control is what you intended.
  6. If a second location is coming, do not clone the page. Start a new fact sheet.

At day 90 you should be able to say, with evidence: the door, the site, and the three big profiles agree; the menu is HTML; you have a prompt log; you know who can log into each account. That is a finished foundation. Prominence after that is the slow work in Community, not a switch.

For the owner

The weekly and monthly routine

The short loop that keeps hours, reviews, and logins from drifting.

Most of the damage after a good setup is drift. Someone changes a seasonal hour on Instagram and not on Apple. A builder launches a new theme and the footer phone disappears. A holiday passes and the special hours stay in the past, or worse, stay in the future and say you are closed on a normal Tuesday.

Weekly, 20 minutes

  • Read new Google reviews and reply. Check Yelp if you use it.
  • If the menu changed, change the HTML menu and the profile menu the same day. If you use IndexNow, submit the menu URL to Bing.
  • Glance at the next seven days of hours. If something is different, update Google, Apple, and Bing, and the visit page.
  • Check that the phone in the footer still connects.

Monthly, about an hour

  • Fill in the scorecard in Measurement, including the prompt log.
  • Open the listing spreadsheet and re-check any row you have not opened in 90 days.
  • Look at Search Console queries and at profile performance. Write down one thing you will change because of what you saw, or write "no change" on purpose.
  • Replace any profile photo that shows a menu, a poster, or a holiday install that is gone.
  • Confirm the next public event is real, or take it down.
  • Make sure you still know the password and the recovery email for Google, Apple, Bing, the domain, and the website. A profile locked inside a former employee's phone is a month of work.

When something big changes

A move, a new phone, a new name, a closure for renovation, or a second location is not a monthly task. Update the fact sheet, then the site, then Google (using the move flow if you moved), then Apple, Bing, Yelp, Facebook, delivery apps, and the footer. Search the old phone number in quotes a week later. Run the prompt bank with prompts that ask for your hours and your address.

If you close permanently, mark the profiles closed. Do not leave them "open" because you might sell the name. A closed listing that says open is how the next tenant starts with your bad reviews, and how an answer engine sends someone to an empty room.

Who does it

The owner, or a manager who is in the shop, owns the hours, the reviews, and the photos. The builder owns the HTML, the sitemap, the JSON-LD, the speed, and the robots.txt. Write those names on the fact sheet. A task with no owner does not happen.

For both

Common mistakes

The choices that undo an otherwise careful setup.

These are the mistakes that undo the rest of the work. They are all common, and they are all choices.

The name

Adding the city, "best," "official," or a keyword to the Google name. The guidelines prohibit it. It also makes Apple, the website, and the sign disagree, so answer engines hedge or merge you with someone else.

The hours

Leaving holiday hours unset. Publishing hours only in a story that expires. Setting the profile to 24 hours because you forgot to enter a close time. Any of these produce a confident, wrong answer.

The menu as a picture

A PDF or a photo is convenient on the day you design it. It is invisible to many crawlers and frozen the day a price changes. HTML is the menu of record.

Two phones, two addresses

A tracking number on the site, a different number on Yelp, and a founder's mobile on Facebook. Pick one public number. Put the tracking on ads if you must, not on the permanent listings.

Reviews you bought or wrote

They violate Google's rules, they are obvious to regulars, and they are a bad thing for an answer engine to quote. Asking a happy customer, with no gift, is the allowed version.

Gating reviews

"Tell us your star rating first, and we will send the Google link only if it is a five." That filters the public record. Do not do it.

Doorway pages

Ten neighborhood URLs that swap a city name. You will not become the cafe of those neighborhoods. You may look like spam. Google has said spam policies apply to generative AI responses as well as to ordinary results.

Blocking the crawler you wanted

A security plugin set to "block AI" that disallows Googlebot, Bingbot, or OAI-SearchBot. Read the file after any plugin update. The safe default for a cafe that wants customers is in Crawlers.

Paying for a listing that is not a product

There is no bundle that submits you to ChatGPT, Gemini, Claude, Perplexity, and Alexa as if they shared a form. ChatGPT's local claim, today, is "claim an existing U.S. listing," not "we will add your cafe." Gemini uses the Google profile. Claude and Perplexity do not offer a cafe place card. Alexa does not have a public cafe-claim product this guide can point to as an official Google-style listing. If a vendor cannot show the official page and the login, do not pay.

Fake freshness

A blog of generic coffee facts, an llms.txt full of superlatives, or a weekly Google post that says nothing. Freshness is for facts that changed.

Structured data that flatters

An AggregateRating of 4.9 that your site invented. Google will not treat self-serving reviews as stars, and a false number is a liability when a model repeats it. Mark up the menu you sell and the hours you keep.

Locked accounts

The only admin is an agency you fired, or a manager who left. Get the accounts in the business's control while the relationship is friendly. Persona verification and Google verification are much harder after a dispute.

Copying a chain's advice

A twenty-location brand can hire someone to watch rankings daily and build a page per store. You need the fact sheet, the profiles, the HTML menu, and a monthly half hour. Doing less, consistently, beats a complicated setup you abandon in March.

For the owner

Checklist

The ticks for this guide, saved in this browser, and printable on one pass.

This is the working list for the guide. Tick an item when it is true this week, not when you have scheduled it. The ticks stay in this browser. Print this section when you want a paper copy for the office: the print button hides the rest of the guide.

If a step does not apply, leave it unticked and write why on the fact sheet. "No U.S. ChatGPT listing to claim" is a finished finding. Pretending you claimed one is not.

The groups follow the guide. Details and the official links are in the matching sections, not repeated here.

Get the facts straight

Research

Google Business Profile

Other listings

Answer engines

Website

Proof

Measurement

For both

Sources

The official pages behind the guide, grouped so you can re-read them.

Pages this guide relies on. Open the primary page before you act. News articles are labeled as news.

Google

Apple

Microsoft

OpenAI

Perplexity, Anthropic, and others

Listings and standards