# How to Get ChatGPT to Recommend Your Cafe (and Why Most Venues Are Invisible to It)

> Getting recommended by ChatGPT, Claude or Perplexity when someone asks for a good local spot is almost entirely an off-page problem: their crawlers do not render JavaScript, and none of them owns a local index the way Google Maps does. The prompts to test it yourself, and where a citation actually comes from.

- Author: Alex Riesenkampff (Super44)
- Published: 2026-08-14
- Canonical: https://super44.ai/blog/chatgpt-recommend-your-restaurant

## Key takeaways

- A March 2026 study of 1,211 independent venues across eight UK cities found 84.7% of independent restaurants and bars got zero mention when AI assistants were asked for a recommendation, rising to 100% in Liverpool (booteek.ai).
- None of ChatGPT, Claude or Perplexity has its own local-business index the way Google Maps does: OpenAI licenses Yelp's and Foursquare's data for ChatGPT, and Perplexity licenses Tripadvisor's. Only Google's own Gemini is grounded directly in Maps.
- Across more than 25 million AI-cited links, Muck Rack found earned media, not brand-owned websites, accounts for 84% of AI citations; paid or advertorial content barely registers at 0.3%.
- US consumer use of AI for local recommendations rose from 6% to 45% in a single year (BrightLocal, 2026), yet Booking Holdings, one of the largest travel platforms in the world, still recorded under 1% of its room nights arriving via AI assistants in its Q2 2026 results.
- A peer-reviewed Princeton and Georgia Tech study (KDD 2024) found the single strongest lever for being cited in an AI answer is adding a direct quote from a named source, a 41% visibility lift, ahead of adding statistics (around 30%).

If ChatGPT, Claude or Perplexity has never named your cafe when someone asks for a good spot nearby, you are the rule, not the exception. **A March 2026 study of 1,211 independent venues across eight UK cities found 84.7% of independent restaurants and bars got zero mention when AI assistants were asked for a recommendation, rising to every single venue checked in Liverpool.** The fix almost nobody reaches for first is the right one: none of the big three assistants has its own database of local businesses the way Google Maps does. ChatGPT's local answers are licensed from Yelp and Foursquare, Perplexity's from Tripadvisor, and none of them so much as looks at most small-venue websites, because their crawlers do not run the JavaScript that builds them. Getting recommended is mostly a question of what other people and platforms say about you, not what you say about yourself.

## Run the diagnostic before you fix anything

**The fastest way to know where you actually stand is to ask the assistants directly, the same way a guest would, and write down exactly what comes back.** Open ChatGPT, Claude, Perplexity and Gemini in separate tabs and run a plain, realistic prompt for your own neighbourhood, not your venue's name. Then run a second prompt that does name you, to check whether the details it states are even correct.

**Discovery prompt: does it name you at all?**

```text
Recommend a good independent [cafe / bar / restaurant / bakery] in [your neighbourhood, city]. I want somewhere small and local, not a chain. Give me two or three options with a short reason for each.
```

**Accuracy prompt: does it get you right?**

```text
What do you know about [your venue name] in [your town or neighbourhood]? Include its current opening hours, what kind of food or drink it serves, and its approximate price range.
```

Run both on a real device, since results vary by session and are not guaranteed to repeat. What you get back sorts into three outcomes, and each one points to a different next step, covered in the decision guide further down: not mentioned at all, mentioned with something wrong (closed when you are open, the wrong cuisine, an old address), or mentioned correctly but only when you ask by name rather than by category. None of the three outcomes is fixed by redesigning your homepage.

## Why your website's schema markup probably isn't the fix

**OpenAI's and Perplexity's own crawler documentation confirms a real-time search bot fetches your page, but neither vendor states whether that bot executes JavaScript, and independent technical testing has repeatedly found that it does not.** OpenAI names OAI-SearchBot as the crawler behind ChatGPT's search features, separate from the GPTBot crawler used for model training; Perplexity names PerplexityBot the same way for its own search results. Vercel's own technical testing of these bots, alongside a broad body of independent audits, found a consistent pattern: they behave as plain HTTP fetchers that do not render JavaScript, even when they download a script file, they read it as inert text rather than running it.

That matters more than it sounds like it should, because a large share of small-business websites, including most templated restaurant sites built on modern frameworks, render their menu, hours and description client-side, in the browser, after the initial page loads. Googlebot solves this by running a full headless browser. The assistants' own search crawlers, by the clearest evidence available, generally do not. A site that ranks perfectly well on Google can still be functionally blank to the crawler ChatGPT or Perplexity actually uses, an empty shell with nothing for the model to read or quote. This is a genuinely different failure mode from the one our piece on [whether a restaurant still needs a website](/blog/restaurant-website-vs-google-profile) covers, which is about whether Google's own AI Overviews can cite an indexed page at all; that argument still holds once a page is crawlable. This one is earlier and more basic: can the bot even read the page in the first place.

## None of them has their own map

**Unlike Google, which built Gemini directly on top of Google Maps and Google Business Profile, none of the other three major assistants owns a comparable database of local businesses, so each has gone and licensed one.** The practical result is that where an assistant's local answers come from is now a matter of public record, not guesswork.

**What grounds each assistant's local recommendations**

| Assistant | What it draws local data from | Owns its own local index? |
| --- | --- | --- |
| ChatGPT (OpenAI) | Yelp (reviews, ratings, and since Aug 2026 US/Canada reservations) and Foursquare's Places API, 100M+ points of interest | No, licensed |
| Perplexity | Tripadvisor (1bn+ reviews and contributions) plus open-web retrieval | No, licensed |
| Gemini (Google) | Google Maps / Google Business Profile | Yes, Google's own index |
| Claude (Anthropic) | No dedicated local-business data partnership found in our research | No known partner as of writing |

*Sources: Yelp (blog.yelp.com, 10 Aug 2026), Foursquare (16 Dec 2024), Hospitalitynet/Tripadvisor (10 Jan 2025), Search Engine Land reporting SOCi's 2026 Local Visibility Index (28 Jan 2026).*

The SOCi Local Visibility Index found the practical consequence of that gap: across roughly 350,000 tracked locations, Gemini's answers matched a business's own Google Maps data 100% of the time, since it is grounded directly there, against only around 68% accuracy for ChatGPT and Perplexity, which are working from a different, licensed dataset one step removed from the source. That single number is the clearest evidence that "optimise your website" is solving the wrong layer of the problem for three of the four assistants a guest might actually use. It also means fixing an error happens at the source: a wrong opening time traces back to whichever of Yelp, Google or Tripadvisor the assistant actually pulled it from, and correcting only your own site leaves the wrong version live everywhere an assistant looks first.

## Where citations actually come from: earned media, not owned content

**Across more than 25 million links cited by ChatGPT, Claude and Gemini, Muck Rack found earned media, coverage you did not write or pay for, accounts for 84% of AI citations, with paid or advertorial content barely registering at 0.3%.** That figure lines up with a peer-reviewed finding from the other direction. Princeton and Georgia Tech researchers, presenting at KDD 2024, tested specific content changes against a 10,000-query benchmark and found that adding a direct, attributable quote from a named source produced the single largest gain of any tactic tested, a 41% lift in a model's likelihood of citing that content, ahead of adding statistics (around 30%). Both findings point the same way: an AI system trusts what other, independent sources say about a subject more than what the subject says about itself, whether that subject is a brand, a claim, or a small venue.

> **84%** — Share of AI citations across 25M+ links that come from earned third-party media rather than a brand's own website (Muck Rack, May 2026)

Kelsey Verdier, VP of Marketing at Marqii, a restaurant-data platform, puts the more familiar advice about structured data in useful perspective: "Schema is basically a translation of what a human sees on the front end of your website, translated into robot speak." A translation only helps once someone is reading the document in the first place. For an independent venue with a thin site and no press mentions, that document is rarely the one an assistant reaches for.

## What this actually means for an independent venue

**The concrete, unglamorous version of "build earned media" is completing the same handful of third-party listings the assistants are already licensing, then adding one or two things genuinely worth citing.** None of this requires an agency or a marketing budget; it requires an evening and then a repeatable, small monthly habit.

**Where to actually spend the next hour**

- [ ] **Claim and complete your Yelp and Tripadvisor listings, not only Google**: ChatGPT's and Perplexity's local answers are licensed from exactly these two platforms; an incomplete or unclaimed listing there is invisible to both, regardless of your own site.
- [ ] **Get your review count and rating past roughly 4.4 stars**: A 2026 US study of 551 AI dining recommendations found this acts as a rough entry threshold below which AI assistants rarely recommend a venue at all; above it, review volume matters more than chasing a higher average.
- [ ] **Earn one genuine local mention**: A local newsletter, a food writer's list, a neighbourhood blog: this is the earned media that Muck Rack's research found behind 84% of AI citations, and it is worth more than any amount of on-page copy.
- [ ] **Don't drop your Google Business Profile to chase this**: Gemini is grounded directly in Maps, so profile work is still the widest single lever any of the four assistants respond to; our Google Maps ranking playbook covers that side in full.
- [ ] **Re-run the two diagnostic prompts monthly**: Results change as the underlying listings and press mentions change; treat this as a habit, not a one-time check.

*None of this requires a rebuild of your website. It requires the listings the assistants already trust to be accurate and complete.*

## How much is actually riding on this right now

**Consumer use of AI for local recommendations is growing fast, but the actual volume flowing through it today is still small even at enormous scale, which is a reason to make time for this, not a reason to panic about it.**

**How much is actually riding on AI recommendations today**

*Source: BrightLocal 2026 and Booking Holdings Q2 2026 earnings call, reported 4 August 2026*

- **Supported:** Use of AI assistants for local recommendations is growing fast in the US, from 6% of consumers to 45% in a single year, and is now the third most popular source of business recommendations overall.
- **Not yet true:** Even the world's largest online travel platform, Booking Holdings, still recorded traffic from AI assistants, paid and unpaid combined, at under 1% of its room nights in its Q2 2026 results. Treat this as getting ahead of a fast-growing channel, not as replacing the channels that already convert.

Super44 connects to your POS and shows you where your money leaks — then tells you what to do about it: https://super44.ai

## What to do with what you found

**Your diagnostic result from the top of this article points to one of three different next moves, and they are not the same fix.**

**What did the discovery prompt actually return?** *(interactive decision helper in the web version)*

**Possible routes:**
- **When you ran the discovery prompt for your neighbourhood, what happened? — My venue wasn't named at all:** Start with the listings, not the website. You have no presence in the datasets these assistants actually license. A redesigned homepage will not change that. Next move: Claim and complete your Yelp and Tripadvisor listings this week, and check your Google Business Profile is verified and current.
- **When you ran the discovery prompt for your neighbourhood, what happened? — It was named, but a real detail was wrong (hours, whether it is still open, the cuisine):** Fix it at the source platform. The wrong detail almost certainly traces back to whichever listing the assistant pulled it from, not your own site. Next move: Correct the specific field on Yelp, Google or Tripadvisor, and re-run the prompt in a few weeks; these feeds refresh on their own schedule, not instantly.
- **When you ran the discovery prompt for your neighbourhood, what happened? — It was named correctly, but only when I asked for it by name, not by category:** You are indexed. Now build what gets you into more answers. Being found by name but not by category means you clear the basic data bar but lack the earned mentions that put a venue into an open-ended recommendation. Next move: Pursue one genuine local press or "best of" mention, and keep your review count climbing past the roughly 4.4-star floor.

For most independent venues, the honest starting point is the first branch: the data the assistants are actually reading from is thin, not wrong, and thin is the cheaper problem to fix. The one thing not to do is treat this like traditional SEO and start stuffing keywords into an about page; none of the research behind this article found a single tactic that worked that way, and a page written for a crawler instead of a guest reads as exactly that to both.

## Where Super44 fits

**Super44 does not audit or fix AI-visibility directly, but it already does the adjacent, unglamorous work that feeds it: syncing a venue's Google reviews automatically and drafting a copy-ready reply for the owner to edit and post.** That keeps replying to reviews a five-minute habit rather than another task competing for an owner's evening, which matters given how directly Pluspoint's research ties review count to the AI recommendation floor. For the wider Google Business Profile picture that Gemini is grounded in directly, our [playbook for ranking higher on Google Maps](/blog/google-maps-ranking-restaurant) covers the rest of that lever. And if the question underneath all of this is which discovery channel is worth an owner's next hour at all, [where diners actually find you](/blog/restaurant-discovery-channels) puts AI assistants in context next to word of mouth, social media, and simply walking past.

## FAQ

### How do I get ChatGPT to recommend my restaurant?

Start with the diagnostic prompts in this article to see what ChatGPT currently says about you, then work on the sources it actually draws from rather than your own website: a complete, accurate Yelp listing (ChatGPT's local answers are licensed from Yelp and Foursquare), a Google Business Profile with a review count and rating above roughly 4.4 stars (a threshold from a 2026 US study), and any genuine local press or "best of" coverage you can earn. Website fixes matter later, not first.

### Does my restaurant need a website for ChatGPT to recommend it?

Not for the initial recommendation, which usually comes from licensed third-party data, but yes for what happens after: once an assistant does look at your own page, it needs something specific to quote. Our piece on whether a restaurant still needs a website in 2026 covers that separate, later-stage problem in full.

### Why doesn't Claude or Perplexity know about my restaurant?

Two reasons, usually together. Their search crawlers fetch raw HTML and do not run JavaScript, so a site built on a JavaScript framework can be functionally invisible to them even if it ranks fine on Google. Separately, neither Claude nor Perplexity has its own database of local businesses; Perplexity licenses Tripadvisor's, and no equivalent partnership for Claude turned up in our research, so its local answers are unusually dependent on whatever it can find on the open web.

### Does having more Google reviews help me get recommended by AI?

It helps up to a point, then stops being the lever. A 2026 US study of 551 AI dining recommendations found roughly 4.4 stars acts as an entry threshold below which AI assistants rarely recommend a venue at all, but above that floor, review volume matters more than chasing a higher star average. Our guide to how many Google reviews a restaurant actually needs covers the same finding for Google Maps ranking specifically.

### Is it worth spending real time on AI visibility right now?

Worth an hour a month, not a rebuilt marketing plan. Consumer use of AI for local recommendations grew from 6% to 45% in the US in a single year, but even Booking Holdings, a travel giant, still recorded under 1% of its room nights from AI assistants in mid-2026. Treat this as getting ahead of a fast-growing channel, not replacing the Google Business Profile and review habits that already convert.

### Can I fix AI invisibility by adding schema markup to my website?

Schema helps a system that has already decided to look at your page understand it correctly, but it does not make an assistant look at your page in the first place if it is not JavaScript-rendered or if the assistant is answering from licensed third-party data instead. Treat markup as a finishing step after the listings and earned mentions below, not the first fix to reach for.

## Sources

1. [Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande: GEO: Generative Engine Optimization](https://arxiv.org/abs/2311.09735) — Peer-reviewed paper, ACM SIGKDD KDD 2024 (dl.acm.org/doi/10.1145/3637528.3671900, paywalled to automated fetches; arXiv preprint linked here). GEO-bench, ~10,000 queries. Quotation Addition +41%, Statistics Addition roughly +30%, Cite Sources +115.1% for rank-5 pages specifically (Position-Adjusted Word Count metric).
2. [Muck Rack: What Is AI Reading? (May 2026 edition)](https://muckrack.com/blog/what-is-ai-reading-may-2026) — Analysis of 25M+ links cited by ChatGPT, Claude and Gemini across 17 industries. Earned media 84% of citations, paid/advertorial content 0.3%, journalism 27%.
3. [Vercel: The rise of the AI crawler](https://vercel.com/blog/the-rise-of-the-ai-crawler) — Independent technical testing of AI search crawler behaviour: plain HTTP fetchers, no JavaScript execution.
4. [OpenAI: Bots and crawlers used by ChatGPT and OpenAI](https://developers.openai.com/api/docs/bots) — OpenAI's own documentation naming OAI-SearchBot as the crawler that surfaces websites in ChatGPT's search features, separate from the GPTBot training crawler.
5. [Perplexity: Perplexity crawlers documentation](https://docs.perplexity.ai/docs/resources/perplexity-crawlers) — Perplexity's own documentation naming PerplexityBot as the crawler that surfaces and links websites in Perplexity's search results.
6. [Search Engine Land: Anthropic clarifies what its three web crawlers do](https://searchengineland.com/anthropic-claude-bots-470171) — Names ClaudeBot (training), Claude-User (live user-triggered fetch) and Claude-SearchBot (search index) as distinct Anthropic crawlers.
7. [Yelp: Yelp and ChatGPT](https://blog.yelp.com/news/yelp-chatgpt-integration/) — Published 10 August 2026. Confirms ChatGPT surfaces Yelp reviews, ratings and business details, and, in the US and Canada, can book Yelp Reservations and Waitlist without leaving the chat.
8. [Foursquare: Foursquare and OpenAI partnership announcement](https://x.com/Foursquare/status/1868713120765448330) — Foursquare's own announcement, 16 December 2024, that its Places API (100M+ points of interest across 200+ countries) powers ChatGPT search's real-world results.
9. [Hospitalitynet / Tripadvisor: Tripadvisor and Perplexity partnership](https://www.hospitalitynet.org/news/4125323.html) — Announced 10 January 2025. Gives Perplexity access to over one billion Tripadvisor reviews and contributions plus Viator experiences.
10. [Search Engine Land: AI local visibility is up to 30x harder than ranking in Google](https://searchengineland.com/ai-local-visibility-report-2026-468085) — Reporting SOCi's 2026 Local Visibility Index. Gemini's answers matched a business's own Google Maps data 100% of the time, against roughly 68% for ChatGPT and Perplexity, across ~350,000 locations.
11. [PR Newswire: In AI-Driven Discovery, Few Brands Are Chosen, Most Disappear](https://www.prnewswire.com/news-releases/in-ai-driven-discovery-few-brands-are-chosen-most-disappear-302672281.html) — SOCi's 2026 Local Visibility Index, published 28 January 2026. ChatGPT recommends 1.2% of brand locations on average, across 2,751 multi-location brands, versus 35.9% average appearance in Google's local 3-pack. Chain data, not independents.
12. [booteek.ai: UK AI visibility research, March 2026](https://booteek.ai/press/uk-ai-visibility-march-2026) — 95 AI probes (30 zone-specific plus 65 Manchester/Bath queries) across 8 UK cities, covering 1,211 verified independent venues. 84.7% of independent venues invisible to AI overall; 100% in Liverpool (0/48), 77.0% in Manchester (135/586).
13. [BrightLocal: Local Consumer Review Survey 2026, AI Trust module](https://www.brightlocal.com/research/lcrs-ai-trust/) — 1,002 US consumers surveyed. Use of AI to find local business recommendations climbed from 6% in 2025 to 45% in 2026, the third most popular source of business recommendations overall.
14. [Investing.com: Booking Holdings Q2 2026 earnings call transcript](https://www.investing.com/news/transcripts/earnings-call-transcript-booking-holdings-tops-q2-2026-forecasts-shares-jump-93CH-4836060) — Reported 4 August 2026. CFO Ewout Steenbergen: traffic from large language models, paid and unpaid combined, is "significantly below 1% of our room nights."
15. [Marqii: GEO 101, A Guide for Restaurants](https://blog.marqii.com/geo-101-a-guide-for-restaurants/) — Published 27 February 2026. Source of the Kelsey Verdier quote on schema markup as a translation layer for AI systems.
16. [Pluspoint: What AI Recommends When Diners Ask Where to Eat](https://www.pluspoint.io/blog/what-ai-recommends-when-diners-ask-where-to-eat) — Published 4 August 2026. 551 AI answers across 12 US cities on ChatGPT, Gemini and Perplexity. Roughly 4.4 stars functions as an entry threshold, not a competitive edge above it.