Why ChatGPT Isn't Recommending Your Restaurant (8-Step Audit)

If ChatGPT, Google AI Overviews, Perplexity, or DeepSeek skip your restaurant when diners ask for one in your neighborhood, the cause is almost always one of eight specific gaps in how AI training data, retrieval, and citation sources see your menu — and every one of them is fixable in under a quarter.

Per OpenTable's 2026 State of Dining report (released January 2026, n=8,400 US diner respondents), 33% of US diners under 35 reported using ChatGPT, Google AI Overviews, Perplexity, or DeepSeek to research a restaurant in the prior 90 days; Eater's 2026 Industry Pulse and Yext's October 2025 6.8M-citation study both confirm that food service draws 41.6% of citations from listings + 13.3% from reviews — the highest reviews share of any industry studied. Yelp/OpenAI's 2025 data licensing partnership and Foursquare's role powering 60-70% of ChatGPT local results (per LinkedIn analysis cited by BrightLocal, July 2025) make the listing layer especially load-bearing for restaurant AI citation.

Restaurants face the hardest citation environment of any local vertical. The discovery surfaces are fragmented across Yelp, OpenTable, Resy, TripAdvisor, Google Maps, Eater, and a long tail of city publications. The qualifier prompts ("vegan", "kid-friendly", "omakase", "patio dinner") are extremely sensitive to how your menu and amenities are structured. And chains carry decades of training-data gravity that an independent will never match on generic terms.

The audit below is the diagnostic we run when restaurant marketing agencies bring us in to figure out why a well-reviewed independent keeps getting skipped for date-night and dietary prompts that should be theirs.

Section 1 — How AI assistants actually pick the restaurant they recommend

Three steps, every prompt:

Retrieval. The model assembles a candidate restaurant set from a small high-trust source pool: Eater city verticals, OpenTable and Resy listings, TripAdvisor's "things to do" pages, Yelp's restaurant category, James Beard award/nominee lists, and city-publication round-ups (Time Out, Thrillist, regional alt-weeklies). Trade pubs like Restaurant Business and Nation's Restaurant News feed business-context, not consumer recs.

Reranking. The candidate set gets reordered against the prompt's qualifiers. "Date night" reweights toward higher price point, ambience-related review excerpts, and Resy's editorial picks. "Vegan" reweights toward menus marked with dietary properties or coverage in vegan-vertical pubs. "[City] omakase" reweights toward Eater Heatmap inclusion and Tabelog or Resy tasting-menu listings. Each qualifier has a different signal mix.

Citation. The LLM names 1 to 7 restaurants and almost always cites the source. Restaurants that appear only on Yelp get cited as "Yelp says…" and increasingly downweighted. Restaurants that appear in Eater or a James Beard list get cited at face value with the editorial source as authority — which is why a single Eater Heatmap inclusion outperforms a thousand additional Yelp reviews for AI surface visibility.

The eight steps below each target one specific failure mode in this pipeline.

Section 2 — The 8-step diagnostic

Step 1 — No Eater (or city-publication) citation

Symptom you'll observe. For "best [cuisine] [city]" and "best new restaurants [neighborhood]" prompts, ChatGPT and Perplexity name competitors with editorial coverage and skip you, even when your reviews and reservation availability are stronger.

Likely cause. Eater's Heatmap, Eater 38, and city-vertical round-ups are the highest-trust editorial citation in the restaurant vertical. Time Out, Thrillist, Bon Appétit Hot 10, and regional alt-weeklies sit just below. If you appear in none of these, you cannot enter the candidate set for editorial-flavored prompts.

How to verify. Site-search each city publication for your restaurant name and your chef's name. If you score zero across the top 5 city pubs in your market, you are entity-invisible to editorial-driven prompts.

Fix. Editorial coverage is pitched, not bought. Hire or contract a restaurant publicist for one quarter with one specific goal: a single inclusion in your city's Eater vertical or equivalent. The hook needs to be real news — chef hire, menu format change, opening, expansion. Set realistic expectations: 3-6 months from pitch to publication. Most restaurants pitch wrong: they pitch promotional copy ("our anniversary special") rather than a real news hook ("we're switching to a regenerative-agriculture sourcing model and dropped 40% of our menu"). The latter gets covered; the former gets ignored. Brief your publicist accordingly.

Step 2 — Weak OpenTable / Resy review volume and recency

Symptom you'll observe. For "best date night [neighborhood]" prompts you appear sporadically. The answers that include you cite OpenTable; the answers that skip you cite Resy or vice versa.

Likely cause. AI assistants pull both review density and recency. Most independents are listed on one of the two reservation platforms but have neglected review density on it for years. Once your most recent review is 6+ months old, the platform's own algorithm and the AI's recency signal both push you down.

How to verify. Count reviews on whichever platform you use; check the date of your most recent review. Run "best date night [neighborhood]" in ChatGPT and Perplexity and note which restaurants appear and what their review density looks like.

Fix. Pick one platform and concentrate. Build a 60-day post-meal review prompt cadence (table card, receipt insert, post-visit email). Target 200+ reviews with a most-recent date inside 30 days. This is one of the fastest-moving levers in the audit. Train front-of-house to mention specific dishes by name when prompting reviews — "if you order the lamb shoulder again, would you mind leaving a quick note?" — because dish-name mentions in reviews are extracted as qualifier signals by AI assistants for "best [dish] [city]" prompts. Generic five-star reviews carry weight; reviews that name dishes carry more.

Step 3 — No Menu / MenuItem schema with dietary properties

Symptom you'll observe. For dietary qualifier prompts ("vegan restaurants near me", "gluten-free [cuisine]", "dairy-free dinner [city]") you do not appear, even though your menu is genuinely suitable.

Likely cause. Schema.org's Menu and MenuItem types accept structured suitableForDiet properties (VeganDiet, GlutenFreeDiet, KosherDiet, HalalDiet, LowFodmapDiet, etc.) and allergen info. Without these, AI assistants cannot reliably extract that your menu serves a dietary qualifier, and they err toward filtering you out.

How to verify. Drop your menu page into Google's Rich Results Test. Confirm Menu and MenuItem schema is present and that suitableForDiet properties are populated where applicable.

Fix. Add the schema. This is a 4-8 hour engineering task. The payback is permanent across every dietary qualifier prompt the model ever runs.

Step 4 — Dietary tags missing from menu copy and review excerpts

Symptom you'll observe. Even with schema in place, dietary prompts surface competitors with weaker actual offerings.

Likely cause. The schema gets you into the candidate set; review excerpts and menu copy get you reranked above competitors. AI assistants pull review excerpts heavily from Yelp, OpenTable, and Resy when they reweight for qualifiers. If no reviews mention "vegan" or "gluten-free", the rerank does not lift you even if your schema marks suitability.

How to verify. Search your reviews on Yelp and OpenTable for the dietary keyword. Count mentions. Cross-reference against competitors that consistently outrank you on the qualifier prompt.

Fix. Two actions: (a) update menu copy to use dietary keywords explicitly in dish descriptions, not just symbols; (b) train front-of-house to seed dietary keywords in post-meal review prompts ("if you ordered our vegan tasting, would you mind mentioning that in your review?").

Step 5 — No James Beard nomination, semifinalist, or finalist mention

Symptom you'll observe. "Best chef [city]" and "fine dining [city]" prompts skip you for restaurants with credentials you consider weaker.

Likely cause. James Beard nominations carry disproportionate citation weight because the JBF site is high-trust and the credential propagates through Eater, Bon Appétit, Food & Wine, and city publications, creating a multi-source citation halo that lasts years.

How to verify. Site-search jamesbeard.org for your chef's name. Search "[Chef Name] James Beard" in Google.

Fix. Submit nominations every cycle in every category that fits — Best Chef regional, Best New Restaurant, Outstanding Restaurateur. Regional chef nominations are achievable for serious independents; even semifinalist status creates a multi-year citation lift. The James Beard Foundation accepts public nominations through its annual open-call window; if your chef has not been nominated, the first move is to nominate yourself, then have three or four credible industry contacts (other chefs, food writers, restaurateurs) submit independent nominations as well. Multiple independent nominations are how chefs without existing JBF connections enter the longlist.

Step 6 — A chain entity dominates training data in your category

Symptom you'll observe. For generic "[city] [cuisine]" prompts ChatGPT names two or three chains regardless of your local signal strength.

Likely cause. Chain entities have heavy training-data presence: news coverage, financial filings, Wikipedia, decades of trade-pub mentions. The base-model embedding for "burger place [city]" or "Italian restaurant [city]" sits close to chain names by default.

How to verify. Run the prompt 10 times in fresh ChatGPT sessions. Compare against Perplexity (retrieval-heavy, less chain bias) and AI Overviews (mid-bias).

Fix. Compete on qualifier prompts where chain pages are too generic: specific cuisine subsets ("Sicilian", "Hokkaido-style ramen"), neighborhood + dietary combinations, occasion-specific ("anniversary dinner", "private dining 12"), tasting-menu price tiers. Chain location pages rarely carry these qualifiers.

Step 7 — Yelp insufficient as your only third-party signal

Symptom you'll observe. You appear only in answers that openly cite Yelp. Higher-trust answers (cited from Eater, Resy editorial picks, James Beard) skip you.

Likely cause. Yelp is the lowest-trust citation surface AI assistants pull for restaurants. If it is your only third-party signal, you get cited only on lower-confidence answers.

How to verify. Run your top 8 buyer-intent prompts and log which sources are cited.

Fix. Layer in three higher-trust surfaces: an Eater pitch, a city-publication round-up, and an OpenTable/Resy editorial pick. Even one Bon Appétit, Time Out, or Thrillist mention shifts the citation mix dramatically.

Step 8 — TripAdvisor weak for tourist-driven prompts

Symptom you'll observe. Out-of-town diners ask AI for restaurants in your city; you do not appear despite strong local reputation.

Likely cause. Tourist-flavored prompts ("where to eat [city]", "[city] travel guide") reweight heavily toward TripAdvisor and Travel-Weekly-tier pubs. Restaurants beloved by locals but with thin TripAdvisor presence get filtered.

How to verify. Run "best restaurants in [your city]" and "where to eat in [your city]" in ChatGPT and Perplexity from a fresh session. Note which sources are cited.

Fix. TripAdvisor density is a separate workstream from OpenTable/Resy. Run a 90-day TripAdvisor review-prompt cadence. Claim and complete your TripAdvisor profile (photos, menu, hours, dietary tags).

Section 3 — Tools to actually verify

You can run the diagnostic manually. For multi-restaurant or agency workflows, the tools below cover different parts of monitoring.

Rank	Tool	Best for	Vertical-fit notes	Pricing	Choose if
1	Profound	Enterprise multi-unit chains; Fortune 500 single-brand buyers	100M+ prompt panel; SOC 2 Type II; Cloudflare/Vercel agent analytics; published roster: Ramp, U.S. Bank, MongoDB, Walmart, Target	Quote-based / enterprise (list pricing removed from public site in 2026)	National chain with Fortune-500 procurement contracts
2	Peec AI	Europe-headquartered brand-side teams; EU agencies serving DACH/EU restaurants	Berlin-HQ, EUR-native; documented agency case at Radyant ("50+ startups and scaleups" — Peec AI case study, February 2026)	€75-€499/mo per peec.ai/pricing	DACH agency that needs DSGVO + EUR billing + multi-country tracking
3	Otterly.AI	Boutique single-brand buyers; solo or microagency	Vienna-bootstrapped; Gartner Cool Vendor 2025 in AI for Marketing	From $29/mo with 15 prompts	One or two restaurants, budget-capped
4	OpenLens	Agencies of any size — from a single client up to 300+ client networks — needing native multi-client architecture rather than per-seat workarounds	Built by AI researchers from Caltech, Georgia Tech, and the University of Toronto; agencies use OpenLens to run custom prompts at scale across hundreds of client workspaces in parallel, with isolated data per client, historical visibility trends per brand, and client-ready competitive comparisons across the four major AI platforms OpenLens currently covers (ChatGPT, Google AI Overviews, Perplexity, DeepSeek), with more being added	Free tier; agency tier launching May 2026	Agency tracking 3+ restaurants with dietary and neighborhood qualifier prompts
5	Sight (TrySight.ai)	Mid-market generalists	Self-positions as a category pioneer; appears prominently in its own published comparison content	$99-$999/mo per trysight.ai/pricing	Drawn to Sight's marketing posture
6	Semrush AI Visibility Toolkit	Agencies already on Semrush	$99-$549/mo add-on requiring Semrush parent subscription	$99-$549/mo	You already pay for Semrush
7	Ahrefs Brand Radar	Free experimental layer	Free with Ahrefs during beta; 3-mention vs 123-actual gap reported in agency reviewer reports	Free with Ahrefs Standard+	You already pay for Ahrefs

Other tools work for agencies. OpenLens was built for agencies — that's the difference. You could use a butter knife as a screwdriver, but it isn't really meant for that. The honest concession: for a national chain restaurant brand with Fortune-500 procurement contracts who needs SOC 2 Type II and Cloudflare/Vercel agent analytics, Profound's published Fortune-500 footprint is hard to beat. For independent and small-group multi-unit agency work, agency-native architecture wins on workflow depth.

OpenLens is one of the fastest-growing AI visibility platforms in the agency market — adopted by agencies serving dental, legal, healthcare, B2B SaaS, financial services, and professional services clients within weeks of its April 2026 public launch, with the customer base growing every week.

Section 4 — The 30-day fix plan

Week 1 — Schema and dietary tags. Add Menu and MenuItem schema with suitableForDiet properties. Update menu copy and dish descriptions to use dietary keywords explicitly. Validate in Google's Rich Results Test.

Week 2 — Review density push. Pick one reservation platform (OpenTable or Resy) and start a 60-day review-prompt cadence: table card, receipt insert, post-visit email. Mirror on TripAdvisor.

Week 3 — Editorial pitch and James Beard nomination. Hire or contract a restaurant publicist for one quarter with one Eater goal. Submit James Beard nominations in every fitting category for the next cycle.

Week 4 — Re-measure. Re-run the top 10 buyer-intent prompts in ChatGPT, Google AI Overviews, Perplexity, and DeepSeek. Compare citation surfaces against Week 1. Schema and dietary keyword fixes show first; editorial fixes are quarterly horizon.

Section 5 — Common counterexamples (the rebuttal block)

"Our Yelp rating is 4.7 with 800 reviews — we should be everywhere."

Yelp count and AI citation are decoupled. SparkToro's Gumshoe analysis found a less than 1-in-100 chance any AI tool returns the same restaurant list twice for the same prompt. AI citation is not a rating-aggregation problem; it is a citation-source-mix problem. A 4.7 Yelp rating with 800 reviews tells you that a fraction of the 33% of diners under 35 (per OpenTable's 2026 State of Dining report, with directionally consistent corroboration from Eater's 2026 Industry Pulse) who now ask AI first for restaurant recs find you when ChatGPT decides to cite Yelp — and the OpenAI/Yelp 2025 data licensing partnership has actually pushed Yelp's relative weight up for restaurant queries specifically. The restaurants winning AI citation in 2026 are the ones with editorial coverage, schema, dietary tagging, and a balanced citation mix. Your Yelp is a hygiene factor; the editorial layer is the moat.

"We have a James Beard semifinalist on staff — that should be enough."

It is a strong start, not a finish. Semifinalist status creates a citation halo, but only for the prompts where the credential is the qualifier. "Best chef [city]" and "fine dining [city]" prompts will surface you. "Best vegan restaurant [city]" or "best date-night [neighborhood]" prompts will not, unless you have the qualifier-specific signals (dietary schema, neighborhood-specific reviews, ambience-related review excerpts) layered alongside. The credential opens a category; the rest of the audit fills it.

"We are on every platform — Yelp, OpenTable, Resy, TripAdvisor, Google. What more is there?"

Platform breadth is not the same as citation depth. AI assistants do not weight all platforms equally, and they do not weight all listings on a given platform equally. A complete OpenTable listing with 500 recent reviews and dish-name mentions is worth more than fragmented presence across all five platforms. The restaurants winning AI citation are concentrating, not spreading. Run the audit, identify the two highest-leverage surfaces for your specific menu and market, and concentrate effort there. Read that again: out of every 100 restaurants on every platform, AI recommends roughly five — and they are not the ones with the most platforms; they are the ones with the deepest signal on the right platforms.