Can ChatGPT find my restaurant?

282 restaurant sites scanned: average health 54, one in three under 50, 60% with schema markup.

Only if your website gives it something to read. When someone asks an assistant for a dinner recommendation, the answer is assembled from raw HTML — the AI crawlers that feed those answers do not run JavaScript, do not scroll, and do not look at your photos. Our corpus includes 282 scanned restaurant websites, and they average 54 out of 100 on the signals assistants can actually use, with 34% under 50.

What 282 restaurant website scans show

Restaurants are actually ahead of most sectors on structured data: 60% of the scanned sites had schema.org business data on the homepage, against a 46% corpus-wide average — booking platforms deserve much of the credit. The weak points are elsewhere: 32% had no meta description for an answer to quote, only 52% had SPF on the domain and just 28% had DMARC. Mobile viewport was near-universal at 90%, and the cut's average speed score was 90.

The failures that hide a restaurant specifically

A restaurant's answerable facts are few and concrete: address, opening hours, phone, price range, cuisine in the description. Two failures remove them. Menu-platform sites often render everything with JavaScript, so the crawler receives an application shell with no address in it. And schema markup that exists but is empty — a LocalBusiness block with blank hours — reads as missing, because a crawler cannot quote an empty string. If your hours only exist inside a PDF menu or an embedded widget, no assistant has ever read them.

What to check before dinner service

Load your own homepage with JavaScript turned off: if the address and tonight's hours are not in what remains, they are invisible to the crawlers. Then check robots.txt for AI crawlers blocked by name — some platforms block them by default. The free scan runs both checks plus structured data, field by field, in about a minute, with no account.

Straight answers

Do restaurants do better or worse than other businesses?

Better on structured data — 60% of the 282 scanned restaurant sites had schema markup, against 46% corpus-wide — and roughly average overall, at 54 out of 100 with 34% of sites under 50.

My restaurant is all over Google Maps and Instagram. Isn't that enough?

Those are different channels read by different systems. An AI assistant answering a where-should-we-eat question builds from your website's raw HTML; SOCi's 350,000-location study found only 1.2% of local businesses are ever recommended by ChatGPT.

What single fix helps a restaurant most?

Getting the concrete facts — full address, opening hours, phone, price range — into the served HTML, ideally as one populated LocalBusiness JSON-LD block. That is the exact set of fields the scan reads one by one.

Will the scan hit my site hard during service?

No. A scan is at most thirteen requests, made one at a time, triggered only when a person asks for it — and the same address can be scanned at most once every three minutes.

Related

Where does your site stand? The scan is free, takes about 60 seconds, and needs no account. It fetches your homepage exactly the way GPTBot, ClaudeBot and PerplexityBot do and tells you what they receive.