Skip to content

Research 8 min read Updated September 26, 2026

What does AI cite when people search for a restaurant?

In short

On 2026-09-26 we ran five restaurant prompts through gpt-5.1 with web search on. It attached 51 citations, the most of any category, and they were mostly best-of list sites, city guides, magazines, OpenTable, Reddit and food media rather than restaurants' own pages. Calgary brunch drew fifteen citations from nine hosts; Chicago deep dish cited Eater, The Infatuation, Axios and Wikipedia alongside one pizzeria's site.

Key facts

  • Five prompts were run once each on 2026-09-26 through gpt-5.1 with the web search tool on: best Italian restaurant in Toronto for a birthday dinner, best sushi in Vancouver, brunch spots in Calgary, best deep dish pizza in Chicago, and best BBQ in Dallas.
  • The model attached 51 citations, more than any other category in the study, and nearly all of them were to list sites, city guides, magazines, booking platforms, food media and Reddit rather than to restaurants' own websites.
  • Calgary brunch cited fifteen times across avenuecalgary.com, calgarylocale.com, bests.ca, opentable.ca, visitcalgary.com, reddit.com and others; Vancouver sushi cited twelve times across eatingvancouver.ca, vanmag.com, vancouverplanner.com, vancouverknow.com, bests.ca and reddit.com.
  • Chicago deep dish cited chicago.eater.com, theinfatuation.com, axios.com, en.wikipedia.org, openpizzamap.com, mlb.com and one pizzeria's site, georgesdeepdish.com; Toronto Italian cited opentable.ca, tastingmenu.ca, restaurantsforkings.com and one venue's site.
  • Dallas BBQ cited six times across firstindallas.com, verybestoftexas.com, restaurantji.com and an unrelated title company page, and named no restaurants in bold at all.

What we asked and how the study works

On 2026-09-26 we sent five restaurant prompts, one run each, to gpt-5.1 with its web search tool on, and recorded every URL the model attached to its answer. We sorted each cited host by type with our own rules: a restaurant's own site, a directory, booking or review site, a news, guide or reference site, Reddit, or social. The sorting is ours and worked poorly here, because most of the cited hosts were local list and guide sites that our simple rules filed as business sites; the hosts themselves are what the model attached.

The prompts were the kind people type on a Friday: best Italian restaurant in Toronto for a birthday dinner, best sushi in Vancouver, brunch spots in Calgary, best deep dish pizza in Chicago, and best BBQ in Dallas.

What each prompt cited

Toronto Italian: seven citations, to opentable.ca, tastingmenu.ca, restaurantsforkings.com and casaloma.ca, a venue. The answer named several restaurants by neighbourhood and grouped them by occasion, including hotel dining rooms for groups.

Vancouver sushi: twelve citations, to eatingvancouver.ca, vanmag.com, vancouverplanner.com, vancouverknow.com, bests.ca, reddit.com and two file-hosting hosts, with the answer grouping restaurants by price tier from cheap to omakase.

Calgary brunch: fifteen citations, to avenuecalgary.com, calgarylocale.com, bests.ca, opentable.ca, visitcalgary.com, urba.now, yusker.com, reddit.com and even calgaryzoo.com, and the answer named a list of diners and cafes. Chicago deep dish: eleven citations, to chicago.eater.com, theinfatuation.com, axios.com, en.wikipedia.org, openpizzamap.com, mlb.com and georgesdeepdish.com, naming the classic and newer spots. Dallas BBQ: six citations, to firstindallas.com, verybestoftexas.com, restaurantji.com and protitletx.com, and the answer named no restaurants in bold.

The pattern: the model reads the lists, not the restaurants

In every city the model built its answer from the pages that already rank restaurants: city magazines, local best-of sites, food media, OpenTable, tourism boards and Reddit. Restaurants' own websites were cited twice in 51 citations. The model then reorganized the lists by occasion, price tier or neighbourhood, which is the value it added.

By our sorting, 38 citations landed in the business bucket, but almost all of those were list and guide sites rather than restaurants; seven were news or reference, three were directories or booking, three were Reddit. Read honestly, roughly 45 of 51 citations were third-party lists and guides.

What a restaurant can do with this

A restaurant does not get cited for its own site here; it gets named because a list the model trusts named it. The work is therefore to be on those lists: the city magazine's best-of, the local food sites, Eater and The Infatuation in US cities, the tourism board's guide, OpenTable with current availability, and the Reddit threads where locals argue. Reviews and a current Google Business Profile feed those lists indirectly.

The restaurant's own site still matters for the second click, when the model or the diner opens it for hours, the menu and booking, but it is not where the answer comes from. SearchPod works on the list and guide presence for restaurant clients, keeps the booking and profile listings current, and tracks which prompts name the restaurant with a fixed prompt panel.

What this cannot tell you

A single model, a single run of each of five prompts, on one date. Another day, model or wording would produce different citations, and our host sorting was weak in this category, which we say plainly above. The restaurants the model named are recorded only as what it said; nothing on this page verifies or endorses any of them.

Related questions

Sources

one model with web search, one run per prompt, one date; host types are sorted by SearchPod's own rules and were unreliable for this category; names are recorded as the model wrote them and are not verified or endorsed; SearchPod sells AI visibility work.

Want to know how your business shows up?

Get a free, no-obligation proposal within one business day. We look at your site and your market and tell you plainly what we would do, and what we would not.

Get your free proposal

Keep reading

More in Research

All 37 in Research