How often AI named one of the groups first
Ask about a signature dish
56%named first
93% named / 56% first. 72 answers.
Signature-category questions (best seafood, best tapas, best sports bar, best BBQ).
Ask about an occasion
7%named first
40% named / 7% first. 60 answers.
Occasion questions (a private dinner for 30 to 50, date night, a client dinner, a family dinner).
How often a platform put one of these groups FIRST when asked where to host a private dinner for 30 to 50 people.
The finding
Six restaurant groups that operate in metro Atlanta. Between them they run more than forty restaurants, and every one of them takes private events. We asked four AI platforms, ChatGPT, Google's AI (AI Mode), Grok and Perplexity, forty questions a diner in Atlanta actually asks, from an Atlanta location, between October 23 and 29, 2026. One hundred sixty answers.
When the question was about what a group is known for, the platforms knew. Best seafood in Atlanta. Best Spanish tapas. Best sports bar. Best BBQ. The groups were named in 67 of 72 answers and named first in 40.
When the question was about an occasion, the same groups mostly disappeared. Where can I host a private dinner for 40? Best date-night restaurants? Where should I take a client to dinner? Named in 24 of 60 answers, first in 4.
Figure 1
The same groups, by the kind of question
Share of answers that named one of the six groups. 160 answers, four AI platforms.
- Named first
- Named, not first
- Not named
See the numbers as a table
| Question type | Answers | Named | Named first |
|---|---|---|---|
| Signature dish or feature | 72 | 67 (93%) | 40 (56%) |
| General | 28 | 20 (71%) | 3 (11%) |
| Occasion | 60 | 24 (40%) | 4 (7%) |
The private-dining question alone: named in 11 of 24 answers, first in 1. ChatGPT named one of the six groups once in six private-dining questions, and then only in an add-on line under its main list. Two of the six groups were never named for private dining by any platform, and both have a private-events page on their website.
Figure 2
Private dinner for 30 to 50: 24 answers, 1 first place
Each dot is one answer. Six private-dining questions per platform, one per group.
- Named first
- Named, not first
- Not named
Date night was worse than private dining: named in 6 of 16 answers, never first. The one occasion the groups won was a client dinner, where one group's restaurants were named first in three of four answers.
Figure 3
The occasions, one by one
Answers that named one of the six groups, by occasion.
- Named first
- Named, not first
- Not named
The 60 occasion answers also include one family-dinner question: 4 answers, none named.
See the numbers as a table
| Question type | Answers | Named | Named first |
|---|---|---|---|
| Client dinner | 16 | 7 (44%) | 3 (19%) |
| Private dinner for 30 to 50 | 24 | 11 (46%) | 1 (4%) |
| Date night | 16 | 6 (38%) | 0 (0%) |
| Family dinner | 4 | 0 (0%) | 0 (0%) |
So the platforms will recommend a group for an occasion. They just rarely did.
What the answers point to
The two kinds of questions drew on different sources. For "best seafood," the answers cited critics and review sites: The Infatuation, Eater Atlanta, the AJC, Atlanta Magazine, the Michelin Guide, OpenTable. Those lists already rank the groups' flagship restaurants.
For "private dinner for 40," there is no critics' list. The answers cited OpenTable's private-dining listings, PartySlate, The Vendry and restaurants' own private-dining pages, and the rooms they recommended nearly always came with a room name and a seated capacity.
Best seafood
Critics and review sites
- The Infatuation
- Eater Atlanta
- The AJC
- Atlanta Magazine
- The Michelin Guide
- OpenTable
Private dinner for 40
Listings and room pages
- OpenTable private-dining listings
- PartySlate
- The Vendry
- Restaurants' own private-dining pages
Our read, not a measurement
Our read, and it is a read, not a measurement: a room is easier to recommend when its name and size are written down where a platform can find them. One of the two groups never named has a private-events page that shows 48 words to a reader without JavaScript; the room sizes load only with it.
But that is not the whole story. Another group in the study states capacities on its pages and was still left out by three of the four platforms. Outside listings appear to matter as much as the group's own page.
What a restaurant group can check
Read your private-dining page with JavaScript switched off.
If the room names, seated and standing capacity, minimums and booking contact are not there in plain text, some AI crawlers do not see them.
Check that the same numbers appear on the outside listings the answers cited, such as OpenTable's private-dining section and PartySlate.
Run the test yourself.
Ask ChatGPT "where can I host a private dinner for 40 in [your neighborhood]" and see whether you are in the answer.
How we tested
40 questions, 4 platforms, 160 answers, one run each
- Six restaurant groups that operate in metro Atlanta. Groups are not named; some are prospects of GW Labs, and the point is the pattern.
- 40 unbranded questions about the Atlanta market (18 signature-category, 15 occasion, 7 general "best restaurants in X right now"), written in a diner's words. Questions that named a group or one of its restaurants, and questions about other cities, were asked too and are not counted here.
- Four platforms: ChatGPT (Temporary Chat), Google AI Mode (a clean test account, personalization off), Grok (Private chat), Perplexity (incognito). One session per question, one run each. Claude was run on two of the six groups and is left out to keep the counts equal.
- One exception to private mode: for two of the six groups, ChatGPT's Temporary Chat was in its Personalized setting, and one answer showed it. That affects 11 of the 160 answers. Our runs since November 2 use the Unpersonalized setting.
- Location: Atlanta, Georgia, checked before each platform. Dates: October 23 to 29, 2026.
- Scoring: a group is "named" if any of its restaurants appears in the answer; "first" if one of its restaurants is the first recommendation. One answer that addressed the wrong city is counted as not named.
- Single-run baseline. Answers change from day to day.
- Full method: gwlabs.app/method
Want to know how the platforms answer the ten questions your customers ask about you?
GW Labs runs it on five platforms and sends a short report, free. No call needed.