Article
Good reviews may not be enough: keep your restaurant facts current
A August 2026 audit of AI venue recommendations suggests strong ratings do not protect you from being left out of an answer or described with stale facts, so the practical response is a routine for keeping your hours, menu, prices and booking links accurate.
Owners of independent restaurants, cafes, bars and event venues in the United States

- Supporting sections
- 5
- Questions answered
- 9
- Sources
- 4
- Research captured
- 2026-10-01
A great rating does not help a customer who is told you are closed, sent to an old menu, or given a dead reservation link. Restaurant and venue owners now have a clearer reason to treat those basic facts as part of customer acquisition rather than clerical cleanup.
A new audit of AI venue recommendations, published in August 2026, found that many established businesses never entered the assistants' answers at all, and that the assistants often disagreed about which places to suggest. The owner decision that follows is straightforward: maintain one reliable source of truth, then check for factual dead ends instead of assuming reputation alone will carry the business into every answer.
This page reflects evidence gathered between August 1 and August 15, 2026, and is dated August 15, 2026. It is the first edition on this topic, so it cannot describe a month-over-month change. It also cannot say how quickly any assistant absorbs an update. What it can do is show where the avoidable ambiguity usually sits and how to remove it.
The argument
The thesis
Being well reviewed and being included in an AI answer are different problems, and stale or contradictory facts are the failure an owner can most directly fix. Keep one owner-approved facts sheet, make your website and Google Business Profile match it after every change, and measure whether more customers reach your menu or finish a booking.
Being well reviewed and being included are different problems
The August 2026 preprint examined 2,208 search-grounded responses from ChatGPT, Claude, Gemini, and Perplexity against 4,776 food-and-drink venues in Canggu and Ubud, Bali. Within those two tourism-heavy markets, 85.6% of venues were never recommended by any tested system. Even among established venues with at least 50 ratings, 72.6% were never recommended.
The useful distinction is between getting into an answer and placing first once included. Review volume, an owned website, listed prices, and third-party mentions were positively associated with entry. Star rating was not positively associated with entry, although it did help predict first position among venues that were already recommended. These are associations, not proof that adding a field or earning another mention will cause inclusion. The study is also a preprint covering two Bali markets, so it is not a benchmark for your city.
The systems showed little agreement over their top-20 choices. A one-time win in one assistant therefore says little about your visibility in another. More importantly, the audit recorded 93 recommendations of permanently closed venues. Outright fabrication was rare; stale information was the more common practical failure observed.
For an owner, that shifts attention away from trying to please one assistant and toward removing contradictions and dead ends at the sources you control. A strong rating cannot rescue an old menu, a dead booking link, or the wrong hours at the moment a customer is deciding.
What the August 2026 Bali audit measured
| Measure | Result |
|---|---|
Search-grounded responses examined | 2,208 (ChatGPT, Claude, Gemini, Perplexity) |
Food-and-drink venues in the census | 4,776 in Canggu and Ubud, Bali |
Venues never recommended by any tested system | 85.6% |
Venues with at least 50 ratings never recommended | 72.6% |
Recommendations of permanently closed venues | 93 |
Agreement across systems on top-20 choices | Little |
Associated with entering an answer (positive) | Review volume, owned website, listed prices, third-party mentions |
Star rating | Not positively associated with entry; did help predict first position among venues already recommended |
Source: 'Invisible to the Machine: Auditing AI Restaurant, Cafe, and Bar Recommendation Against a Complete Market Census,' arXiv preprint, accessed August 15, 2026. Findings are associations from two tourism-heavy Bali markets, not causal effects and not a rate for other markets. The figures describe whether a venue was recommended at all, not how often it appeared in positions 1 to 3.
Build one source of truth, then make each surface match
Start with a simple owner-approved facts sheet for every location. Include the real-world name, precise address, primary category, phone, current hours, closure status, menu and prices where relevant, reservation or ordering links, and the date each fact was last checked. Compare that sheet with your location page and your Google Business Profile whenever an operational change happens.
Google currently tells restaurant owners to maintain those core profile details, and its restaurant guidance emphasizes accurate contact details, hours, links, attributes, photos, and menus. Its representation rules require the profile to match the real-world business. Google also identifies searches, menu clicks, and completed bookings as measurements owners may be able to watch.
Hours need judgment. Google's representation guidelines say restaurants should use dine-in hours as primary hours when available, followed by takeout, drive-through, or delivery hours when dine-in does not apply. Event venues with schedules that vary generally should not publish regular hours. A tidy-looking but misleading schedule is worse than an honest absence.
On the website, appropriate LocalBusiness or Restaurant structured data can state hours, menu URL, cuisine, phone, location, and price range in a machine-readable form. Use the most specific business type, validate the markup, and inspect the rendered page. This helps describe the business; Google does not promise a special search appearance, and its documentation does not establish visibility in AI assistants.
None of this is proof that completing the audit raises rankings, bookings, revenue, or AI visibility. The case for it is narrower: wrong hours or a dead booking path can stop a customer before a visit, regardless of how the business was discovered, and consistent visible content with valid markup reduces ambiguity at a source you own.
Facts to keep on the owner-approved sheet, and where they must match
| Fact | Why it matters | Where to reconcile |
|---|---|---|
Real-world name, address, primary category | Google requires the profile to represent the real-world business | Location page, Google Business Profile |
Phone | A wrong number is a dead end before the visit | Location page, Business Profile, structured data |
Current hours and closure status | Hours and closures are the facts assistants most visibly got wrong; the audit recorded 93 recommendations of closed venues | Location page, Business Profile, structured data |
Menu and prices | Menu URL and price range can be stated in structured data; menu clicks are a measurement Google points to | Menu page, Business Profile, structured data |
Reservation or ordering links | A dead link stops the booking; completed bookings are a measurement Google points to | Location page, Business Profile |
Date each fact was last checked | Shows which facts are due for review | Facts sheet |
Built from Google Business Profile Help guidance for restaurants, Google's representation guidelines, and Google Search Central LocalBusiness structured data documentation, all accessed August 15, 2026.
A maintenance rhythm worth testing
Assign one person to update the facts sheet, website, and Business Profile after any of these events: a closure, a seasonal-hours change, a menu or price change, a sold-out date, or a reservation-provider switch. One owner for the routine matters because the audit's practical failure was stale information, and stale information usually comes from a change that reached one surface but not the others.
Once a month, ask the same small set of authentic customer questions in Search, Maps, and whichever assistants your customers may use. Log factual errors and dead ends rather than inventing a visibility score, and correct the sources you control.
Track menu clicks and completed bookings where they are available. If your facts are consistent but customer actions do not improve, do not assume more profile fields are the answer. Test the offer, the page experience, or the booking path separately.
Channels where no reliable evidence was available this month
This edition does not make claims about four other channels. There was not enough reliable evidence on paid search, Local Services Ads, Google AI Overviews, or AI assistants as a group to say anything useful about them this month, and no comparable evidence exists to describe a month-over-month change. A measurement gap should not be mistaken for evidence that a channel improved or declined.
For each of them, use current account-level evidence before changing spend, content, or operating priorities. The Bali audit is the only assistant evidence on this page, and it covers two markets and four named assistants, not the channels listed below.
Channel status for August 2026
| Channel | Evidence this month | Month-over-month comparison | What to do |
|---|---|---|---|
Paid search | Not enough reliable evidence to make a claim | Insufficient evidence | Use current account-level evidence before changing spend, content, or priorities |
Local Services Ads | Not enough reliable evidence to make a claim | Insufficient evidence | Use current account-level evidence before changing spend, content, or priorities |
AI Overviews | Not enough reliable evidence to make a claim | Insufficient evidence | Use current account-level evidence before changing spend, content, or priorities |
AI assistants | Not enough reliable evidence to make a claim beyond the Bali audit described above | Insufficient evidence | Use current account-level evidence before changing spend, content, or priorities |
Evidence window August 1 to August 15, 2026. This is the first edition on this topic, so no previous-edition comparison is available.
How the evidence was assembled and what it cannot tell you
This page combines one current preprint about AI venue recommendations with current Google guidance for Business Profiles and LocalBusiness structured data. The two kinds of source answer different questions: the audit describes observed recommendation patterns, while Google's documents describe supported fields and representation practices. Public sources were included if they were accessed during the August 1 to 15, 2026 window and directly supported the owner decision.
Both findings are best read as inference at low confidence. They have not been tested against a second independent source. The audit supports source-of-truth maintenance as a sensible risk-control step, but it does not show that profile updates cause assistant inclusion, and it does not establish a rate for markets outside Canggu and Ubud. Google's documents support the maintenance workflow, not a claim that completing it improves ranking, bookings, revenue, or AI visibility.
Paid search, Local Services Ads, and Google AI Overviews were left out of the analysis because no relevant measurement was available. No client observations or identifying business details are used, and this page does not present a client benchmark or claim any internal corroboration of the audit's results.
We also do not know how quickly any assistant will absorb an update. What an owner can do is reduce avoidable ambiguity: keep the facts current, remove dead ends, and measure whether more customers reach the menu or finish a booking.
Takeaway
Strong reviews are necessary but not sufficient. In the Bali audit, most venues were never recommended and closed venues were still being suggested, so the controllable risk is stale or contradictory facts. Keep one approved facts sheet per location, make your website and Google Business Profile match it after every change, check a fixed set of customer questions monthly, and watch menu clicks and completed bookings.
Start here
Write the facts sheet
List name, address, category, phone, hours, closure status, menu and prices, booking links, and the date each was last checked, for every location.
After every change
Reconcile the website and Business Profile
Update both after any change to hours, menu, prices, booking links, phone, address, or closure status. Use dine-in hours as primary hours for restaurants, and do not publish regular hours for event venues with variable schedules.
Monthly
Ask the same customer questions
Check Search, Maps, and the assistants your customers use. Log factual errors and dead ends, then fix them at the sources you control.
Measure
Menu clicks and completed bookings
If facts are consistent but customer actions do not improve, test the offer, page experience, or booking path separately.
Questions
Frequently asked
If my reviews are strong, do I still need to worry about AI recommendations?
Does the study show how likely my restaurant is to be recommended?
Do the AI assistants agree with each other?
Do assistants make up restaurants?
What should go on the facts sheet?
Which hours should a restaurant list on Google?
Will adding structured data get my restaurant into AI answers?
How do I know whether the maintenance is paying off?
What about paid search, Local Services Ads, and AI Overviews?
Keep reading
Related
4 sources, captured 2026-10-01 · show· hide
- Invisible to the Machine: Auditing AI Restaurant, Cafe, and Bar Recommendation Against a Complete Market Census (arXiv preprint, accessed August 15, 2026)
- Get started with a Business Profile for your restaurant (Google Business Profile Help, accessed August 15, 2026)
- Guidelines for representing your business on Google (Google Business Profile Help, accessed August 15, 2026)
- Local Business (LocalBusiness) Structured Data (Google Search Central, accessed August 15, 2026)
Vendor pages change. Figures are as published on the capture date; confirm before deciding.