TLDR: Give every location one approved fact sheet that separates what the group shares (name, story, reservation pattern) from what is true only there (dishes, hours, price, setting, access), and write every page and listing for that branch from it. Then ask the same questions about each branch every month and log the mix-ups.
It is 8:30 on a Friday night and you are at the pass when a table of four asks for the crab linguine. They came because an AI assistant told them this was the place for it. The dish is real, but it is on the menu at your other location across the river, and tonight your floor team has to explain the mistake.
The assistant did not invent the linguine. The likeliest explanation is a mix-up between two branches, because the pages and listings it can find may describe both sites in the same words or describe one site in ways that contradict each other. In our view that makes inconsistency across locations more a description and menu-data problem than a ranking problem. Many groups read brand consistency as identical copy on every site. We think identical copy can cause as many wrong answers as letting each venue improvise, because an assistant can only tell two branches apart by what differs between them.
Why do AI assistants mix up your locations?
An assistant can mix up branches because it builds each answer from several sources at once, and for a group many of those sources describe more than one site. Google says AI Overviews and AI Mode may use a "query fan-out" technique, "issuing multiple related searches across subtopics and data sources" to develop a response. It also lists keeping Business Profile information up to date among the fundamentals that still apply (Google Search Central).
ChatGPT works in a similar way. OpenAI's help page explains that ChatGPT search rewrites a question into one or more targeted queries for its search partners. It may share a general location with providers of nearby business listings, and it can show reservation times from OpenTable, Resy or Yelp. The same page warns that "Search results and citations can be incomplete, outdated, or incorrect" (OpenAI Help Center). Which listings matter differs by market; our guide on whether diners use ChatGPT instead of Yelp sets out the reservation and review platforms that appear in the US, the UK and Germany.
The most useful evidence on what goes wrong comes from a preprint that audited 2,208 search-grounded answers about 4,776 cafes, restaurants and bars in two Bali markets. Outright fabrication was rare, but the systems recommended permanently closed venues 93 times, and the author concludes that "staleness, not hallucination, is the practical failure mode" (Pitenin, arXiv, August 2026). The study is not peer reviewed and covers no US, UK or German market, so it gives no rate for your market. Read it as a pointer: in that census the practical failure was stale data, not invention, so the first place to look for a wrong answer is what is published.
For a group, that leaves two ways to fail. Sameness: two branch pages carry the same paragraph, so nothing tells the sites apart. Contradiction: one branch's sources disagree, so the answer can follow either version. Both also explain why one site can appear in answers while another stays invisible, which our guide to why some of your restaurants show up in AI answers covers. Both can start in the sources, which raises the next question: which of yours disagree?
Which sources disagree about each branch?
Four kinds of source usually describe each branch, and a mix-up can start where two of them disagree or where two branches share one description. Take a practical example: a two-location group with a flagship in SoHo and a newer location in Williamsburg, where the crab linguine is served only in SoHo. An audit of the Williamsburg sources might look like this.
| Source | What it says | Problem |
|---|---|---|
| Website location page | Group About text and a PDF of both menus | Nothing marks which dishes are Williamsburg's |
| Google Business Profile | Description copied from SoHo; menu link to a delivery app | Names SoHo's linguine; Google does not allow that menu link |
| Reservation listing | Correct hours, price range from before the menu change | Price contradicts the website |
| Press and guide pages | A review of SoHo headed with the brand name only | Easy to read as a review of the whole group |

The website page and the press review are sameness problems: they describe both sites as one. The profile and the reservation listing are contradictions: they disagree with what is true at the branch. Google's own rules already point at the fixes. The profile should link to "a website that represents your individual business location". Menu links cannot be direct links to third-party ordering or delivery services, and the description allows no links and must not focus on promotions, prices or offers (Google Business Profile guidelines). So the Williamsburg profile should link to the Williamsburg page and the Williamsburg menu, and its description should name what is true in Williamsburg. Google adds that businesses with complete and accurate information are more likely to show up in local results (Google Business Profile Help), which is a second reason to fix the Williamsburg profile first.
Fixing each listing one by one does not hold for long, though. The next menu change puts the old copy back somewhere. What holds is a rule for which facts must match across the group and which must differ.
What should be the same everywhere, and what must be local?
Lock the facts that identify the group, and keep local the facts that make a guest choose one branch over another. Google decides part of this for you. Its guidelines for chains say all locations in a country must have the same name, with "The Home Depot" rather than "The Home Depot at Springfield" as the example. Locations that provide the same service must also share a category (Google Business Profile guidelines). Since the name cannot separate your branches on Google, the local facts have to carry the whole load. Google's attributes, such as outdoor seating, show on the profile and can bring a branch into searches for places with that feature (Google Business Profile Help), so they belong to each site, not to the group.
| Fact | Locked or local | Why |
|---|---|---|
| Name format on profiles | Locked | Google asks chains for one name per country |
| Main category | Locked | Same service, same category under Google's rules |
| Group story and founding | Locked | One version stops two histories blending |
| Reservation route | Locked pattern, local link | Same steps everywhere; each link opens its own branch |
| Signature dishes and menu | Local | The dish a guest travels for must sit with its branch |
| Hours and exceptions | Local | Late nights and closures differ by site |
| Price range | Local | A remodel or new menu moves one site only |
| Setting and seating | Local | Terrace, counter or private room are building facts |
| Access | Local | Wheelchair-accessible entrance and restrooms depend on the building |

Where the answer changes: a tightly standardised chain with one identical menu should say the same thing about its food everywhere, and for that chain the menu row moves into the locked column. Its problem is location facts only, which means hours, access, address and reservation link. The table above is written for the group in between, with a shared core menu and a handful of dishes per site.
A rule like this only works once it is written down for each branch and has someone who keeps it true.
What goes on a location fact sheet, and who owns it?
A location fact sheet is one approved document per branch that every website page, profile and listing copies from, with a named owner and clear triggers for an update. It needs two blocks:
- Locked block, pasted unchanged from the group master: name format, category, group story in two sentences, reservation route pattern.
- Local block, confirmed in the building: address, this branch's reservation link and menu page, dishes served only here, hours with exceptions, price range, setting and seating, access, which delivery apps carry this site. Add a description of two or three sentences that names at least two things true only here.
Group marketing owns the locked block. The general manager confirms the local block, and one person signs off each change. Update the sheet before any surface whenever the menu, hours, price range, reservation system or building changes. Our checklist for Google Business Profile governance across multiple restaurant locations covers the routine for hours, menus and access edits.
The branch page on your website should carry the same facts in its structured data. Google's local business markup example gives its restaurant a location-specific page address and a menu link (Google Search Central), so give each branch page its own. On schema.org, the Restaurant type has properties for the menu and cuisine, and parentOrganization links a branch to its group (schema.org). Google also asks that structured data match the visible text (Google Search Central), so the markup comes from the sheet too. If you have not yet decided how the group and its venues should appear as entities, our guide on whether AI sees your restaurant group as one brand covers that choice first.
A fact sheet tells you what should be true. Only the answers tell you whether the assistants now agree.
How do you check that the fix held?
Ask the same questions about each branch every month, in the assistants your guests use, and log every answer that gives one branch another's facts. A simple routine:
- Write six questions per branch that use its neighborhood: what to order, late opening on Sunday, wheelchair access, a table for six on Friday, outdoor seating, and price.
- Ask each question in ChatGPT, in Google's AI Mode and in one more assistant your guests use, and record the answer and the sources listed alongside it.
- Mark three kinds of mix-up: a borrowed fact (true of another branch), a stale fact (true of no branch now) and a wrong door (right fact, but the link or reservation route opens another branch).
- Trace each mix-up to the sources listed alongside the answer, correct the fact sheet first and then the source.
- Ask the same questions next month and compare.
For the two-location example, six questions in three assistants make 36 answers a month. Those numbers are our assumptions for illustration, not a benchmark. Answers vary from run to run. A US study by SparkToro of product and brand recommendation prompts found under a 1 in 100 chance that ChatGPT or Google's AI gives the same list of brands in any two of 100 runs (SparkToro, January 2026; one author works for an AI tracking company). So ask each question twice, which makes 72 answers. If 6 of the 72 carry a mix-up, your rate is 6 in 72, about 8%, and next month's number tells you whether the sheet worked.
A source listed beside an answer does not prove the assistant relied on it, but it is the best lead you have. A wrong-door mix-up usually needs the fix our guide to turning AI restaurant recommendations into bookings describes: each branch's reservation link checked from the answer through to the confirmed table.
This monthly check is the easiest step to drop after the first cleanup, and it is the work we do every week for restaurant groups at Schmitdy, as part of an ongoing managed service. We check the answers for every venue, trace each mix-up to the page or listing shown alongside it, and work with you to fix it. That covers Google Business Profile, Search Console, menu pages and the location pages themselves, with a person checking every change. It is ongoing because menus, hours and prices keep moving. If you want to see where your branches are mixed up today, the free AI search audit checks local facts branch by branch.
What changes for the table of four?
Once each site has its fact sheet, the Williamsburg profile names Williamsburg's dishes and the SoHo page names the linguine. The next guest who asks for it is pointed to SoHo, reserves through the SoHo link and walks through the right door. The table that came on Friday still deserves an apology, and next month's check tells you whether theirs was the last one.
Start with one branch this week. Write its fact sheet, split locked from local, and compare it with that branch's Google profile and menu link side by side.
Sources
- Google Search Central, AI features and your website, last updated 10 December 2025, checked 13 November 2026.
- OpenAI Help Center, ChatGPT search, checked 13 November 2026.
- Vladimir Pitenin, Invisible to the Machine: Auditing AI Restaurant, Cafe, and Bar Recommendation Against a Complete Market Census, arXiv preprint, 7 August 2026.
- Google Business Profile Help, Guidelines for representing your business on Google, checked 13 November 2026.
- Google Search Central, Local business structured data, last updated 8 September 2026.
- schema.org, Restaurant and Menu, checked 13 November 2026.
- SparkToro, AIs are highly inconsistent when recommending brands or products, 28 January 2026.
- Google Business Profile Help, Tips to improve your local ranking on Google, checked 13 November 2026.
- Google Business Profile Help, Manage your business attributes, checked 13 November 2026.

