TLDR: An agency, an in-house lead or both can manage AI visibility for a restaurant group. Whoever does it, the job is monthly work: measure a fixed set of diner questions per location, ship changes to pages, profiles and press, and report the two together. A retainer that only reports is monitoring.
You are three months into an AI search retainer. The third monthly report arrives, and it is the same chart as the last two with a new date at the top. You open the menu page for your busiest location and it says exactly what it said the week you signed. Nothing in the contract has been broken, and nothing on your pages has changed.
That gap answers the question of who manages AI visibility for a restaurant group. An agency, your own marketing team or a mix of the two can do it, but managing means measuring and changing things every month. The scope should name three things: the questions it tracks for each location, the changes it puts live on your website, Google profiles and in the press, and a report that ties the two together. If the report never points at a change, you are paying for monitoring and should price it that way.
Our view is that you should pay for shipped changes rather than dashboards. Some groups take the other view for good reasons: they buy a cheap tracking tool and do the work themselves. Both can work, provided someone owns the changes.
Who can manage AI visibility for a restaurant group?
Three setups work, and the right one depends on who can change your location pages and profiles every month.
- A specialist agency runs the whole loop. It measures, ships changes and reports, while your team confirms facts and approves copy. This suits groups with no marketer who can edit the website and every Google profile each week.
- Your own team runs it with a tracking tool. A marketing lead owns the question set and the changes, and the tool supplies the counts. This suits groups that already publish pages and manage profiles in-house.
- A split. Corporate keeps the facts, the accounts and the approvals, and an agency measures and ships. This tends to suit groups with more than a handful of locations.
If you are still choosing between one partner for the whole group and separate help for each location, our guide on whether a restaurant group should hire one AI search agency or handle it site by site covers that decision.
When is a tracking tool and a quarterly audit enough?
A tracking subscription can be all you need when your in-house team already ships location pages, menu updates and profile fixes every week, or when a strong local SEO partner does that work for you. Then the missing piece is a fixed count, and a restaurant group AI search audit each quarter can check the work from outside. A tool is not enough when nobody on your side has the time or the access to act on what it shows. Whichever setup you choose, the first thing to put in writing is what gets measured.
What should be measured every month?
A monthly retainer should measure the same written set of diner questions for each location and market, in the same AI engines, and report raw counts with their denominators and dates.
The set stays fixed because AI answers change from one run to the next. In SparkToro's January 2026 study of the three most popular AI tools in the US (ChatGPT, Claude and Google's AI), 600 volunteers ran 12 prompts 2,961 times. The authors put the chance that ChatGPT or Google's AI gives the same list of brands in any two of 100 responses at less than 1 in 100. They suggest asking 60 to 100 times and averaging. One co-author works for an AI tracking company, which the post discloses. A 2026 census of 4,776 cafes, restaurants and bars in Bali also found low agreement between ChatGPT, Claude, Gemini and Perplexity, so each engine needs its own count.
Take a hypothetical group of eight restaurants in Chicago and Austin. Say the scope sets 12 questions per location, such as brunch near Wicker Park for one restaurant and private dining for 20 in the West Loop for another, and checks each one five times a month in three engines. That is 12 Γ 3 Γ 5 = 180 answers per location and 1,440 for the group. A useful report line reads like named in 41 of 180 answers for the Wicker Park location, October 1 to 31, not visibility up 12%. These figures are our illustrative assumptions, not a benchmark or a result. Five runs is well below what the SparkToro authors suggest for a single prompt, so treat one question's count as noisy and judge each location's total across several months.
Google adds a gap of its own. It counts appearances in AI Overviews and AI Mode inside the overall Web figures in Search Console's Performance report, so they never show up as a separate line there. The question set fills that gap. It tells you where each location stands; it does not move anything, which is why the next line of the contract matters more.
What should be shipped every month?
Each month should end with a dated list of changes that are live on your pages, profiles and in the press, each tied to the question it is meant to move.
The evidence on which changes matter is still young, but it points one way. The Bali study, a preprint covering two areas of one island (Canggu and Ubud), found that venues with more reviews, a website of their own, listed prices and mentions on other websites were associated with a higher chance of entering AI answers. Star rating was not linked to getting in, although among venues that were recommended, a higher rating was linked to being named first. It also caught the four systems recommending permanently closed venues 93 times, so stale facts are a real failure, not a theory. Peer-reviewed work on generative engines by Aggarwal and colleagues, accepted to KDD 2024, found that changes to page content could raise a source's visibility in generated answers by up to 40% on the authors' benchmark. Google, for its part, says AI Overviews and AI Mode need no special optimization beyond sound SEO fundamentals.
That gives a monthly work list with an owner on your side for each line:
| Work | What goes live in a month | Your owner |
|---|---|---|
| Location pages | One page per location: address, hours, reservation link, structured data with menu URL | General manager confirms facts |
| Menu pages | Menus as text pages, not PDFs, updated when dishes change | Executive chef or operations |
| Google profiles | Hours, holiday hours, categories, photos, corrected facts | General manager; corporate applies |
| Reviews | One routine for asking every guest, and replies | General manager |
| PR and content | A pitch, article or guide answering a question in the set | Marketing lead approves |
| Technical | Crawler access, sitemap and broken link fixes | Whoever runs your website |
Google's local business structured data has a menu property that takes the full web address of a food business's menu, so a menu that sits on its own page gives you an address to point it at. On access for AI engines, OpenAI says OAI-SearchBot is the crawler used to surface websites in ChatGPT search, and it can be allowed separately from its training crawler. The review routine has to stay within the law. The FTC's Rule on the Use of Consumer Reviews and Testimonials, 16 CFR Part 465, in effect since October 21, 2024, bans fake reviews and buying reviews. An agency that offers to write or buy them is a reason to end the contract.
More is not better. Google's spam policies define scaled content abuse as generating many pages mainly to manipulate rankings without helping users. Ten useful changes tied to ten questions beat fifty thin pages.

Every line of that table needs someone with access to your accounts, which raises the question of who holds them.
Who owns access and decisions?
The restaurant group should own every account, and the agency should work on its own logins with the lowest role that lets it do the job.
On Google Business Profile, a profile can have several owners but only one primary owner. Owners can add and remove users; managers can do almost everything else but cannot add or remove users or remove the profile. Keep primary ownership in a company account and invite the agency as a manager. For many locations, Google's business groups work like a shared folder of profiles, and Google recommends them over sharing a username and password. Ownership also matters for bulk verification, which needs 10 or more profiles from the same business and is not open to agencies holding more than one business in one account.
In Search Console, only property owners can grant permissions, and a full user sees all data while a restricted user sees most of it. Google's own advice on hiring an SEO is to grant read access only while a vendor is auditing; a retainer that ships may need a full user, but ownership stays with you. In Google Analytics, adding or changing users needs the Administrator role, so keep that in-house and give the agency a lower role. The measurement belongs to you as well: ask for the question set and the raw answer counts as a monthly export to your own drive, so a new partner can start from your baseline rather than building another.

Decisions need the same clarity as logins. General managers confirm facts within an agreed number of days, corporate approves anything in the brand's voice, and the agency proposes, ships and logs. If you are putting this into an RFP, our guide on how to brief an AI search agency for a restaurant group sets out the brief. Once access and approvals are named, every delay has an owner, and that makes month three far easier to judge.
How do you judge month three and month six?
Judge both by setting the shipped list beside the question set: for each change, which question it was for, and what that question's count in ChatGPT, Google's AI and the other engines in the set did over the same window.
By month three, expect a baseline, three months of counts with their denominators, and a shipped list long enough that every location has had something change. The counts may barely have moved; the list must not be empty. By month six, you should be able to compare the questions that received work with those that did not. If the two groups look the same, the plan changes, not only the report.
Four signs mean the scope is wrong:
- A guarantee. Google says no one can guarantee a #1 ranking, and for its AI features that indexing and serving are not guaranteed. A promise of a place in ChatGPT makes the same claim about a different engine.
- A percentage with no denominator or dates. Visibility 34% means nothing until you know how many answers, which engines and which month.
- Reports with no changes. Three charts and an empty shipped list is monitoring.
- A shipped list padded with thin pages. Volume without a question behind it is the scaled content Google warns about.
If you are still at the first call, the ten questions that show whether a restaurant marketing agency understands AI search test for these before you sign.
What should a one-page monthly scope say?
Take this list into your next renewal and ask the agency, or your own team, to fill in each line:
- Questions: the written set for each location and market, fixed for at least a quarter, with any change logged.
- Engines: the named AI engines checked, the same every month.
- Counts: raw numbers with denominators and dates, per location and per engine, exported to you.
- Shipped list: every change with its date, its link and the question it serves.
- Owners: who on your side confirms facts and approves copy, and how fast.
- Access: your company holds primary ownership of every profile, Search Console and Analytics; the agency works on its own logins and loses them on exit.
- Exit: the notice period, and confirmation that the pages, content, question set and data stay with you.
Beyond an offer to write or buy reviews, two findings justify canceling: a third monthly report with nothing shipped, and numbers with no denominator or a guaranteed result.
Restaurant AI Search at Schmitdy is built on the same loop. We are a GEO and AEO specialist rather than a general SEO agency, and the service is ongoing. We check the agreed questions daily across the main AI engines and implement changes with your team every week: menus rebuilt as text pages, a page for every location, editorial pitches and review routines. Pricing starts at $1,650 a month. Each 30-day month stands alone with 30 days' written notice. Half the fee is due at the start of the month, and the other half only when one of two agreed monthly goals is met.
If you want to see the starting point first, request the free AI search audit and bring your list of locations. You will get recorded answers for the questions that matter in your market and a 30-day plan you can hold any partner to, us included.

