On 21 October we asked ChatGPT the questions customers ask, for 30 independent venues in south-west London. 16 were named first, 11 came up behind a competitor and 2 didn't appear at all. One restaurant was listed first but not recommended: ChatGPT picked another, noting that the first one's opening hours differed across its listings. If you run three, five or ten sites, each one faces that test separately. Whether it passes comes down to something unglamorous: whether the information about that site is clear, current and the same everywhere it appears.
That's harder for a group than for a single restaurant. One restaurant has one set of hours, one menu and one Google listing to keep right. A group has all of that once per site, plus a group website that often blurs the sites together. Every new opening adds another place for things to drift.
In September we checked 100 restaurants in Notting Hill. In those preliminary findings, 40% of the websites we could assess had hours flagged as different from Google, and more than half had a menu that was a PDF, an image or missing. A group can repeat each of those problems at every site. Read the findings and method.
We've also tested it directly. On 28 September we asked ChatGPT, Gemini and Google's AI Mode four questions a hirer might ask, each one a good fit for a 550-capacity event venue in west London that we look after. That gave 12 answers. The venue appeared in none of them, while 46 other venues did. Two of the assistants even said a 300-person venue near Ladbroke Grove would be hard to find. The space existed. The information about it didn't.
What the assistants used instead was telling. Google listings were behind every AI Mode answer, and one venue with just three reviews made the list because its Google category matched a word in the question. Booking-site listings came up in 5 of the 12 answers, sometimes quoted word for word. If your own details are thin, someone else's version is the one that gets read.
Where the answers come from
AI assistants don't have a secret list of good restaurants. They build an answer from what's already out there about you: your Google listing, your own website, what the press and guides have written, your reviews, and the booking platforms you use. Where those sources agree, the assistant can describe you with confidence. Where they disagree, it tends to hedge, pick one, or leave you out.
Booking is now part of it too. Since April, Google's AI Mode in the UK can find a table and hand people straight to a booking page. It works through eight booking platforms: OpenTable, SevenRooms, ResDiary, TheFork, DesignMyNight, Mozrest, Foodhub and Dojo. (ChatGPT added table booking in August, through OpenTable, Resy and Yelp; availability varies by provider and country.)
None of this is a new channel you need to buy into. It's the same information you already publish, read more strictly.
Five ways a group goes missing
1. Every site shares one page
Many group websites have a homepage, a group menu and a "Find us" page listing every address. That reads fine to a person who already knows you. But an assistant answering "Italian in Notting Hill" needs a page about the Notting Hill site: its address, hours, menu, prices and booking link, all in one place.
Google's own guidelines ask for the same thing. The website on each listing should be one that "represents your individual business location". If every listing points to your homepage, none of them has a page of its own to back it up.
2. Hours drift, and Google now fills the gap
Hours are the easiest thing to let slip. One site changes its Sunday hours, Christmas hours go on the website but not on Google, or the group page shows one set of hours for everyone.
Since late September this matters more. When Google notifies you of a suggested change that requires your review, you have 4 days to accept or reject it. If you don't respond, Google says it "may automatically publish the update if it is supported by other publicly available information", such as your website. So if your website is wrong or vague, it can support the wrong update. Multiply that by every site, and by everyone in the team who might or might not see the email.
3. The menu is a PDF
A PDF or a photo of a menu is easy for a guest to open. It's much harder for Google or an assistant to read reliably, and it often isn't the current menu. In Notting Hill, 56% of the websites we could assess had a PDF, an image or no menu at all. Some groups go a step further, with one group menu that doesn't say which dishes are served where. Prices matter as much as dishes: half of the Notting Hill websites we could assess (51%) had no prices found. Put each site's menu on its page as plain text, with prices.
4. Names and links that wander
Google asks that locations of the same business in the same country use a consistent name, with exceptions for genuine differences in real-world branding. "Your Name", "Your Name Borough" and "Your Name Kitchen & Bar" can make one group look like three separate businesses if those variations don't reflect the names used in person. Check the link on each listing too. In Notting Hill, 86 of the 100 listings had a website domain matching the one recorded for the restaurant. Some of the others pointed to booking or social pages; a different domain isn't automatically an error. Two pointed to an old website the restaurant had since replaced.
Check the main category on each listing as well. In our venue test, a place with three reviews made an AI answer because its Google category matched the word in the question. If a site is listed as "Restaurant" when people search for "Italian restaurant" or "wine bar", it's harder to match.
5. Booking the assistant can't use
If people can only book by phone or email, or through a system Google doesn't work with, an assistant can recommend you but can't book you through that integration. In the UK, Google's AI Mode books through eight platforms, listed above. If your sites use different booking systems, check each one. One site can be bookable while the next isn't.
Opening a new site
If you're expanding, the cheapest time to get this right is before you open.
- Put the opening date on Google early. You can add an opening date up to a year ahead. Once the date is entered and the business is verified, the listing appears on Google 90 days before you open. Until opening day, the public can't suggest edits or leave reviews, so it's yours to set up properly.
- Launch the site's own page the same day. Include the address, the opening date, a sample menu with prices, and how to book. Point the new listing at that page, not the homepage.
- Use the same name and booking system as the rest of the group, unless there's a real reason not to.
- Get the press to link to the site's page. Openings are when you get the most coverage. If the articles point to a page with the right details, that's some of the best backing an assistant can find.
A 20-minute check for every site
Do this once per site, with the Google listing open in one tab and your website in another. Anything you can't tick is a fix to make.
- The name on Google matches the sign, and matches your other sites unless the branding genuinely differs.
- The main category on Google matches what people search for, such as "Italian restaurant", not just "Restaurant".
- The website link on Google goes to this site's own page, not the homepage.
- Opening hours are the same on Google, on the site's page and in your booking system.
- Christmas and bank holiday hours are set on Google as special hours.
- The menu is on the page as text, with prices, and it's the current one.
- The booking link on Google opens this site's availability, not a group page.
- Someone is named to answer Google's suggested-edit notifications within 4 days.
- Search "[your cuisine] near [the site's area]" in Google and see what comes up.
In short
Nobody can promise that an assistant will recommend you. What you can do is make sure every site has one clear, current, consistent version of itself online, so that when it is the right answer, nothing stands in the way.
If you'd rather we checked every site for you, that's what our Can They Find You? audit does. We look at each site the way Google and the AI assistants do, and tell you what to fix first.
Damoy Robertson is the founder of Red Hills Lab, a London studio that works with restaurants, bars, gyms and venues on brand, web and marketing. The Notting Hill figures come from the studio's W11 AI-readiness report, September 2026. These are preliminary findings, not fully checked by hand; differences in kitchen and venue hours can explain some flagged mismatches.
