Opening hours, address, offering: How contradictory data costs you the AI recommendation

Why the AI punishes contradictions rather than ignoring them

Language models like ChatGPT, Gemini or Perplexity draw their answers from many sources at once: Google Business Profile, TheFork, Tripadvisor, menu portals, your own website. When these sources say the same thing, trust arises. When they contradict each other, uncertainty arises, and an AI that's uncertain prefers to recommend the restaurant whose data fits together smoothly. You aren't sorted out because of bad cuisine, but because of unclear facts. This is exactly what most restaurateurs overlook, because they think about food and not about data sets.

This is the big difference from the classic Google search. Previously the guest himself compared five links and forgave inconsistencies. Today the AI formulates a single, confident recommendation. In this one answer there's no room for '10 p.m. per the website, but 9 p.m. per Google.' The machine resolves the conflict by leaving you out and instead naming a business whose information is contradiction-free. For you as a restaurateur, that means: consistency is no longer a tidiness obsession, but a direct revenue lever.

Concretely, you encounter this with questions like: Which restaurant in Schwabing is still open after 9 p.m. today and has a table for four? Exactly here the businesses part ways. Whoever has the same kitchen-closing time listed everywhere gets mentioned. Whoever has three different times online falls out of the selection, because the AI doesn't take the risk of a wrong recommendation. It prefers to protect the user from the locked door than to protect you from the loss of revenue.

The three data points where restaurants most often fail

Opening hours are the classic. Hardly any industry changes them so often: rest day shifted, kitchen closing earlier than bar closing, lunch break, holiday rules, seasonal hours in the beer garden. Each of these subtleties has to appear identically at every location. In practice the boss maintains Google, the service staff enters something else on TheFork, and the website has been frozen since the last relaunch. And there you have three truths for one question, and the AI can't pass on any of them with a clear conscience.

The address sounds trivial, but it isn't. Move to the neighboring street, new building entrance, spelling with or without an addition like 'in the courtyard' or 'rear building,' typo in the house number. To the AI, 'Ristorante Bella at Marktplatz 3' and 'Bella Ristorante, Marktplatz 3a' are, when in doubt, two different businesses. Then reviews and signals get spread across two ghost entries, and neither is strong enough to be recommended. A clean, identical address including spelling is the basic prerequisite for all signals paying into one name.

The offering is the underestimated third point. Do you offer gluten-free pasta? Is there a vegan menu, a lunch special, weekend breakfast, a delivery service? If that only stands on a chalkboard in front of the door, it doesn't exist for the AI. And if the website still shows the old menu with dishes you haven't cooked in a year, the AI might recommend you for something you no longer have at all, and the disappointed guest writes a bad review.

A real example: the lost Friday evening

Imagine a family restaurant that's open until 11 p.m. on Fridays in summer. On Google it still shows the winter time until 9 p.m., because in autumn no one thought to change it back. TheFork shows 10 p.m., because that's where the last reservation is recorded. The website names no time at all, only the phrase 'hot food all day.' Three sources, three statements. Solvable for a human, a warning signal for an AI.

Now if someone asks their AI on Friday at 9:30 p.m. for a restaurant that still serves families with children at that hour, this place drops out. Not because it's closed – it is open – but because the data situation suggests it might be closed. The competing restaurant that has its 11 p.m. time listed the same everywhere gets the table for four. The revenue is gone, and the restaurateur never learns why it was emptier than expected that evening.

The bitter punchline: the restaurant did nothing wrong except forget its data maintenance. The cuisine was top, the service warm, the price fair. But all of that was invisible, because the machine had already bailed out at the fact-check. Exactly for this reason, data consistency isn't an IT topic you can delegate and forget, but belongs on the same note as the grocery shopping list.

How to create a single source of truth

The most important step is a simple document that stands nowhere on the internet: an internal master-data list. On it belong the exact name, exact address with spelling, phone number, all opening and kitchen hours including rest day and holiday logic, plus the core offerings like vegan, gluten-free, lunch special, terrace, dogs allowed. This one list is your truth. Every portal will in future be reconciled against it, not the other way around. Without this reference you maintain from memory each time, and that's how contradictions creep in.

After that you go through the platforms once thoroughly, portal by portal, and enter exactly the same values everywhere. Order by impact: Google Business Profile first, then reservation portals like TheFork or OpenTable, then review portals like Tripadvisor and Yelp, then delivery services, and last your own website. Pay meticulous attention to small things like umlauts, abbreviations in the street name and uniform time formats. What a human dismisses as a detail, the machine treats as a difference.

Why your website has to remain the anchor source

Portals come and go, change their rules and display your data however it suits them. Your own website is the only source that belongs entirely to you and that you can steer exactly at any time. That's why it should be the anchor source that all other information aligns with. When AI systems, in doubt, look for an official source, that's ideally your domain, and exactly there the times, address and offering have to be flawless.

Technically, structured markup helps you enormously. With machine-readable information per the restaurant schema you can store opening hours, cuisine, price level and address so that search systems read them unambiguously. This isn't marketing frippery, but the language in which machines understand your facts most reliably. A web developer sets this up in a few hours, and afterward you have a source the AI can hardly misunderstand.

The discipline afterward is important. When something changes, it changes first on your website and in the master-data list, then everywhere else. That way the anchor stays stable, and you avoid that of all sources the one the AI trusts most is the outdated one. A well-maintained website is, in the GEO era, no longer the digital business card, but the foundation of your findability.

The holiday and season pitfall

Restaurants live on exceptions, and exactly those give the AI trouble. Closed on Easter Sunday, only lunchtime on Christmas Eve, closed for the winter break in January, longer terrace hours in summer. If these special rules only exist in the restaurateur's head, the AI shows the normal hours and sends guests to a locked door, or it overlooks that you're open by exception on a holiday while the competition is closed. Both cost you, once the review, once the full house.

The solution is well-maintained special hours directly in the portals. Google and the big reservation platforms offer fields for deviating holiday hours. Enter them in good time, ideally for the whole year in one go. Set yourself a fixed appointment, for example in January, on which you store all holidays and vacation periods for the coming twelve months. This one hour of work prevents dozens of disappointed guests and protects your recommendation rate on exactly the high-revenue days.

How to check what the AI really says about you

You don't have to guess whether your data is right, you can test it. Ask ChatGPT, Gemini and Perplexity about your restaurant yourself, the way a guest would. 'Name me good Italian restaurants in my city,' or directly: 'When is the Ristorante Bella at Marktplatz open and does it offer vegan dishes?' Compare the answers with your master-data list. Every deviation is a trace to a source that delivers wrong data.

If you find a mistake, work backward. If the AI names a wrong time, check Google first, then the reservation portals, then the website, until you've found the source of the contradiction. Often it's exactly one outdated platform that poisons the whole tone. Repeat this test every few weeks and especially after every change to your times or your offering. That way you notice yourself whether a correction really landed everywhere.

Treat this check like a hygiene inspection in the kitchen: it's not a one-off project, but a routine. Whoever goes through completely twice a year and briefly cross-checks after every change of hours has a crystal-clear, contradiction-free presence toward the AI. And clarity is, in the generative era, the currency in which recommendations are paid out.