What your restaurant website needs in 2026

In September 2026 we ran more than 900 individual checks across restaurants, cafés, guesthouses and hotels in Germany and Austria. Every business got the same treatment: nine live local searches, a full pass over the website and the Google profile, and a set of questions put to Google's AI Mode and to ChatGPT while logged out. This is what came back, including the places where it contradicts the advice everyone else gives.

If you would rather start with the work, skip to what to fix first.

Does the better restaurant rank higher?

The businesses we checked hold a median Google rating of 4.75 stars, and 78% sit at 4.5 or better. Tested against nine local searches each, they are absent from 67% of them. One house with 4.9 stars shows up in every single search. One with 4.4 shows up in none.

Julian Rauch, who ran these audits, put it in one line before the counting started: "It is rarely the genuinely best restaurant that sits at the top."

The pattern holds outside our sample. Local Falcon analysed 10,000 US restaurants across nine grid points in May 2026 and found that businesses with 1,000 or more reviews were still absent from Google's AI recommendations 70.9 % of the time, with the 4.5 to 4.7 star band the most visible of all (Local Falcon, "The Restaurant AI Visibility Index", 2026, retrieved 2026-09-22). That study is US-only; ours is German and Austrian. Both land in the same place.

Reviews clear a filter. Clearing it changes nothing else on its own.

What does a website audit actually measure?

Each business got about fifty individual checks, more than 900 in total. Of the checks that returned a result, 42% failed. Every business also went through nine live local searches drawn from its own cuisine, category and catchment, and two thirds of them were put to Google's AI Mode and to ChatGPT while logged out.

Scores landed between 32 and 65 out of 100, with a median of 48. Not one business reached 70.

Three limits belong on the table before any of this gets quoted.

What follows is sorted by how often each check failed, against the businesses where that check actually ran.

Which mistakes show up most often?

Three readings change what you would work on first.

The menu is not the main problem. A clear majority already publish it as readable text. What is missing is the label: three quarters carry no structured data saying the business is a restaurant or that the page is a menu.

The crawler question is settled. More than eight in ten already let AI crawlers through. One business blocked everything except Googlebot, and it is the only one where that section of the usual advice applied at all.

The expensive failures are missing from the table entirely, because they sit behind the enquiry rather than on the website. Those checks only run where there is something to check, so the base is too small to put a percentage on. The pattern is not subtle: almost every business where we could test it had no system catching enquiries, no automatic confirmation, no review request running, and nothing posted to its Google profile inside a month.

Why does a hotel with a restaurant vanish from half the searches?

Two owners we audited, 700 kilometres apart, ran into the same constraint and solved it in opposite directions. Both lost half their demand doing it, and Google's own documentation already contains the answer neither of them found.

The operator of a hotel with a restaurant in northern Germany stayed in the hotel category. He described the consequence himself: guests searching for a restaurant in his town do not find him, "because I am not listed under restaurant, I am listed under hotel." When he tried to add the second profile, Google refused it, and he read that refusal as the rule.

An innkeeper in eastern Germany went the other way. She picked restaurant over hotel, and her reason was specific: under the hotel category she could not enter opening hours.

She was right about the constraint. Google documents it in one sentence: "Hotels do not have the ability to edit their business hours because hotels are open 24 hours to their guests" (Google Business Profile Help, "Manage your hotel's details", retrieved 2026-09-22). The general hours page carries no such carve-out, so the restriction lives only on the lodging documentation, which is why almost nobody finds it.

The part both of them missed sits in Google's representation guidelines, and it is explicit. Under co-located businesses Google writes: "The following types of co-located businesses should each have their own profile. If you need to use both categories for the same business location, create two profiles instead." The first example Google lists is "A Restaurant/Cafe/Bar inside of a Hotel/Motel" (Google Business Profile Help, "Guidelines for representing your business on Google", retrieved 2026-09-22).

A restaurant inside a hotel is one of exactly four named cases where Google's own rules point at two profiles. The conditions are real and worth reading before you try: each profile needs a distinct name from the main business, its own categories, and has to face the public as a separate entity. The guideline is permission to make the case, and approval still runs through verification.

Why this one is worth the hour it costs: in Whitespark's 2026 survey of 47 local search practitioners scoring 187 factors, the primary Business Profile category ranks as the single strongest local pack factor at 227 points, ahead of proximity at 225. "Business is open at time of search" sits fifth at 189 (Whitespark, "Whitespark's Official 2026 Local Search Ranking Factors Report", retrieved 2026-09-22). That is an expert panel rather than a measurement, and Google itself never states that the primary category outweighs the additional ones. Google does confirm the mechanism in general terms, that "the categories you select affect your local ranking on Google", and caps additional categories at nine (Google Business Profile Help, "Manage your business category" and "Create a bulk upload spreadsheet for Business Profiles", both retrieved 2026-09-22).

The category check failed at 53% of the businesses where it ran, which makes it the highest-leverage item on the whole table.

Can a machine tell that your menu is a menu?

A majority, 57%, already publish their menu as readable text. Three quarters carry no Restaurant or LocalBusiness structured data. The dishes and the prices are legible, and nothing on the page states what they are.

That ordering matters, because the widely repeated version of this advice points somewhere else. One of the top-ranking guides on restaurant local search calls PDF-only menus "the single most common mistake in restaurant local SEO". Measured against everything else on the list, it does not come close. The missing label does.

Two things are worth separating here, because the gap between them is where most structured-data advice goes wrong.

What markup does: it hands Google a machine-readable statement of what the page means. Google describes structured data as providing "explicit clues about the meaning of a page" (Google Search Central, "Introduction to structured data markup in Google Search", retrieved 2026-09-22). For a restaurant it can feed the knowledge panel and a business carousel.

What markup does not do: produce a menu rich result, because Google documents none. Its structured data gallery lists 25 supported types as of June 2026, including local business. Menu is not among them (Google Search Central, "Structured data markup that Google Search supports", retrieved 2026-09-22). Google's own policy page states the boundary plainly: "Using structured data enables a feature to be present, it does not guarantee that it will be present" (Google Search Central, "General structured data guidelines", retrieved 2026-09-22).

The text itself carries weight for a separate reason. Vercel's 2024 crawler analysis found that the major AI crawlers do not execute JavaScript (Vercel, "The rise of the AI crawler", retrieved 2026-09-22). A menu that only assembles in the browser is a menu they never see.

Ship the markup because it is cheap and unambiguous. The evidence for it improving AI citations is absent, and that belongs in the next section.

What happens after someone enquires?

This is the layer that does not appear in the table, because these checks only run where there is something to check. Wherever we could test it, almost every business took enquiries into no system at all, had posted nothing to its Google profile inside a month, ran no review request after a visit, and sent no automatic confirmation to the person who had just written in.

A hotel owner in the Austrian Alps described what fills that gap. He carries a diverted line and answers it himself, day and night, because "I have no artificial intelligence that can work my phone well enough to hold that conversation." He knows his own website should show a price and it does not: "that is so much work, I have not managed it yet."

Another operator had tried asking for reviews by hand and stopped. His count: "maybe one in twenty left one."

Why do these survive an agency?

Roughly nine in ten of the hospitality businesses we audited had already paid an agency for visibility at some point. That figure is our own observation across the sales conversations behind these audits rather than a counted sample, and it should be read as one. What is consistent is the shape of the complaint: changes take too long, and every change costs again.

One owner gave the full arc. Close to €1,000 a month for a year, then cancelled: "I never saw the result, so after a year I cancelled again."

The scale of the affected population is documented, at least in the US. Clutch surveyed 406 small business owners in August 2025 and found 83 % have a website, with 45 % of those sites built by an outside partner and 8 % by the owner (Clutch, "Clutch Report: No-Code Tools Fuel Website Growth, Yet 17% of Small Businesses are Still Offline", 2025, retrieved 2026-09-22).

How long those owners then wait for a change has no published benchmark at all. We looked in German and English. What circulates instead is a set of figures attributed to a "Clutch Agency Survey 2024" and a "Clutch SMB Digital Operations Survey 2024", neither of which appears in Clutch's own index of publications. The single source for both is a vendor selling website-update services.

That absence is the reason the next section is sorted by effort instead of by importance.

What should you fix first?

Sorted by who has to do the work, because that is the part that decides whether it happens.

This week, by yourself, at no cost

1. Check your Google category, and count your rooms. If you run a restaurant inside a hotel, read Google's co-located businesses rule and apply for the second profile. Distinct name, distinct categories, publicly facing as its own entity.
2. Fill your additional categories. You can hold nine beyond the primary one. Use the ones a guest would type.
3. Publish one Google Business Profile post. Then put a recurring 15 minutes in the calendar for the same day each month.
4. Read your own name, address and phone across every directory that lists you. Correct whichever is wrong, character by character, including the domain spelling. One operator we audited discovered he had been running two versions of his own domain for years: "at the start we just did not pay attention."
5. Put your menu on your own domain as text. Dish names and prices as characters on the page, with whatever tool you already publish with. Keep the designed version as a download.
6. Show one price. A range is enough. Guests who cannot find a price on your site find it on a portal instead.
7. Book a table at your own restaurant, on your own phone, with a stopwatch running. Start at a cold search for your own name and stop when the booking is confirmed. One operator who tried this for the first time needed over four minutes. Write the number down before you change anything, because it is the one measurement on this list that does not need a tool.

Once, with someone technical

1. Add Restaurant or LocalBusiness markup with address, opening hours and a menu URL. Ship both hasMenu and menu, because schema.org supersedes the older property while Google's documentation still names it.
2. Fix mobile load time if your own phone takes longer than two and a half seconds to show the page.
3. Make an enquiry completable without a phone call. A form with a date and a party size counts. Nearly three quarters of the businesses we checked had no such path, which is usually where the stopwatch in step 7 runs out.

Continuously, and this is where a system earns its place

1. Every enquiry gets captured with a status, not only an inbox.
2. Every enquiry gets an automatic confirmation, so the guest knows it arrived.
3. Every guest gets the same review request after the visit, through one link.
4. The nine searches get re-run monthly and written down.

Steps 1 to 10 need no product and no agency. Steps 11 to 14 are the ones that keep costing time forever, which is exactly why they are the ones that were missing almost everywhere.

What can you skip?

Four pieces of common advice that the evidence does not support.

An llms.txt file. Ahrefs examined 137,000 domains in May 2026 and found 97 % of llms.txt files received zero requests (Search Engine Journal, "97% Of llms.txt Files Got No Requests, Ahrefs Data Shows", 2026, retrieved 2026-09-22).

Schema markup as a route to AI citations. A matched test across 1,885 pages that added JSON-LD found AI Overview citations down 4.6 %, with the AI Mode and ChatGPT movements not statistically significant (Search Engine Journal, "Schema Markup Didn't Move AI Citations In Ahrefs Test", 2026, retrieved 2026-09-22). Google states the same thing from its own side: "There's also no special schema.org structured data that you need to add" to appear in AI Overviews or AI Mode (Google Search Central, "AI features and your website", 2025, retrieved 2026-09-22). Add the markup for the reason in the menu section above. Expect nothing from it in AI answers.

Collecting more reviews as a visibility lever. The Local Falcon figure at the top of this piece puts the ceiling on it: 1,000 reviews and still absent 70.9 % of the time.

Adding another channel. One operator we audited listed what a single menu change already costs him: publish it to social, upload it to the cloud folder, rebuild it on the review portal, push it as a Google post. "That is four or five channels I have to touch every time." A fifth channel makes the next change slower.

And a fifth, halfway. Your robots.txt is probably already fine. It was at more than eight in ten of the businesses where we checked. Open the file, look for a line disallowing OAI-SearchBot, and move on.

How do you check your own restaurant in ten minutes?

Paste this into ChatGPT, Claude or Gemini. Fill in the brackets and change nothing else.

Act as a local search analyst. My business is [name], a [restaurant / hotel /
guesthouse] in [town], serving [cuisine]. My website is [url].

1. Search the way a guest would: nine variations of [cuisine] plus [town],
   plus two neighbouring towns and two occasions (celebration, business lunch).
   For each one, list every business named, in order, and the source you used.
2. Tell me in how many of the nine I appeared. Name the businesses that
   appeared most often instead of me.
3. Open my website. As plain text only: can you read my menu with dish names
   and prices? My opening hours? A price range? Is there a way to make an
   enquiry without calling?
4. Look up my Google Business Profile. What is my primary category, and how
   many additional categories are set? If my business contains both a
   restaurant and rooms, tell me which of the two the profile is optimised for.
5. List every fact about me that contradicts another source you found.
6. Rank the fixes by how much each one would change your own answer.

Run it on the first Monday of each month and keep the answers in one document. The change between months is the only visibility measurement a single location needs.

How do German guests actually search?

A YouGov survey of 2,000 respondents in Germany in January 2025, commissioned by METRO and its subsidiary DISH Digital Solutions, found 70 % use digital platforms to find a restaurant, 51 % use search engines specifically, 50 % have booked a table online, and 38 % have used a digital menu (Tageskarte, "Umfrage - Mehrheit nutzt digitale Tools für Restaurantsuche, Reservierung und Bezahlung", 2025, retrieved 2026-09-22). The commissioning parties sell digital tools to restaurants, which is worth knowing while reading it.

Put that next to our own measurements without any inference step. Half of German diners start at a search engine. Two thirds of the local searches we ran against these businesses returned nothing.

What that gap is worth has published numbers on both surfaces, and they are not the same number. On German organic results, Sistrix measured the click-through rate on position one at 27% on a normal result page and 11% once an AI Overview sits above it, across more than 100 million keywords (SISTRIX, "AI Overviews in Deutschland: So stark sinken die Klickraten wirklich", 2026, retrieved 2026-09-22). In the map results the numbers run lower and flatter: 14.8% for the first slot, 12.6% for the second, 11.9% for the third, and 7.3% for a business named inside an AI answer (First Page Sage, "Google Click-Through Rates (CTRs) by Ranking Position in 2026", retrieved 2026-09-22). That second set comes from an agency's own client and panel data rather than a random sample, so read it as an order of magnitude.

The useful part is the shape rather than the exact figure. Those three map slots sit within three percentage points of each other. Appearing in the pack at all is the jump, and moving from third to first is a rounding error next to it. Most of this article is about the first jump.

For rooms the starting point has already moved. SiteMinder's 2026 traveller report, covering 12,000 travellers across 14 countries including Germany, records booking platforms overtaking search engines as the primary research starting point for the first time, at 26 % against 21 % (SiteMinder, "SiteMinder's Changing Traveller Report 2026", retrieved 2026-09-22).

That shift has a price attached. In Germany, 59.7 % of hotel bookings ran direct in 2025 and 31.9 % through online intermediaries, with 70.5 % of that intermediary share flowing through a single corporate group (HOTREC, "European Hotel Distribution Study 2026", 2,713 hotels across 28 countries, reported by AHGZ, 2026, retrieved 2026-09-22).

On the food side the published rates are specific. In 2024 Foodora put its maximum commission at 30 %, while Lieferando stated an average commission of 13 % per order plus a further 17 % when the platform handles the delivery entirely. A Vienna restaurateur quoted in the same report drew his own line: "we can live with ten percent, anything beyond eighteen percent is not financeable" (wien.ORF.at, "Wirte protestieren gegen Lieferdienste", 2024, retrieved 2026-09-22).

Two owners we spoke to had done this arithmetic themselves, arriving at 16.2 % including bank fees and 17 % respectively. Both figures land almost exactly on the platforms' own published logistics surcharge.

Where does a business system fit into this?

Steps 1 to 9 above are a weekend. Steps 10 to 13 are a system, and the audits show what happens without one: the follow-up layer is where nearly every business fell down.

Knowlix is an all-in-one AI Business Platform that replaces 50+ apps, and the part that applies here is where a fact lives. Enquiries, contacts, invoices and the pages built by the AI Website Builder sit on one shared set of records, so an opening time changes once and changes everywhere that record feeds. An AI Teammate drafts the work ahead of that, and every step is approval-gated, so a person signs off before a guest sees anything. The same drift shows up as disconnected business systems in every other part of a small operation.

One limit stated plainly, because this article turns on it: the website builder does not emit menu markup by itself. hasMenu, servesCuisine and the Restaurant type are hand-built work, and in our own paid service they sit in week two of a four-week plan.

No platform can promise you a position in a search result or an AI answer. What you can control is whether the facts a machine needs exist as text, agree with each other, and sit where the machine actually looks.

What would your own score look like?

Everything in this article came out of one process, and you can have it run against your own business. The Online Health Score checks your website, your Google profile and your listings, runs the nine local searches your guests actually type, puts the same questions to the AI assistants, and returns a score out of 100 with the findings behind it.

It takes a website address. Everything else we look up ourselves.

Two things worth knowing before you click. The score on its own moves very little from one business to the next, so the number is the door and the list underneath it is the answer. And a low score is ordinary. Not one business in this round reached 70.