The UK Restaurant AI Visibility Index (2026): Who AI Actually Names

In our review of public restaurant AI studies before this index first appeared on 26 July 2026, the published market evidence we found was US-focused. We wanted a bounded London read of which restaurant brands answer systems actually named and which sources they retrieved. Over 9 consecutive days, from 12 to 20 July 2026, we tracked 28 London restaurant and wine-bar brands across around 85 real diner questions on ChatGPT and Google AI Overviews. We retained 2,138 usable answers after excluding empty responses, errors and non-answers. Every brand was scored against that same 2,138-answer denominator. The short version: even the most AI-visible brand appeared in only about one in eight answers. This is a young London aggregate, not a UK census or a restaurant-quality rating.

TL;DR

• Across 2,138 retained AI answers to London dining questions, the most-named brand appeared in 12.5%. Every brand used the same denominator; no tracked brand cleared one in seven.
• The median across all 28 brands was 4.19%. All 18 brands outside the top ten were below 5.6%; Padella was highest among them at 5.10%.
• Hawksmoor led the index, followed by Blacklock, Dishoom, Noble Rot and Humble Grape.
• Source retrieval was concentrated: Time Out was retrieved in 42% of answers, ahead of SquareMeal, thatsup, Reddit and The Infatuation. Editorial, community and listing sites occupied much of the retrieved evidence set.

What we measured, and how

We tracked a purposive cohort of 28 London restaurant and wine-bar brands, from steak and Indian fine dining to natural wine bars, across around 85 diner questions covering neighbourhood, occasion, group size, price, cuisine, dietary needs, access, private dining and comparison. This was not a random or representative market sample. Brands were admitted when they had an official London restaurant or wine-bar presence, then selected to cover multi-site and independent operators across different cuisines, price points, neighbourhoods, occasions and operating models. The public measurement receipt names all 28.

Those questions ran daily on ChatGPT and Google AI Overviews for 9 consecutive days, from 12 to 20 July 2026. A response was retained when the engine returned a usable answer to the dining question; empty responses, errors and non-answers were excluded. That left 2,138 responses. For every brand, visibility means mentions divided by the same 2,138 retained responses, with every response carrying equal weight. We also recorded which source domains each answer retrieved.

The receipt fixes those definitions, the engine set, tracked universe, full 28-brand distribution, retrieval table and exact run dates. The underlying prompt library remains in the Peec project. The public receipt does not republish raw answer text.

Who does AI actually name? The London leaderboard

Visibility here means the share of the same retained London dining answers that mention a brand. Read the top number twice: the most-named brand appears in only 12.5% of this tracked prompt portfolio.

HHawksmoor

BBlacklock

DDishoom

NNoble Rot

HHumble Grape

KKiln

SSt. JOHN

GGymkhana

BBig Mamma

FFallow

The ceiling is the finding. First place still means being absent from roughly seven of every eight retained answers in this London prompt portfolio.

The order describes this prompt portfolio and measurement window. It does not prove that one brand has a better restaurant, marketing team or commercial result. Hawksmoor tops the index with the largest measured answer share of any tracked brand, but the stronger conclusion is the ceiling: no brand was close to appearing consistently.

What the full distribution adds

The median across all 28 brands was 4.19%. All 18 brands outside the published top ten were below 5.6%; Padella was highest among them at 5.10%, followed by Flat Iron at 4.72% and BAO at 4.68%. The complete distribution, mention counts and shared denominator are in the receipt. That makes the long tail inspectable without pretending the cohort represents every restaurant in London. The mechanics behind restaurant invisibility are covered separately in why your restaurant is invisible on ChatGPT, and the wider sourced numbers in the restaurant AI visibility statistics.

Two external benchmarks help frame the scale without being blended into this index. Local Falcon's May 2026 US audit used a different local-grid method across 10,000 restaurants and found 74.9% absent from Google AI Overviews. SOCI's vendor-published Local Visibility Index covered more than 350,000 locations and found ChatGPT recommended 1.2% of brand locations against 35.9% appearing in Google's local three-pack. Both are US studies and both come from companies selling into local visibility. They are directional comparisons, not validation of this London cohort.

Which sources did the engines retrieve for London dining?

The source aggregate is the most useful operating queue in this index. When an engine answered these London dining questions, it repeatedly retrieved a small set of domains. Here is how often each appeared across the same 2,138 retained answers.

Retrieved sources form the working evidence set. Retrieval is not the same as an inline citation, and neither measure proves that one page caused a restaurant mention.

Three things stand out. First, editorial sources occupied much of the retrieved set: Time Out appeared in more than four in ten retained answers, with several UK titles behind it. Second, community was material: Reddit and thatsup appeared frequently enough that a restaurant source plan cannot treat community as a footnote. Third, a restaurant's own website was rarely the displayed source in this aggregate. That does not make the venue page unimportant. The first-party page resolves facts; outside sources provide independent corroboration.

What this index can and cannot tell you

Use the index as a directional baseline, not a league table of restaurant quality. The public measurement receipt fixes the window, sample size, definition and top-ten aggregate so the number can be checked against the same contract later.

If you run a restaurant or group and need to turn this market view into a check of your own venues, use the Restaurant AI Visibility Audit. It separates the public research question from the operational one: what an engine can verify about your restaurant today, which claim or source is weakest, and whether a larger programme is justified. Start with the free diagnostic when you want a quick evidence-led first pass.

The index can answer four bounded questions:

• Which tracked brands appeared most often in this retained London prompt portfolio?
• How large was the gap across the published top ten?
• Which domains were retrieved most often alongside these answers?
• Which source and venue-record gaps deserve a closer, prompt-level diagnosis?

It cannot show incremental bookings, prove that a specific source caused a mention, or tell a restaurant that it should fit every question. Prompt choice matters. A steak group should not win a vegan tasting-menu question, and a neighbourhood wine bar should not be judged against every city-wide fine-dining prompt. Engines also vary by location, account, time and model state.

That boundary changes the operating response. Do not chase the leaderboard as one blended score. Split the prompt set by venue fit, inspect the exact answer and sources, then repair the weakest evidence layer.

What the index means if you run a London restaurant

Two practical reads follow. If a relevant venue is absent from a prompt, diagnose whether the gap is first-party clarity, outside corroboration, current reviews or genuine fit before creating anything new. If you run a group, keep every venue independently resolvable so the engine does not blend locations, which we cover in why your restaurant group's venues cannibalise each other.

The source table is a research queue, not a mass-pitch list. Open the exact pages retrieved for the prompts that matter. Work out why the page was useful. A menu launch, chef appointment, local data point or access improvement may support a real editorial idea. A generic request to be included does not.

Start with the free Restaurant AI Visibility Diagnostic to inspect one live venue page. For a prompt-level baseline and the source graph behind it, see Schmitdy's restaurant AI search system.

Sources

1. Schmitdy, public UK Restaurant AI Visibility Index measurement receipt: retained methodology, definition, top ten and limitations.
2. Hawksmoor, official website: official group identity and current venue record.
3. Blacklock, official website: official group identity and current venue record.
4. Dishoom, official website: official group identity and current venue record.
5. Noble Rot, official website: official group identity and current venue record.
6. Humble Grape, official website: official group identity and current venue record.
7. Kiln, official website: official venue identity and current record.
8. St. JOHN, official website: official group identity and current venue record.
9. Gymkhana, official website: official venue identity and current record.
10. Big Mamma, official website: official group identity and current London venue record.
11. Fallow, official website: official group identity and current venue record.
12. Time Out London, Restaurants: live editorial restaurant section from the most frequently retrieved domain in this aggregate.
13. SquareMeal, London restaurants: live London restaurant discovery section.
14. Visit London, Food and Drink: official destination source for London food and drink discovery.
15. Michelin Guide, London restaurants: live London guide inventory.
16. OpenTable, London restaurants: live London booking and venue discovery inventory.
17. Local Falcon, Restaurant AI Visibility Index: separate US local-grid benchmark across 10,000 restaurants; method and market differ from this London prompt study.
18. SOCI, AI for Local SEO: How Agents Improve Rankings for Multi-Location Brands: vendor-published summary of its 2026 Local Visibility Index across more than 350,000 locations; directional rather than independent validation.