We Audited 3,642 UK Hospitality Websites.
101 Passed.
Not 101%. 101 websites. Out of 2,789 we could measure, 101 showed their main content within the 2.5 seconds Google treats as acceptable. The median took 9.2 seconds.
How we measured it
Sample
3,642 independent UK restaurants, pubs and cafés across 50 ceremonial counties, sourced from OpenStreetMap. Chains are 0.3% of the set.
Checks
137 per site — speed, structured data, local SEO, metadata, email authentication, accessibility, AI readiness.
Speed
Cold TTFB (first uncached request — what a visitor and Googlebot pay) plus Google PageSpeed Insights for rendering.
Failures
611 sites could not be scored — dead domains, unreachable hosts, bot challenges. They are counted, not silently dropped.
1. Roughly one in twelve is already gone
Before measuring anything, 611 of 3,642 sites couldn’t be audited at all. The largest group — 313 businesses, 8.6% — have a domain that no longer resolves. Listed with a website, but there is nothing there. We'll be re-checking these dead domains quarterly — if any come back to life, we'll update the study.
Verified by DNS resolution outside the audit engine, so no scoring bug can inflate it. These businesses are usually still trading — the website simply lapsed and nobody noticed.
2. The page-render problem is far worse than the server problem
Two different measurements. Time to first byte is how quickly the server answers. Largest Contentful Paint is how long until the visitor can actually see the menu. The first is mediocre. The second is catastrophic.
⚡ Server response (cold TTFB)
Median 808ms · 36.5% under 600ms
🐢 Page render — Largest Contentful Paint (LCP)
✗ Only 3.6% meet Google’s 2.5s threshold · median 9.2s
56.0% take longer than eight seconds to show their main content. Google’s own research puts the probability of a mobile visitor leaving at over 100% higher once a page passes three seconds. Most of these sites lose the customer before the menu appears.
3. A quarter of them need a CDN, not new hosting
This is the finding we couldn’t see until we measured cold and warm separately. A cold request pays DNS, TCP and TLS setup. A warm one reuses the connection. The gap between them says where the time goes — and it splits the population into groups needing completely different fixes.
24.6% of these businesses have a perfectly healthy server sitting behind a slow first connection. Their hosting is not the problem and upgrading it would change nothing. They need a CDN — an afternoon’s work, often free.
Standard SEO tooling reports a single speed number and recommends all three fixes at once — better hosting, caching, a CDN. For a quarter of this market, two of those three are wasted money.
Which of the four is your site? The audit reports cold and warm separately, so you can see whether it's the server or the connection.
Run a free audit →4. The invisible restaurant
Ranked by percentage of available points — not raw score, since categories carry different weights — the weakest areas are not the ones owners worry about.
Avg score: 2.6 / 4 pts
Avg score: 7.9 / 12 pts
Avg score: 2.1 / 3 pts
Avg score: 3.6 / 5 pts
Avg score: 3.9 / 5 pts
Avg score: 9.4 / 12 pts
Avg score: 9.0 / 10 pts
Avg score: 9.1 / 10 pts
Only 15.1% of these venues publish LocalBusiness structured data — the markup that tells Google an address, opening hours and venue type. These are businesses whose customers search “restaurant near me” with immediate intent, and roughly six in seven have omitted the one piece of code written for exactly that search.
Performance shows the largest absolute loss (4.1 points of the roughly 22 the average site gives up), but AI readiness is proportionally worst at 64.1% of available.
5. The twelve most-failed checks
Pass rate across all 2,850 scored sites.
Some of these are severe (Largest Contentful Paint, DKIM). Others are cosmetic (Dark Mode). We publish the full list rather than the flattering subset.
6. The maintenance marker
llms.txt is a small file
telling AI assistants what a site is about. 20.4% of these sites have one — and they score
markedly higher.
This is association, not causation. Adding the file will not raise your score by 5.57 points. It is a proxy: its presence means somebody is still tending the site, updating the CMS, running a current SEO plugin.
The strongest evidence for that reading is what happens when you look at file quality. Among sites that have one, a genuinely useful file scores no better than boilerplate — the difference is not statistically significant. Presence predicts a 5.57-point gap; contents predict nothing.
Adjusted for the fact that llms.txt is itself one of the 137 scored checks, the conservative gap is +3.71 points (t=22.5). We report both.
110 businesses have no HTTPS at all
110 sites (3.9%) have no working certificate. Every visitor sees a browser “Not Secure” warning, and Google has treated HTTPS as a ranking signal since 2014.
The counter-intuitive part: they are the fastest sites in the study — median 334ms against 817ms for everyone else. They are simple, static, hand-built pages from an earlier era. Nothing is wrong with them technically except the missing certificate.
These sites were invisible in our first pass — the engine only tried HTTPS, so a site without a certificate was recorded as unreachable and written off. Adding an HTTP fallback surfaced them.
Where everyone lands
Mean 78.2 · median 79 · 54.2% score C or below
Five things worth doing this week
Put a CDN in front of it
If your server answers quickly but the page still feels slow, you are in the 24.6% that a CDN fixes. Cloudflare's free tier is enough.
Compress your images
The gap between a 808ms server response and a 9.2s render is almost entirely images and scripts. This is the single biggest win available.
Add LocalBusiness schema
Fifteen percent of your competitors have it. It is roughly twenty lines of JSON and it is what "near me" searches read.
Check your domain still resolves
313 businesses in this study are paying for a website nobody can reach. It takes ten seconds to check.
Fix your email authentication
Only 6% pass DKIM. Without it your booking confirmations are trivially spoofable and increasingly land in spam.
What we’re measuring next
Hospitality was the first vertical because it is large, independent and local-search dependent. The same 137-check audit is now being run across adjacent categories, and the interesting question is whether these patterns are a hospitality problem or a small-business problem.
🛏️ Stay
Hotels, B&Bs, guesthouses, self-catering. Higher transaction values and heavier booking integrations — we expect better speed and worse structured data.
📅 Bookings
Venues, event spaces, activity operators. Almost entirely dependent on being found for a dated, local query.
✂️ Services
Hairdressers, salons, trades. In our sampling only about 5% even have a website — the interesting number may be absence, not quality.
🛍️ Retail
Independent shops. The one category where a slow page has a directly measurable cost per visit.
We already hold contact and website data for roughly 2,000 non-hospitality businesses gathered alongside this study. Those audits are queued.
What this study cannot tell you
- • Speed does not predict our score. We tested it. Once you remove the components speed itself feeds, the correlation is r = +0.05 — not significant. Slow sites are not automatically bad sites.
- • No traffic or revenue data. We measure what a site does technically, not what it earns. Nobody should infer lost covers from a score.
- • Homepages only. A missing FAQ means missing from the homepage, not absent from the site.
- • One snapshot. August 2026. These figures are frozen deliberately so they stay reproducible.
Which group is your site in?
Run the same 137 checks on your own site, or see where you sit against the cohort without revealing your domain publicly.
Methodology notes
- • 3,642 independent UK hospitality sites audited August 2026 across 50 ceremonial counties. 2,850 returned scoreable results; 611 are reported as failures rather than dropped.
- • Speed: cold TTFB (first uncached request). Warm figures reported separately and never conflated.
- • PageSpeed: Google Lighthouse mobile strategy via API (2,789 of 2,850 returned data).
- • llms.txt: independently verified by re-fetching every domain with soft-404 detection. Raw gap +5.57 (t=30); conservative, adjusted for circular scoring, +3.71 (t=22.5).
- • Category ranking normalised to percentage of available points — raw sub-scores are not comparable across categories of different weight.
- • Chains and multi-site brands are 0.3% of the sample and were not excluded.
- • Speed is measured cold, on a fresh connection. Averaging repeat samples measures a warm connection and reports figures several times faster than a real first visit.