Open research
Almost half of the 200 dental websites we checked never tell search engines they are a local practice
We checked the websites of 200 dental practices drawn from across Great Britain, between 30 October and 5 November 2026. 95 of them carry no machine-readable description of the practice — no address, opening hours or phone number that a search engine or an AI assistant can read without guessing. Nobody was surveyed and nothing was judged by eye — only what can be measured in an ordinary browser.
- 95sites give search engines no local business detailsof the 200 checked — 47.5%
- 116sites offer no messenger contactof the 200 checked — 58.0%
- 100sites take over 4 seconds to show their main contentof the 200 checked — 50.0%; of the 137 whose speed could be measured — 73.0%
In our Ukrainian study the same figure is 46.5%. Two countries, two separately drawn samples, and the share lands within one point — so whatever keeps this description off dental websites, it is not peculiar to one market.
What we measured with
Every number here was gathered by
our own auditor
AI X-RAY
It is the same check we run on every site before it goes to a client: over a hundred points across six areas — speed, visibility in search, security, accessibility, code quality and the path to getting in touch. Speed is measured for us by Google PageSpeed Insights; everything else is checked by our own analyser.
How we did it
200 sites out of the 1,183 drawn from across Great Britain
The query to the open OpenStreetMap database covers all of Great Britain, not selected cities. It lists 6,617 dental practices, and 1,184 of them give a website address — about one in six, against one in ten in Ukraine. Every candidate came from the map; no directory was added. After removing one domain from outside the market, 1,183 were left, and we checked 200 of them. In Ukraine the checked sites were most of the list; here 200 is a sample of a larger, better-filled list. That cuts both ways, and the limits below say how.
- 6,617Dental practices on the open OpenStreetMap database, all of Great Britain
- 1,184Of those, give a website address — about one in six
- 1,183Left after removing one domain from outside the market
- 200Checked
- 138Opened and allowed the full check
What was intended and what came out
| Type of place | Intended | Actual |
|---|---|---|
| London and the largest metro areas | 30% | 30% |
| Large cities | 40% | 40% |
| Smaller towns | 30% | 30% |
The quota was met exactly: 60, 80 and 60 sites. Unlike our Ukrainian study, there were enough eligible sites in smaller towns, so nothing had to be filled from the large cities. London alone accounts for 60 of the 200.
Every percentage on this page is counted over all 200 checked sites unless stated otherwise beside it. Where a figure can only come from the sites that answered, we say so — "100 of 137", for instance. A site we could not reach is counted neither as having the problem nor as free of it: nothing is known about it.
Speed
Half the sites keep a visitor waiting over four seconds
We measured the time until the main block of the page appears — the large heading or the photo at the top. That is the LCP metric, and the 2.5-second threshold is Google’s. The measurement itself is made on our request by Google PageSpeed Insights: mobile profile, five runs, median taken. These are lab figures, not the real-visitor data Google collects separately in its own reports — the two are usually close but not identical.
How long the main content takes · all 200 sites
Whether the page jumps while loading · all 200 sites
Among the 137 sites we could measure, 117 stay still, and the median shift is effectively zero. The trouble is not here — it is in how long the first screen takes.
- 7,695milliseconds to the main content — medianthat is 7.7 seconds — about three times Google’s limit
- 400milliseconds for the server to answer — medianunder half a second — the server is not where the time goes
- 68points out of 100 for speed — medianGoogle’s score for mobile
The same picture can be counted another way. Over all 200 sites, 100 take over 4 seconds — 50.0%. Over only the 137 whose speed could be measured, it is 100 of 137 — 73.0%. One site opened and was checked in full, but PageSpeed Insights returned no speed result for it, which is why 137 is one fewer than the 138 sites that allowed the full check. Both framings are correct; they answer different questions.
The server answers in 0.4 seconds, yet the main content takes nearly eight. So the time goes not into the server’s reply but into what happens after it — heavy images, fonts and scripts. What exactly is to blame on any one site only a separate diagnosis can show.
Technical risk
Availability is not the problem here: 13 of 200 did not open
Served no working home page
In 8 cases the domain has no DNS record, 4 servers returned a 404, and 1 answered with a registrar’s placeholder instead of a site. None are blocks on our auditor — those are counted separately
136.5%95% CI: 3.8–10.8%
Server files left open
Files that are normally closed off are reachable over the web. We do not publish which files or where
52.5%95% CI: 1.1–5.7%
Domain expires within two months
Unless it is renewed, the site disappears and joins the very first row of this list
136.5%95% CI: 3.8–10.8%
Running without a secure connection
The browser marks the address as not secure, and anything typed into a form travels in the clear
31.5%95% CI: 0.5–4.3%
Certificate expiring with no auto-renewal
Fewer than 60 days left and no automatic renewal configured
31.5%95% CI: 0.5–4.3%
Signs of a break-in
Injected text, redirects to other sites, spam
0of 20095% CI: 0.0–1.9%
Why 62 sites could not be fully checked
If a site took us for a robot, asked us not to visit, or redirects somewhere else, that is our inability to check it rather than a flaw in the site, and it is left out of the counts. Of the 15 under Other, 11 served an empty shell — a real page head over a body with no text, which is what a site behind an interstitial or an unrendered single-page app serves — and 4 were connection failures the sites themselves produced: three timeouts and a reset. They fell on both 30 October and 5 November rather than together, so they were not one outage on our side. The 20 robot blocks are a limitation in their own right; see below.
Visibility in search
Almost half the practices never tell search engines they are a practice
Besides the text for people, a page can carry a short description for machines: what kind of business this is, where it is, when it is open, what number to call. Google and AI assistants can read it, and it helps them understand the page correctly. It is not a condition of being in search — a site is indexed without it — but with it there is less room for the business to be described wrongly. A generic Organization block, which SEO plugins add to every site, is not counted: it carries no address and no opening hours.
No business description for search
Address, opening hours and phone are not given in machine-readable form, so search has to pick them out of the text
9547.5%95% CI: 40.7–54.4%
No machine-readable description at all
To a search engine the page is plain text with no hints about what is on it
4321.5%95% CI: 16.4–27.7%
No page description
Google writes the caption in results itself, taking a passage from the page
2311.5%95% CI: 7.8–16.7%
No list of pages for search engines
Search finds pages through links as well; a list only helps it happen sooner
147%95% CI: 4.2–11.4%
Closed off from search
An explicit rule forbids showing it in results
10.5%95% CI: 0.1–2.8%
No page title
The line shown in search results and on the browser tab. None among the sites that served a page
0of 20095% CI: 0.0–1.9%
Getting in touch
More than half offer no way to message the practice
How to get in touch · all 200 sites
How it looks on a phone · all 200 sites
Seven sites have no viewport setting for phones. We count the setting only when it carries width=device-width or initial-scale=1, not merely a viewport tag — and that checks the setting, not the whole experience on a phone. What is left is smaller: tiny text and images without descriptions.
A practice that cannot be messaged, an enquiry with no form, a number that will not dial from a phone: none of it needs a new website. And this is exactly where a visit that advertising already paid for can go missing.
How precise these figures are
A different draw would give slightly different figures
Any sample of 200 gives an estimate, not an exact value. That is why each share in the risk and visibility sections carries a 95% confidence interval: the range that, with that confidence, holds the true share among all 1,183 sites on the list. "No business description for search", for instance, is 47.5%, with an interval of 40.7 to 54.4%. The data file gives the interval for every figure. The same reading applies to the comparison with Ukraine: 46.5% there and 47.5% here sit well inside each other’s intervals, so the honest statement is that the two are about the same, not that one is higher.
What we do not know
What these numbers do not tell you
- Only those the open map knows are visible. About one dental practice in six (1,184 of 6,617) has a website address recorded in OpenStreetMap, which is not an official business register. A practice absent from it, or with the field empty, never entered the check — so this is a slice of the practices that marked themselves in open data, not of the whole industry.
- A sample of the list, not most of it. We checked 200 of the 1,183 eligible sites. In Ukraine the checked sites were most of the list; here they are a sample of a larger one, so every figure is an estimate and carries the interval shown beside it.
- One site in ten refused our crawler. 20 of the 200 sites blocked our auditor — 12 behind an unnamed filter, 8 behind Cloudflare — four times the 5 in our Ukrainian study. Part of the cause is us. The sample was crawled more than once while the auditor was being corrected, and some hosts escalated against our address: eleven sites hosted by SiteGround began answering with an HTTP 202 "Robot Challenge Screen" that never clears. A block here is partly a fact about our measurement, not only about the site.
- Three sites could not be confirmed by hand. Three sites stayed behind a self-reloading "One moment, please…" page when we checked them by hand from a clean browser, and the Internet Archive’s own snapshot of one of them shows the same wall. Their figures come from reads that did get through, and Google PageSpeed Insights — which fetches from Google’s own infrastructure, not ours — returned real scores for them. That is the corroboration we have; we could not confirm them a second time by hand.
- This is one particular week — 30 October to 5 November 2026 (56 sites measured on 30 October, 144 on 5 November), with the list fixed on 20 October. Some of the 13 sites that did not open may have come back the next day, and some live ones may have gone down.
- Dental practices only — and not their dentistry. These percentages cannot be carried over to other sectors or countries, and nothing here measures whether a practice is good at dentistry.
- We looked at the home page only. Service and practitioner pages were not checked; they may be better or worse.
- Lab figures, not real visitors. Speed is Google’s lab measurement in a mobile profile, not data from real visitors. In your city and on your connection the seconds will differ; the overall picture holds.
- An automated check is not a full audit. The auditor sees what is visible from a browser. It does not replace a manual review, and it does not prove the absence of other problems.
- We did not count anyone’s money. We have no data on these practices’ calls, bookings or revenue, so we do not claim a slow site cost anyone a particular sum.
The data behind this study
Every aggregate figure on this page in a single file, so it can be checked or recounted your own way.
Found a mistake?
If you think one of these numbers is wrong, write to us. We will check it, and if we got it wrong we will correct it on this page and note the date.
Have your own site checked the same way
The same AI X-Ray check, but on your site: what we found, how much it matters, and where to start.
30 October – 5 November 2026 · AI X-Ray · research-xray-lab 1.2 · run RGFEAzwt2kt5 · seed 4410. These figures may be reproduced with a link to this page.