The dental practice websites that rank in Yorkshire, measured

The short version. On the evening of 22 September 2026 I took the first two pages of Google’s organic results for “dentist in Leeds”, “dentist in Bradford” and six other Yorkshire towns, kept the pages that belong to a dental provider, and ran each through the analysis engine behind this site’s Website Visibility Checker. 118 websites were measured. The median Website Score is 86 and 61% score 80 or more, which says the technical basics are in hand: every site was served over HTTPS and 78% scored a perfect 100 on the technical component. The gaps are in what a machine can read about the business. 22% of the sites carry no structured data at all, a further 14% carry some but no business entity, and only 35% declare themselves a Dentist in the way Google documents. 19% have no meta description; 14% show a phone number that cannot be tapped. And the position a site holds in Google’s list has no measurable relationship with any of this: the rank correlation between organic position and Website Score across the 118 sites is 0.008.

Why this sample

Professional judgement Most “state of the industry” studies sample the businesses that exist. This one samples the businesses that already rank, because that is the set a patient actually meets, and because it lets one question be asked cleanly: among websites that Google has already chosen to show, does the quality a machine can measure vary, and does it line up with position? A sample of what ranks cannot say anything about what does not rank; that limitation is stated again below.

Method

1. 01
   Search

   Eight queries of the form “dentist in [town]” for Leeds, Bradford, Harrogate, Huddersfield, Wakefield, York, Halifax, Sheffield, on google.co.uk in English with the country set to the UK, from a signed-out desktop browser in West Yorkshire on 22 September 2026. Organic results only, pages 1 and 2 (positions 1 to 20). The map pack, ads and AI features were not recorded.

2. 02
   Inclusion

   Across the sixteen results pages, 155 entries were recorded, one per domain per search. 28 were removed because they are not a dental provider’s own site: NHS pages (present in every town), Healthwatch, council directories and listing sites such as WhatClinic. That left 127 pages, of which 118 could be fetched. A practice website that ranked in two towns counted once; a chain’s separate practice pages counted separately and are flagged as chain pages in the data.

3. 03
   Measurement

   Each page was submitted to this site’s own /api/analyse endpoint in site mode, engine version 1.1.0: the initial HTML is fetched once, as Google’s first pass and most AI crawlers read it, and the contact and about pages linked from it are fetched too where they can be found. No JavaScript is run. Thirty-plus checks produce seven component scores and an overall Website Score out of 100, under the published methodology.

4. 04
   Not fetched

   9 of 127 pages could not be measured: 8 pages on one national chain’s domain returned nothing to our fetcher at all, and one site redirected in a loop. One further site answered with a bot-challenge page (HTTP 403) and is included with the score that page earned, which is 12. Those are findings in themselves: a plain crawler is what an AI system or a new search engine looks like.

5. 05
   Provenance

   The searches were run and the pages submitted from the author’s own browser and account; the collection, the statistics and the first draft of this note were produced with AI tooling under his direction, and every figure on the page is recomputed from the published dataset each time the site is built. The author is accountable for every line.

6. 06
   Publication

   One row per website, carrying the town, the search position, whether it is a chain page, every score and every flag. No domain names, because no practice agreed to be named. The method is reproducible by anyone with the checker and a browser.

Results

The distribution

Measured observation These are good websites by the standard the checker applies, and that is worth saying first. Only 2 of 118 fall into the weak band, and one of those is the bot-challenge page. Professional judgement A patient who lands on almost any of these sites will find a practice that looks like a practice. The trouble is not what a person sees.

Where the marks are lost

Structured data: the biggest gap, and the most fixable

Measured observation 36% of the sites that rank give a machine no structured statement of what the business is. The pattern behind the second row is familiar from audit work: an SEO plugin emits a graph describing the web page and the website, and nobody adds the node describing the dentist. Google guidance Google’s documentation for local businesses asks for a business type with a name, address and telephone, and lists the dental type among the ones it recognises. Measured observation 17 sites carry rating markup on their own homepage, which Google’s review-snippet documentation says is not eligible when the rating is self-serving; it does no good and, in the audit reports we write, it goes on the fix list. Professional judgement Structured data does not move rankings on its own, and this note does not claim it does. It is the cheapest way to remove ambiguity for the systems that increasingly answer on a patient’s behalf, and 64% of the practices that rank have not done it fully.

Search presentation and conversion

Measured observation 17 sites, 14%, print a phone number a phone cannot dial. On a device where most local searches now happen, that is a customer asked to memorise a number and retype it. Professional judgement It is a five-minute fix and it appears in the top three of almost every report the checker produces, which is why it heads the worked example on the checker’s own page. Measured observation Booking links are nearly universal (86%); WhatsApp is not (14%). The median landing page carries 892 words of visible text and 142.5 KB of initial HTML; 18 pages are over half a megabyte before a single image loads.

AI readiness

Measured observation The AI-readiness component averages 75.9, and 20 sites score under 55 on it. Its weakest check is machine-readable identity, which 43 sites (36%) fail for the same reason as the structured-data table above; heading structure is the next, failed by 42. Only 6 sites answer questions in a form a machine can lift. Professional judgement Nothing here measures whether any assistant actually names these practices; that is what the live test is for, and it was not run on this sample. What this component measures is whether the raw material is there to be read, and for a third of the sites it largely is not.

Chains and independents

Measured observation The overall scores are close. The structured-data column is not: the chain pages average 57.5 against 72.8 for independents, and several of the chain pages in the sample, the location pages of one national group in particular, carry no structured data at all. Professional judgement A corporate template that omits a business entity omits it for every practice on it, which is the one place in this data where scale works against the patient’s ability to be found. The chain sample is small (20 pages) and one group accounts for several of them, so read the column as a pattern to check rather than a verdict.

By town

Measured observation Leeds has the highest mean (88.1) and Bradford the lowest (74.6). Professional judgement With twelve to eighteen sites per town the differences between neighbouring towns are within what a couple of poor sites can cause, and no town-level conclusion should be drawn from them beyond this: the gaps described above are present in every town, not concentrated in one.

Position and quality

Measured observation Page-one sites do not score higher than page-two sites, and the Spearman rank correlation between a site’s organic position (1 to 20) and its Website Score is 0.008, which is no relationship at all. Professional judgement Two readings are honest. The first is that Google is not ranking these pages on the things the checker measures, which is what Google itself says: the checker measures legibility, trust and conversion readiness, not authority or relevance to a query. The second is that within a set Google has already filtered, what separates position 3 from position 14 is not visible in the HTML. Either way, the practical point for a practice is that a good position does not mean the site is doing its job once the click arrives, and this data shows a great many well-ranked sites that are not.

Limitations

• The sample is the businesses that rank in organic search for one phrasing, on one day, from one location. It says nothing about practices that do not rank, about the map pack, or about “near me” and service phrasings.
• Organic results vary by location, history and time. A signed-out desktop browser in West Yorkshire was used; a search from another town or on a phone would produce a different top 20.
• The instrument reads initial HTML only. A site that builds its content with JavaScript reads poorly here and would also read poorly to most AI crawlers, but a person on a browser would see a complete page. The checker’s methodology page lists every check and its weight.
• One national chain’s pages could not be fetched at all (8 of the 127 sampled pages). They are absent from every figure, and the chain-versus-independent comparison is therefore about the chains that could be read.
• 118 sites is enough to describe a pattern and not enough to compare towns. The town table is published for completeness, not for ranking towns.
• Structured data was read from the landing page that ranked, which for some chain pages is a location page rather than a homepage. A practice may carry its entity markup elsewhere on the site.
• The Website Score is HighRegard’s own instrument, version 1.1.0, and its weights are a judgement. The raw flags in the dataset (tel link, JSON-LD blocks, business entity, description length and so on) do not depend on those weights.

What follows from it

Professional judgement For a practice, the fixes this data points at are the cheap ones: a business entity in the markup with the name, address, phone and hours; a tel: link around the number; a meta description; one H1. None will move a ranking by itself and none is a substitute for reviews, citations and a complete Business Profile. But they are the difference between a site that a search engine, an assistant or a patient can read cleanly and one it has to guess at, and on this evidence a third of the practices that already rank in Yorkshire are being guessed at. Working hypothesis The question this note cannot answer is whether the legible third are the ones assistants name. That is the first planned study on the research page, and this dataset is its starting panel.

HighRegard research The full dataset is published as a CSV under a Creative Commons Attribution licence. If you run a practice in the sample and would like your own row identified and explained, get in touch; the row is yours.