Where the facts come from, and how we show what we don't know.

Counted 22 Oct 2026. Out of 13,370 UK practices.

Areas counted 22 Oct 2026. Cut-offs are practices per 10,000 residents, calculated 22 Oct 2026. Figures a country never holds (health insurance, poverty and shortage areas are US census and HRSA measures; general health and deprivation are UK ones) are not listed for it, and the product says so on every page.

We hold a Google rating for 12,051 of 13,370 UK practices, mostly from our sister directory’s copy of the Google listing; we checked Google directly for West Yorkshire only. The date on a rating is when the directory last updated, which is the latest it could have been read. Where we have no listing the page says ‘not checked yet’. We don’t hold review text, so we don’t track what reviews say or how ratings move.

Counted 22 Oct 2026.

A Facebook advertising tag or a Google Ads tag on a website tells us the tag is there, nothing more. We do not know whether the practice is running adverts or what it spends. The product labels these as tags.

Separately from the tags, we look each practice up on the two advertising registers the platforms publish themselves — Google’s Ads Transparency Centre and Meta’s Ad Library. Where an advert is running we can show you which platform it is on, the platform’s own start and end dates, how many adverts, whether the words mention implants or offer finance, and one real sentence out of the advert itself. We never paraphrase an advert.

Three things this does not do, all of which the product says out loud rather than hiding. We do not know what anyone spends: neither register publishes it. Where an advert was found through a website or a Facebook page a practice shares with others, the advert belongs to the group and the row says so instead of crediting it to that practice. And where we could not look a practice up — Meta refused most of our lookups in the first national sweep — we print ‘we could not look’ and never ‘without an identified ad in the checked source’.

The website quality score is 0 to 100, from nine things we read on the page: what the site says about implants (16), whether prices are published (14), online booking (12), whether it is set up to measure its marketing (12), a scanner, 3D X-ray or lab equipment named (10), ways to pay (10), contact details on the page (10), the team named (10) and an implant brand named (6). Two of the nine only apply to a site that shows implant work. Every part prints with the score on the practice’s profile, with the date of the read it came from, so you can disagree with any of it by opening their website in a tab.

The weights are not an opinion about good websites. They are how far each sign separates one practice from another: a sign nearly every practice has cannot rank anything, and nor can one almost nobody has. And a website we have never read, or that did not load, or that answered with nothing we could grade, has no score at all — never 0. A 0 out of 100 reads as a verdict on the practice, and in all three of those cases what happened is that we did not read their site. The sentence saying which of the three it was is printed in the score’s place.

In the United States a metro or micro area is the Census Bureau’s own definition: whole counties grouped by commuting to a shared urban centre. In the United Kingdom the closest real geography is the travel-to-work area, which the Office for National Statistics draws so that most residents both live and work inside it. Both are built from where people travel, which is why they sit together in the product — and they are not the same construction, so we never call a travel-to-work area a metro.

Neither sits on the ordinary ladder of sizes: 59 US metros cross a state line, and a UK travel-to-work area routinely spans several councils and more than one region, so neither has a parent to drill up into. Where a published table says which ZIP area belongs to which metro, we use it and say it is published. Where none exists — a UK neighbourhood can straddle a boundary — we place it in the area holding most of it, label it a best fit, and print the share behind the choice.

‘UK only’ means we never look for it in that country. ‘Not checked yet’ means we would, and haven’t yet for this practice. We keep the two apart.

No page you open calls a model. Every fact on this site is read by code from a register, a listing, an advert library or a practice’s own website, and carries where it was read and when. A model is used in four places, all in batch, all after the fact is read, all with a spend cap, and each answer is checked by code before it is stored: an answer that names a fact the code did not give it is thrown away.

• Prices on fee pages. When a crawl stores a fee page, the page text goes to a small model in a batch to pick out the fees; each price is then checked against the page text and a plausible range before it is kept. Switched off for the monthly re-read of UK websites since 27 September 2026, so that run spends nothing.
• Team pages the parser could not read. Names, roles and registration numbers are read by rules. Only when the rules find nobody on a team page does the page text go to a model, and its answer is kept only where the names appear on the page.
• Which link is the fees or team page, when a website’s menu gives no clue. A one-line question, one link picked.
• The fit verdict (implant reps, imaging reps, labs, marketing agencies): the model reads a numbered list of facts we already hold and says whether it sees an opportunity for that audience; the code keeps the verdict and the facts it cites, and prints those facts, never the model’s prose. Run once for a pilot of six postcode areas; a wider run is a separate decision, with its cost stated first.

The weekly implant grade, the website score, market share, key contacts, the market map and the new-sites ranking use no model at all. “Ask a question” reads your words with rules; a model can be switched on for that, and is not. Two other paid services, neither a model, run only on demand: a screenshot of a practice’s website, and search rankings for a town.

After 90 days for most facts, six months for loans, 13 months for register entries and Google ratings, 18 months for first-seen and registration dates. Old facts stay visible and show their age. We never quietly keep an old fact as current.

Things we cannot yet detect, so we claim no figure: 3D printers, milling machines, how common membership plans are, advertising activity, what reviews say, how ratings move, how many implants anyone places.