SEO
Our own Google snippet was junk. Here is the ledger.
We sell SEO to clinics. On 3 October 2026 I found our own site failing the test we set for clients, so we fixed it in a day and wrote it all down.
By Pete Flynn · 3 October 2026 · 8 min read
Last week I googled "Gold Coast healthcare marketing" to see where we sat. We were on page four, and the text under our title said our page was built for a 50 km coastal strip and the M1 spine. No clinic owner has ever said those words. If I were a physio on the Gold Coast reading that, I would have kept scrolling. We are about to start running SEO for clinics, so the only honest move was to run the same check on ourselves first and publish what we found.
What the snippet told me
Google writes the text under a search result from the page's meta description when it fits, so that line was ours. Someone, and that someone includes me because I signed off on it, had written the Gold Coast page for a map rather than a person.
One bad description is a typo. The question was whether it was one, or whether the whole site had the same disease. So I sent three agents through every page: all 172 URLs in our sitemap, the code that generates them, and a month of Search Console data.
What they found on 172 pages
68 of 172 meta descriptions were longer than 155 characters, which means Google cuts them off mid sentence. One of the article descriptions ended mid sentence even before Google got to them.
Seven of our eight city pages used the same sentence shape, "built around the X and the Y", with roads and rail lines filled in. Sydney had a corridor, Perth had a freeway, the Gold Coast had its spine.
37 pages had no social sharing description of their own and fell back to the homepage's. Two legal pages pointed their canonical tag at the homepage instead of themselves.
Our schema markup told Google this agency had a five star rating from 200 reviews. The reviews are real, but they belong to Clinic Mastery, the parent community, not to this agency, and Google does not accept a rating a site publishes about itself in any case. The same markup listed two kinds of clinic we do not serve as services. We have never had a page for any of them.
The 33 profession pages were built from one skeleton. Twenty two of them had a section heading that opened with "Most agencies". Slugs had leaked into the copy: the web address form of a profession name, hyphen and all, sitting in a sentence. The American spelling of enquiry, 52 times.
On the city pages, 33 of 53 factual claims had no source we could point to, and five contradicted another city page. Two cities were each described as having the lowest cost per click in the country.
The Search Console numbers matched the copy. In September the eight city pages earned 1,383 impressions and 2 clicks, with average positions between 22 and 68. The audiology and osteopathy Google Ads pages, written by hand for a reader, sat at positions 9 and 6.
Presence checks are not quality checks. Every page had a title, a description and an H1. Almost none of them would survive being read aloud.
What we changed on 3 October
Head hygiene first. Every title under 60 characters, every description between 120 and 155 characters ending in a full stop, written to be read by a person. Every page given its own sharing text and its own canonical. The self published rating and the claims for services we do not sell removed from the schema. That was PR #171.
The eight city pages rewritten as if a clinic owner in that city were reading them. Every remaining factual claim now has a source listed in the pull request, and the claims we could not source are gone. Hobart is no longer the "smallest mainland equivalent capital", because it is not on the mainland. PR #179.
The 33 profession pages given their own opening paragraphs. Slugs out of sentences. Australian spelling throughout. The slogan headings on the home and service pages, lines like "Four steps. No mystery.", replaced with headings that say what the section is. Filler words trimmed from the articles, by deletion, so no new claims crept in. PR #180.
The rules written down as code. A lint now runs on every build and fails it if any page breaks a rule: description length, title length, banned infrastructure words in a title or description, one H1, own canonical, no self rated schema, no em dashes, no model vocabulary, no slugs in prose, no American spelling, caps on filler words, clinic counts only from an approved list. PR #176. The same lint is what we will run on client sites. The same lint now runs on every clinic website we build before it goes live.
The before and after, in numbers
Before: 68 descriptions over length, 37 pages on the homepage sharing text, 2 wrong canonicals, 1 self published rating, 2 services we do not sell named in schema, 7 city descriptions on one template, 22 pages with a heading opening with the same two words.
After: 0 descriptions over length, 0 pages on the homepage sharing text, 0 wrong canonicals, 0 self published ratings, 0 services we do not sell, 0 city descriptions on a template, 0 pages with a heading opening with "Most agencies".
The crawl files behind both lines, and the code sweep counts for the schema, opener and claim figures, are committed to the site's repository in the SEO ledger folder dated 3 October 2026, so anyone can check the counts rather than take my word for it.
What we are watching from here
Baseline
September 2026, eight city pages
1,383 impressions, 2 clicks, average positions 22 to 68, from Search Console for 3 to 30 September.
Signal
Position on the city queries
"Clinic marketing" and "healthcare marketing" with each of the eight city names. These are now in our weekly Search Console report.
Read date
31 October 2026
Twenty eight days after the rewrite. I will append the numbers to this article on that day, whatever they say.
What would count
Clicks, not impressions
Impressions can rise while nobody clicks. The real measure is whether a clinic owner in one of these cities reads our result and chooses it.
What this does not prove
Rewriting metadata does not move rankings on its own, and I am not claiming it will. A five month old domain with about 25 clicks a month is not going to outrank agencies that have been publishing for a decade because its descriptions got shorter.
What it does is remove the reason a person who finds us would not click. That is the part of SEO we can control in a day. The rest is the slow work of writing things worth ranking, and this article is one of them.
If you run a clinic and want to know whether your own site passes the same test, send us the address and we will run the lint on it and send you the result. No pitch deck attached.
Your site
Run the same check on your clinic website.
We will send back what the lint finds, in plain words.
Send us your websiteThe 28 day read, 31 October 2026
I said I would append the numbers on 31 October, whatever they said. Search Console's final figures run three days behind, so this read covers 4 to 28 October 2026, which is 25 days. The September baseline covered 28 days, 3 to 30 September. October had three fewer days to collect anything, and I have not scaled the numbers to make up for it.
The eight city pages had 2,039 impressions and 4 clicks, against 1,383 impressions and 2 clicks in September. Canberra had 2 clicks, Melbourne 1 and the Gold Coast 1. Sydney, Brisbane, Perth, Adelaide and Hobart had none.
Page by page, September first and October second. A lower position is better, and position 10 is the bottom of page one. Sydney went from 346 impressions to 354, and from position 51.0 to 48.5. Melbourne went from 293 to 455, and from 66.8 to 57.7. Brisbane went from 458 to 727, and from 61.5 to 56.8. Perth went from 36 to 162, and from 67.6 to 39.5. Adelaide went from 108 to 223, and from 38.3 to 49.6. The Gold Coast went from 71 to 48, and from 50.2 to 35.0. Canberra went from 31 to 41, and from 34.7 to 38.1. Hobart went from 40 to 29, and from 22.1 to 10.3.
So six of the eight pages sat higher on average than in September, and Adelaide and Canberra sat lower. Across the eight, average positions ran from 22 to 68 in September and from 10 to 58 in October.
On the searches I said we would watch, the Melbourne page showed for "healthcare marketing melbourne" 57 times at an average position of 22.9, against 40 times at 59.5 in September. The Sydney page showed for "healthcare marketing sydney" 50 times at 32.5, against 52 times at 39.4. The Brisbane page showed for "healthcare marketing brisbane" 11 times at 22.8, against 69 times at 25.7. None of the watched city searches brought a click in either month. No search for "clinic marketing" with a city name appeared in the rows Search Console gave us in either month. Google holds back rare searches, so that means very few people searched it, or none.
Across the whole site it was 5,148 impressions and 28 clicks, against 3,221 and 22 in September. The audiology and osteopathy Google Ads pages, the two written by hand, sat at positions 7.3 and 4.7.
Four clicks is too few to draw a line from. It is two more than September, over three fewer days, and two of them came from one city. Positions moved up on six of the eight pages, and that is as far as I will take it. Rankings move for reasons that have nothing to do with us, and one month cannot tell those apart from our changes. The measure is still the one I set on 3 October: clicks from clinic owners in these cities. October gave us four.
The Search Console export behind every number here is committed to the site's repository in the SEO ledger folder dated 31 October 2026, next to the September baseline. The same lint ran on the live site that day and found nothing on 179 pages.
Common questions
