The SEO agency that never did its own SEO: a public experiment

I run a search agency. For twelve years we’ve built other people’s visibility, and it’s gone well enough that our average client stays 3.2 years and one of them puts £1.5M of revenue down to the organic search we built. Over the same sixteen months in which every client property we measured improved its average search position, our own website earned roughly twenty clicks a month.

I want to be precise about why, because the reason isn’t interesting and the lesson is.

Why we never did it

We were busy growing other businesses. That’s the whole of it. There’s no clever excuse underneath. The work that pays is client work, so the site that gets the research and the baseline and the monthly report is the client’s site, and ours sat there for years doing what most agency websites do, which is exist. It wasn’t neglected on purpose. It was just never treated as a client, and a site that isn’t treated as a client never gets the one thing that makes the method work. Somebody measuring it and ruling on it every month.

I’d guess most agencies are in the same spot and most would rather not say so. I’d rather say so, because the state of the site is what makes the experiment worth running. If our own website had been fine, we’d have nothing to prove. It wasn’t, and we do.

What the site actually looked like

Here’s the baseline, declared on 6 July 2026 from sixteen months of our own Search Console data. I’m publishing it because a public experiment with a hidden starting point isn’t an experiment.

Strip out non-buyer geography and it was around twelve UK clicks a month, for a studio that sells visibility. Our two most-checked commercial pages drew over 53,000 impressions between them and 28 clicks. The searches were happening. We were not being chosen. Eleven thousand impressions for “seo aylesbury”, our own town, at position 20, and zero clicks. The homepage carried 77% of every click the site got, which means the pages that should’ve been doing the selling were doing almost none of it.

One bright spot, and it matters for what comes next. The one page about how we think, /osof/, sat at position six with a 1.9% click-through rate, the best-performing page on the site relative to its exposure. The page that explained the method was the page that worked. Everything else was a brochure.

What we are proving, exactly

The claim isn’t “SEO works”. Nobody needs another agency to say that. The claim is narrower and you can test it. The method we sell, applied to our own business with the same discipline we’d apply to a client’s, moves the same numbers it moves for clients, and we’ll publish the result whichever way it goes.

The method has a name, OSOF, and it’s what we run for every partner. In plain terms it goes like this. Go to market before you write anything, and count what the businesses winning the searches you want are actually doing. Build one substantial page per thing a buyer might want from you, and per place they might want it. Put the proof on the page, character for character, and never a figure you can’t show. Measure all four outcomes a page has to earn, visibility, clarity, trust and conversion, not just the flattering one. Then re-cut on a schedule you announced before you started, so you can’t quietly move the goalposts.

That last bit is the experiment. The site went live on 31 July 2026. The re-measurements are on 6 October 2026, 6 January 2027 and 6 July 2027, same data, same method, against the baseline above, published whether they flatter us or not. I wrote those dates down in July. They aren’t moving.

What we did

We researched the search agencies in the UK whose visibility has been rising for three years, and we counted their pages. How long, how they’re structured, what proof they run, what they ask you to do next. Not to copy them. To know the weight of the field before we wrote a word. Then we rebuilt the whole site to that standard over July, every page from research, every decision recorded and numbered. One page per service. One page per town we serve. A research layer on how the AI assistants choose who to cite, because that’s where a growing share of buyers now ask. And a journal rebuilt to the length and apparatus the field actually publishes at, rather than the length we’d been publishing at, which turned out to be about half.

Then we switched it on and started measuring.

What the first seven weeks say, and what they do not

Here’s the honest read, in the order the numbers move.

The number of buyer searches where we hold a first-page position with meaningful exposure went from 10 to 93 between one 28-day window and the next. Impressions went from roughly 8,400 a month at baseline to 59,277 in the latest window. UK clicks are running at about three times the baseline rate. The homepage’s share of clicks fell to 61% because the service and location pages started carrying their own, which is what the rebuild was for. Three of the four answer-engine research pages reached page one on their first crawl.

Now the part I’d rather write myself than have someone else point out. Most of that impression growth is footprint, not visibility. The site went from a dozen indexed commercial pages to over a hundred, Google is sampling the new ones deep in the results, and the average position got worse before it gets better. Seven weeks isn’t a result. And our most valuable old URL, the Aylesbury SEO page, spent five weeks redirecting to the wrong new page because of a build mistake, which cost that term about half its impressions until we found it in the data and fixed it the same day. That’s in the report too. A public experiment that edits out its own errors is a brochure with a chart on it.

What you can take from this now

You don’t have to be an agency to be where we were. Most businesses have a website that was built, launched and then left, while everyone got on with the work that pays. If that’s you, three things from our own baseline are worth an hour of your time.

1. Find out whether your problem is visibility or being chosen. Ours was the second. We were seen over fifty thousand times and clicked twenty-eight. Search Console tells you this for free, and it changes what you fix first.
2. Look at which single page is doing your acquisition work. If it’s the homepage, the pages that should be selling aren’t, and no amount of traffic fixes a page that can’t convert it.
3. Write your re-measurement dates down before you start, and tell someone. The discipline isn’t the work. It’s the schedule you can’t wriggle out of.

The rules of the experiment

So there’s no ambiguity later, here’s what I’ve committed to.

• The baseline is published, with its method, and it doesn’t change.
• Every re-cut uses the same export, the same window logic and the same calculations, and goes out on the dates above.
• Every figure in our own copy traces to a locked source. Anything we can’t yet verify shows as a visible placeholder, not a guess.
• Build mistakes get published alongside results, with the date we found them and the date we fixed them.
• If the method doesn’t move our numbers, that’s the finding, and it gets published too.

Watch it happen

The report lives at our research library and the next re-cut lands on 6 October 2026. If your own website is where ours was, the free audit runs the same first diagnosis we ran on ourselves, and we’ll tell you which of the four outcomes is the one you’re losing.