TheToolistReviews50 Days With an SEO Operator
Case Study·October 2026·9 min read·James Merrick

I Ran an Autonomous SEO Operator for 50 Days. The Last 8 I Did Not Look At All.

Real telemetry from test.cleoseo.com, 15 August to 3 October 2026. The score falls in places. That is staying in.

Full disclosure

Full disclosure, before the numbers: Cleo is our own product. Pure Light Ltd builds it. This is not a review of a stranger. It is the raw telemetry from our own test site, published in full because the only case study worth reading is one that shows the dips. If you want to check my homework, the free scorecard at cleoseo.com/scorecard runs a deterministic structural scan (robots directives, schema, llms.txt, sitemap health) on any domain and scores it 0 to 100. Yours included.

What it is

Cleo is an autonomous SEO operator. It crawls your site every morning, works out what is broken, fixes what it can in the HTML as the page is served, and files a readout before you are awake. No plugins, no tickets, no "we will get to it this sprint".

What I did

On 15 August I pointed it at test.cleoseo.com, our WordPress test bed, and left it running. No tuning, no babysitting. The site changed underneath it during the run: new pages landed, the page count went from 4 to 6, and every one of them brought fresh problems with it. Fifty mornings later, here is the whole log.

The story in three chapters

Eighteen days of steady state. Score 72. Four pages, six known issues. The operator held the line.

2 September: 90. Three issues. The cleanest the site has ever been. Fixes had stacked up and stuck. I want to be clear that I did nothing that morning. It did the work.

3 September: the site grew. A new page landed and the score fell to 58, because the new page arrived with fresh problems. The crawler found them the same morning. Nobody pointed it at the new URL. It found the page because that is what crawling means.

23 September: a sixth page. Same story. Caught the same morning, issues from 8 to 10, score 54.

24 September: the worst day in the log. Eleven issues, score 50. By the next run it had one less problem and 4 points back.

25 September to 3 October: I did not log in. Eight days. No settings changed, no nudges, no fixes from me. The last seven runs are identical: 54, 54, 54, 54, 54, 54, 54. This morning, with the run landing at 07:01 UTC, it found 6 pages, 10 issues, and it handles 8 of those 10 automatically at the edge on every single request. Two want a human. Fair enough.

The full 50 days

Date (run)HealthPagesIssues
2026-08-157246
2026-08-167246
2026-08-177246
2026-08-187246
2026-08-197246
2026-08-207246
2026-08-217246
2026-08-227246
2026-08-237246
2026-08-247246
2026-08-257246
2026-08-267246
2026-08-277246
2026-08-287246
2026-08-297246
2026-08-307246
2026-08-317246
2026-09-017246
2026-09-029043
2026-09-035858
2026-09-045858
2026-09-055858
2026-09-065858
2026-09-075858
2026-09-085858
2026-09-095858
2026-09-105858
2026-09-115858
2026-09-125858
2026-09-135858
2026-09-145858
2026-09-155858
2026-09-165858
2026-09-175858
2026-09-185858
2026-09-195858
2026-09-205858
2026-09-215858
2026-09-225858
2026-09-2354610
2026-09-2450611
2026-09-2554610
2026-09-2654610
2026-09-2754610
2026-09-2854610
2026-09-2954610
2026-09-3054610
2026-10-0154610
2026-10-0254610
2026-10-0354610

Health is a 0 to 100 structural and technical score. Issues are problems found on that morning, not a cumulative count.

What is good

It does not miss mornings. Fifty runs in fifty days, including eight when I was nowhere near it.

It finds new problems on the day they arrive. Both page additions were caught the same morning they appeared.

The fixes are real HTML. They are applied in the response the browser and the crawler actually receive, not sprinkled on top with JavaScript after the fact.

The score is honest. It went down when the site got messier and stayed realistic. A tool that only ever shows me a rising line is a toy.

What is not

It fixes eight of ten issues, not ten of ten. Two want a human or changes at the origin. It is an operator, not magic.

The score is a structural read, not a ranking promise. Nobody can promise you rankings. Walk away from anyone who tries.

54 out of 100 is not a testimonial number. I could have opened this piece with 2 September, at 90, and stopped there. That would be lying with numbers.

It is our product. You are reading our dogfood run, not an independent lab. Which is exactly why that scorecard link exists at the top: run it yourself, on your own domain, and compare the shape of the numbers.

Verdict

The category is the takeaway. Monitoring tools tell you what is wrong and leave you to it. This is the other kind, the kind that does the work while you sleep. If you run a WordPress site and you are the bottleneck between "known issue" and "fixed issue", look at it properly. Start with the free scorecard. If your worst problems are things you can fix yourself this afternoon, go and do that first. Skip this if you enjoy alt text.

If it is good, I will tell you. If it is not, I will tell you louder.

Questions, answered

What is an autonomous SEO operator?

A monitoring tool tells you what is wrong and leaves you to fix it. An autonomous SEO operator does the work itself: it crawls your site every morning, finds broken structure, and applies real HTML fixes in the response your pages are served with. Cleo is one of these. The distinction that matters is who does the fixing, you or the software.

Is this an independent review?

No, and the article says so at the top. Cleo is our own product, built by Pure Light Ltd. This is our dogfood run on our own test site, published in full with the dips left in. The free scorecard at cleoseo.com/scorecard runs the same deterministic structural scan on any domain, so you can check the shape of these numbers against your own site before believing a word of it.

What does the health score measure?

Health is a 0 to 100 structural and technical score: robots directives, schema, llms.txt, sitemap health and related signals. Issues are problems found on a given morning, not a cumulative count. It is a read of site structure, not a ranking promise. Nobody can promise you rankings.

Check your own site before you trust my numbers.

The free scorecard runs the same deterministic structural scan on any domain. No account, no card.

Run the free scorecard