On 8 July our own site scored 53 out of 100 on the GEO audit we sell. On 11 August the same audit returned 77. Below is what changed between those two numbers over five weeks — including the work that produced almost no measurable movement at all.
We publish this for a simple reason: an agency that sells measurement should show its own measurements. The first audit is here if you want the starting point.
The two scores, side by side
| Category | 8 July | 11 August | Change |
|---|---|---|---|
| AI platform readiness | 60 | 79 | +19 |
| Content quality / E-E-A-T | 72 | 92 | +20 |
| Technical foundation | 55 | 82 | +27 |
| Structured data | 70 | 90 | +20 |
| Brand authority | 5 | 42 | +37 |
| Composite | 53 | 77 | +24 |
What moved the number most
1. Putting a human on the site (+20 on content)
The single biggest content jump came from the least technical change. Every article was signed "Mentio". Now every article is signed by a named person, with a Person node in the structured data, linked by @id to the organisation as its founder.
Nothing else about the articles changed. Same words, same structure. The score moved because E-E-A-T is not a writing-quality metric — it is an accountability metric. An unsigned page has nobody standing behind it.
2. Existing on more than one website (+37 on authority)
Brand authority went from 5 to 42 purely by existing in places we did not control. A company page, a Crunchbase entry, business directories, a review platform, a social profile — six in total, each one saying the same name, the same description, the same URL.
The number is still low, and it should be. Forty-two out of a hundred is an honest score for a brand that is a few weeks old — six profiles exist, but not one of them holds a review yet. Presence is not authority. This is the category that cannot be rushed.
3. llms.txt, FAQ markup and language versions (+19 and +20)
These are the standard technical wins and they behaved exactly as expected. A machine-readable summary of the site, question-and-answer markup that mirrors how people actually ask, and six language versions correctly cross-linked with hreflang.
The stubborn category — and what finally moved it
Technical foundation spent weeks refusing to move. We put real time into it and gained five points, because our host could not send custom HTTP headers at all. No amount of on-page work fixes a limitation that lives one layer below your site. The fix was not on the site: we put a proxy in front of it. Security headers now ship on every page, and the category jumped from 55 to 82 — the largest single gain on the board. The trap we hit doing it is a separate story.
Publishing more articles, on its own, moved almost nothing. The blog grew from nine posts to sixteen and the content score noticed barely at all. Volume is not a ranking input for AI systems the way it sometimes is for classic search. Past a point, the only thing that moves the number is what other people say about you.
What is still broken
Zero reviews. No AggregateRating, no Review markup, because we have no real reviews to point at. We will not invent them: fabricated ratings get the markup stripped from search results, and for an agency selling trust that is the worst possible trade. This stays broken until real clients leave real feedback.
No verified business listing. Partially a choice, partially a platform limitation for our location.
The honest lesson
On-page GEO is a sprint you control. Crawler access, structured data, llms.txt, content shape, even the header problem that needed a proxy — every on-page category jumped at least nineteen points in five weeks.
Off-page authority is a season you schedule. It went up thirty-seven points and still sits at forty-two, because it is built from other people's websites mentioning you, and you cannot deploy that on a Tuesday.
If someone tells you they can fix both in the same sprint, ask which one they are lying about.
Want to know how AI assistants see your website? Run the free check — we reply with a summary of what ChatGPT, Perplexity and Gemini currently say about your business.
Check my website for freeRead next
Case study: we audited our own site. Score: 53/100 — and that's the point
We ran the same GEO audit we sell on mentio.agency itself, two days after launch. Full category scores, what earned 88-98, what earned 5, and why the gap is the roadmap.
Security headers without breaking AI visibility
Adding a CDN proxy for security headers can silently block AI crawlers and break your own site. The two traps we hit doing it on mentio.agency, and how to avoid both.