I Took My Own Medicine: 97 out of 100, and the Three Points I Missed
The question I knew was coming
I was talking tech to a friend online recently when aeocheck.net came up. He asked, thinking he'd catch me off guard, "So what does your site score?"
I had to be honest. I did not know. I had run it against earlier versions of the scoring engine and I remembered those numbers, but the rubric has changed a lot since then and I had never gone back and re-scored it.
So I said, "Let's see."
I ran aeocheck.net through the actual Full Audit pipeline. Same crawler, same scoring engine, same $39.95 product anybody else buys. No special cases, no thumb on the scale. It came back 97 out of 100, grade A.
That number impresses me a lot less than it might impress you.
Grading your own homework
I wrote the rubric. I know every check in it, in order, by heart. And then I built a site while looking at that rubric.
Scoring well on a test you wrote yourself is not evidence of much. It means I can follow my own instructions. The useful part here is not the 97. It's the breakdown underneath it, and the fact that I still dropped three points with the answer key sitting open on the other monitor.
Where the points went
| Module | Score |
|---|---|
| AI Crawler Access | 15/15 |
| Structured Data | 28/28 |
| Content Structure | 17/20 |
| Entity & Trust | 20/20 |
| llms.txt | 5/5 |
| Technical Health | 12/12 |
| Total | 97/100 |
Five modules clean, one not.
Crawler access came back 15 out of 15. GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, and Bingbot are all allowed in robots.txt, and the homepage is indexable. This is the module most sites blow, and they almost never blow it on purpose. It's a blanket Disallow that came with a theme, or a CDN toggle somebody flipped in 2023 and forgot. Ten minutes to fix, and it gates everything else. If GPTBot can't read your site, nothing further down this list matters even a little.
Structured data came back 28 out of 28. SoftwareApplication schema with all the expected fields, FAQPage schema with five question and answer pairs, AggregateRating with a real rating value and review count. If you only do one thing on this whole list, do the FAQPage block. It's the format the answer engines reach for most readily and it's one chunk of JSON-LD.
Entity and trust, 20 out of 20. Pricing is public, the legal pages exist, there's a named human with a byline and author schema attached to an actual credential, the third-party review links are in the schema, and there's an About page. All of that is answering one question on the engine's behalf, which is roughly "is there a real company back there, or is this a parked domain with opinions?"
llms.txt came back a perfect 5 out of 5. File's present, H1, blockquote summary, a linked section. And here's the honest limit on that one: it's a five-point module because llms.txt adoption is still early days. It's a cheap early-mover signal. It is not load-bearing, and I'd be selling you something if I said otherwise.
Technical health, 12 out of 12. HTTPS, sitemap, canonical tag, mobile viewport, date signals, main and nav landmarks. Exactly one H1, no skipped heading levels, and an internal link graph with no orphans and nothing buried more than three clicks deep.
The part that bugs me
Content structure, 17 out of 20.
Four of the five checks passed. Six question-format headings, all six answered directly underneath, three of them in a list or table, and a visible FAQ section. Good.
Then the fifth check:
Pages average 311 words and 2 of 6 are under 300 words.
That's straight out of the report, word for word. Our pages are thin. Two of the six pages the crawler reached are under 300 words and the site-wide average of 311 clears the bar by about the width of the bar.
And yes, six pages. We're a small site with a short menu. I ran the Full tier anyway because I wanted every check in the thing to fire, not because we needed 50 pages of coverage.
I'm not going to talk my way out of that one. It's a fair hit. A 300-word page crawls fine. It just doesn't give an answer engine much to work with when it's picking between your page and somebody else's on the same subject. Thin pages don't get you blocked. They get you passed over, which is worse, because nothing looks broken.
What I'm doing about it
Nothing yet. On purpose.
I wanted the score published before the fix went in, rather than quietly patching the site over a weekend and then posting a nice clean 100 like it had always been there. That's the kind of thing that makes a dogfooding post worthless.
The fix the report hands me is the obvious one. Take the two thin pages past 300 words with actual substance: feature explanations, use cases, the questions people keep asking. Not padding. It's a writing job. No developer required. When it's done I'll come back and note the new number here.
If 97 turns into 100 for about an hour of writing, that's sort of the whole point. Most AEO problems are small, specific, and fixable in an afternoon, once somebody tells you exactly which afternoon and exactly what to type.
What this proves, and where it stops
A 97 proves the pipeline runs end to end against a real site and comes back with a specific, actionable finding. That's it. That's the claim.
It does not prove our site is well known, well trafficked, or getting cited by AI engines all day long. Different question entirely. Technical readiness is a precondition for that, not a substitute for it.
I lived through internet bubble one and bubble two, and both of them ran on people conflating those two things. Being findable is not the same as being famous. This audit measures findable. A site can be perfectly findable and still be invisible, because nobody links to it, nobody's heard of it, and there's nothing on it worth quoting.
What a clean audit buys you is narrower than that, and I'd rather say the narrow true thing. When an answer engine has a reason to cite you, nothing on your end is standing in the way.
Anybody promising you the other thing, that they'll get you into ChatGPT, is selling you an outcome they do not control.
Go run yours
The free Grade Check scans one page and hands you the same 0 to 100 score and letter grade off the same engine that produced that 97 up there. Takes about a minute. Costs nothing.
Run it on your own site before you run it on a client's. If it comes back an A, cool, go read the module breakdown anyway and find out which checks you passed by accident, because some of them you did. If it comes back a C or lower, something specific is broken and it is almost certainly sitting on the list above.
Either way you'll have a number. Which is more than most sites can say.