Being unreadable is worse than ranking badly.

When a buyer asks an AI assistant which tool solves their problem, find out whether your site can even be fetched, read and quoted — so you stop losing recommendations you never knew you were in the running for.

That is the whole job. Not a score to improve: a specific uncertainty nobody can currently resolve. Am I even in the room?

Is this you?

If none of these is familiar, this is not your problem and you should close the tab. That is a serious offer — the free check below exists so you can find out in a minute rather than take anyone’s word for it.

• Someone asked and you couldn't answer. “Are we showing up in AI search?” There is no Search Console for this, so the honest answer is “I don’t know how to check.”
• You did the checklist and nothing happened. llms.txt, schema, a sitemap. Weeks passed. You cannot tell whether you did the right thing, the wrong thing, or nothing at all.
• Your robots.txt says yes and your CDN says no. Allow: / for GPTBot, and a 403 to it for six months, because nothing anywhere tells you and every audit tool reads the file rather than the response.
• The page is perfect in a browser. To a crawler that doesn’t run JavaScript it is an empty div, and nobody on the team has ever looked at it that way.
• A 40-page audit arrived and you can’t check a line of it. You suspect parts were generated. You have no way to find out.
• You ship tomorrow. “Is it discoverable?” is a question nobody can answer with evidence, and you would rather not find out afterwards.

Two people commission this, for different reasons

You run the site. A boss asked why you aren’t in ChatGPT answers and you need evidence within the hour, not a theory. A competitor keeps getting named. Traffic is falling while rankings look level, which means the channel your analytics cannot show you.

You run an agency. A client asks why they aren’t showing up, and you need an audit they can verify themselves — so you bill for judgment rather than for a document nobody trusts. Your name goes on the cover, so it has to survive being picked apart line by line.

What arrives

• Every AI crawler tried against your server, by name. 28 of them, one request each, with the status code your server actually returned and which layer returned it — the edge or the file. Most tools read the file. This reads the answer.
• Every page fetched twice, once running JavaScript and once not, with the specific passages that disappear when it isn’t.
• A plain-language report with a limits page listing what was not measured, and every finding carrying its owner and an effort estimate.
• Files you can ship — robots.txt, structured data, sitemap, markdown twins — containing no claim that isn’t traceable to your own page.
• A verification script. After your team deploys, it re-checks that the files resolve and that real crawlers can now reach them. Generated is not deployed, and this is how either of us proves the difference.
• The raw evidence — per-page parity, per-crawler access, and every measurement as CSV and JSON, so anything in the report can be re-derived.

What you’re probably thinking

“It’ll invent things about my product.” Nothing enters a generated file unless it traces to a span on your own page. Any numeric or currency token has to appear literally or the claim is dropped. Dropping a real-but-unmatched fact is the acceptable failure; publishing an invented one is not — and the ratio is shown to you.

“It’s another checklist I’ll ignore.” Two findings are gates, and they multiply rather than add. If AI crawlers cannot reach the page, the other twenty improvements change nothing, and the report says so on page one instead of listing all twenty.

“How much setup is this?” None. The measurement runs with no model credentials at all — the score, the findings and the generated robots.txt are byte-identical with every API key removed, and a test asserts it.

“Isn’t this llms.txt hype?” We generate llms.txt and score it zero. Ahrefs checked 137,210 domains in May 2026: 28% publish one, and 97% of those files were never requested that month. Google states its Search and AI surfaces do not use it.

What else you could do instead

Worth being straight about, because most of these are reasonable and one of them is what nearly everybody actually does.

• Assume it’s fine — Free, no effort. The invisible-block failure mode is specifically invisible.
• Classic SEO tools — Rankings, backlinks, crawl health. They read robots.txt, not per-bot responses, and never diff rendered against no-JS.
• An llms.txt generator — Fast and cheap. Optimises the one artifact with the least evidence behind it.
• Citation trackers — Measure outcomes across engines. They tell you that you’re absent, not the mechanical reason.
• A hand-written audit — Human judgment, breadth. Slow, expensive, and its claims cannot be checked.

What is left unoccupied is the diagnostic layer: can you be fetched, and can you be read? Citation trackers tell you the score. This tells you the reason.

How you’ll know it was worth it

• A trustworthy verdict in under two minutes from pasting a URL.
• Whether a blocker you didn’t know about turned up. That moment is the value — and if nothing is wrong, the report says so rather than manufacturing a problem.
• Fixes you can verify by status code afterwards, not by waiting and hoping.
• Files you can ship without embarrassment: no invented pricing, no fake reviews.

The term

If any number in your report cannot be reproduced by running the commands printed beside it, the invoice is refunded in full.

Deliberately about the document rather than about a result. Whether your score improves depends on whether the fixes get deployed, which is not ours to do — promising an outcome nobody here controls would be exactly the kind of claim this audit exists to catch other people making.

What it does not do

It does not measure whether assistants are quoting you today. Citation measurement covers two engines that have been checked against their real APIs; two more have adapters written and never run, and every figure says which is which. It does not deploy anything — the files are handed over and your team ships them. It reports off-site presence as a floor, not a census.

All of that is on the report’s own limits page, which survives white-labelling, because an audit that overstates its coverage is the failure it was hired to find.

Start with the free one

Run the audit — no account, no API key, nothing to install. It reads your site as every AI crawler and shows you the status codes. If it finds nothing, you have your answer and it cost you two minutes.

If it finds something and you want the full engagement — every page, the report, the artifacts and the verification — roydell2b@gmail.com. Tell me the domain and how many pages matter.