Ten days ago I published a 10-point audit for AI readiness and noted that most enterprise systems I look at pass three or four points. Then I applied the same checklist to davidglazier.com, the site that published it.
Baseline score: 5 pass, 1 partial, 4 fail. Call it five and a half out of ten. I write about this every day, and my own site barely cleared half.
This article is the full account: what failed, what it was quietly costing me, and what it took to fix. Every number here traces to the two audit documents I froze before and after the fix pass. If the checklist is the theory, this is the lab report.
What did the audit actually catch?
Most of the failures were the unglamorous kind: no structured data on any money page, so the case-studies page presented verified outcome numbers as unanchored strings a machine cannot attribute. No per-field error text on the contact form. No canonical tags anywhere, so clean URLs and .html URLs both answered with identical content and nothing told machines which one was the URL of record.
Then there was the finding that mattered.
The contact form is this site's only conversion path. The audit found it broken three ways at once. The form tag was missing the attribute that tells Netlify to capture submissions, so posts went to a page where nothing received them. The submit handler treated any resolved network response as success, so a visitor saw the thank-you message even when the submission went nowhere. And with JavaScript disabled, submitting was impossible.
No design review would have flagged it. The form looked perfect. The thank-you message was warm and well typeset. Every visual signal said this works, and nothing worked.
What did it cost, honestly?
Here is where the story gets lucky, and I want to be precise about the luck.
The broken form shipped in a site rebuild on July 12. The audit caught it on July 13, roughly 18 hours before a LinkedIn launch was scheduled to drive traffic at that exact form. The exposure window was about one day, before any real traffic arrived. When I checked the capture dashboard's history, the previous form had recorded 8 submissions between March and April, every one of them my own tests. No real inquiry was lost.
But notice what else that history says: no notification was configured, which is why months of test submissions went unnoticed. If the audit had not run, the launch would have sent real prospects at a form that thanked them and dropped their message. I would have read the silence as a failed article instead of a broken pipe. That misread is the real cost of machine-illegible failure: you do not just lose the lead, you learn the wrong lesson from losing it.
What changed in one evening?
The audit's own priority rule is fix what blocks transactions first. That ordering did the thinking. The fix pass landed in three commits the same night, and the after-pass re-run scored 9 pass, 1 fail. What changed:
- The form is now captured natively by the platform, and I verified it the strict way: a raw HTTP submission with zero JavaScript, confirmed landing in the dashboard with a timestamp.
- The success page only appears when the platform actually accepts the submission. No more self-reported victory.
- An email notification fires on every submission, so silence now means silence.
- Every form field got text error states with proper roles, so failures are legible to screen readers and agents alike.
- All six money pages got Person and service structured data, canonical tags went sitewide, and the last JavaScript-dependent content was removed.
The one remaining fail is the machine-readable component inventory, deferred on purpose: for a site this size it is real work for marginal return, and honest scoring beats a vanity ten. The point of an audit is a prioritized list, not a perfect grade.
What should you take from this?
Three things, in order of how much they are worth.
First: audits find a different class of bug than reviews. Everyone on a design review sees the thank-you message. Only a machine-eye pass, reading what the code says rather than what the screen shows, catches that the message is a lie. The failures that cost transactions are exactly the ones that look fine.
Second: knowing the checklist does not protect you from shipping the bug. I wrote the checklist. I shipped the bug the same week. The protection is not expertise, it is the habit of running the audit against production, especially right after a rebuild, which is precisely when confidence is highest and regressions are freshest.
Third: the fix is fast when the finding is precise. Five points flipped in one evening because the audit said exactly what was broken, what it blocked, and what could wait. The expensive version of this story is the vague one: something feels off with leads, and a quarter disappears into guessing.
Frequently asked questions
What does it mean to fail an AI-readiness audit?
It means machines cannot reliably find, understand, or operate your product through the surfaces you ship. Scoring is per point, so a fail is specific and fixable: mine included a conversion form that reported success on submissions nothing received.
How do I know if my contact or checkout form is silently failing?
Three checks. Submit it with JavaScript disabled and see whether anything is captured. Confirm the success message only appears when the server actually accepted the submission, not whenever a network call resolves. And verify a notification fires somewhere a human reads. If any of the three fails, you may be thanking visitors for messages you never got.
How fast can AI-readiness failures be fixed?
Faster than most teams expect once the problems are named and ordered. My fix pass went from 5 passing points to 9 in one evening, because the audit said what blocked transactions and what could wait. Larger systems typically land the same class of fixes over a quarter.
Run it on yourself before someone's agent does
The checklist that caught this is public: the 10-point AI-readiness audit. It cost me an uncomfortable score and saved my launch. Grade your own product honestly, starting with the flow your revenue runs through.
If you would rather have the audit run with real agents on your real flows, with findings ordered by what actually blocks transactions, that is the work I do.
Book Advisory Intro