The AO Audit · Published July 18, 2026
Our free tool at aoaudit.com grades how visible a website is to AI engines — ChatGPT, Gemini, Claude, Perplexity, Copilot. On July 15, 2026, we pointed it at our own website. It scored us 69 out of 100. We published that. Two days later a fresh scan said 71. We published that too — along with the reason the +2 wasn't real progress.
Because the ruler changed, not just the site. Between the two scans we added two new checks to the scanner — both of which our site happened to pass — so the score rose partly because the test grew. Our report says this in print: "treat this as the new baseline, not a +2 trend." Scores only mean something when the ruler holds still. Most tools quietly change their scoring and let you feel the progress. We flag it, because a number you can't trust is worse than no number.
The same thing both times: zero linked profiles. No YouTube channel, no LinkedIn, no X. Twenty of twenty-one automated checks pass — the schema, the crawler access, both AI "kill switches" verified off — and the one failure is the one that requires a human to go build something real. The machines can read us perfectly. They just have almost nothing else on the internet to verify us against. (Our own reports tell clients this exact thing weekly. It stings correctly.)
For about four minutes on July 17, our scanner said we passed 21 of 21. A perfect score — and false. The check that looks for social-profile links had matched the words "youtube" and "facebook" inside the scanner's own code, not actual links on the page. We caught it by re-checking against real links, published the honest 20-of-21, and rebuilt the check. If we'd shipped that false pass to a customer, every score after it would be worth less.
Because our whole product is built on one promise: we verify everything the machines read, we test what the machines actually say, and we don't fake the parts nobody can measure. A grading tool that won't grade itself in public is asking you for trust it won't spend on itself. Here's ours, spent: 71, one failing check, one caught false-positive, one honest asterisk on the trend line.
Want your own number? The scan is free, takes about a minute, and shows every check — including the ones you fail.