flagsmith

The Ghost test is a short Q&A pass on pages we sample. We ask ChatGPT the same question twice: once using the page's raw HTML (what you get from a simple fetch, before JavaScript runs—like a basic crawler), and once using the visible text after the page's JavaScript has run (what we capture with a real browser). If the raw-HTML pass can barely answer or says the information isn't there, but the fully rendered pass can answer in a clearer, fuller way, we flag hallucination risk—because an AI that only saw the static HTML could answer very differently from one that sees the page the way a visitor does.

The AIO score combines four pillar scores—how discoverable the site is, how reliably content shows up for automation, how clear the layout and wording are, and how well titles and previews match the page. The total is shown as a single 0–100 number so you can compare runs over time.

Confidence reflects how complete and consistent the evidence was in this audit. Higher confidence means we had enough stable signal to treat the results as a stronger guide; lower confidence means you should treat it as directional and validate anything critical.

Structure is about organization and readability. Hallucination flags compare whether answers match across different ways automation reads the same page. A site can look well organized yet still show different facts in different readings, which is how inconsistent AI answers slip through.