Why this page exists
Company reports show numbers and short verdicts. This page is the liturgy behind them: what we are actually looking for when we say a domain has a clear Spirit—or is still crowded with phantoms.
Nothing here is a promise about every URL or every AI product. Each report is a Witness from one audit run on a sample of important pages, not a review of the entire site. Use coverage and certainty on the report as your guide: they tell you how complete and stable that Witness was.
The AIO score (0–100)
The AIO score is a single number that rolls four kinds of evidence into one judgment. Higher means the sampled presence looks easier for answer engines to quote, navigate, and trust. Lower means gaps in passage, flesh, bones, or seals—see below.
On every profile you will see the same four pillar scores underneath the total:
- Technical & crawlability — Can messengers reach the right rooms? We look for whether paths, permissions, and maps make the important parts of the site discoverable, instead of walled off or lost in the maze.
- Renderability & JS — Does truth appear when the "flesh" is stripped away? Heavy interactive layers (content that only appears after the page fully loads) can leave automated visitors reading an empty shell while humans see something else.
- Structure & readability — Are the bones legible? Headings, lists, early summaries, and stable layout help both people and models follow intent without guessing.
- Metadata & consistency — Do titles, descriptions, social previews, and structured labels agree with what the page actually shows? Mismatched seals invite wrong answers.
How to use it: Treat the total as a directional signal from that run. Prefer the four lines underneath when you decide what to fix first—often structure and consistency move the experience even when the headline score lags.
Hallucinations detected (Ghost Test)
Models do not "see" your site the way a designer does. They receive different readings of the same place—what sits behind the scenes in the page record, what a visitor would read on the surface, and plainer text mirrors derived from each. If those readings disagree about facts, an answer engine may confidently cite the wrong one.
The Ghost Test asks the same questions across those readings. We look for rows where one view is empty or evasive while another speaks clearly—a pattern that suggests possession by contradiction: the brand thinks it said one thing, but the record available to automation says another.
What you see on the report: the line Hallucinations detected = X / Y — X is how many sampled questions tripped that inconsistency in that run; Y is how many questions were asked. It is a count of risky disagreement, not a claim that a specific model hallucinated on a live query.
A parable: If a detail is missing from the behind-the-scenes record but obvious on the living page, an engine that trusts the wrong layer might invent a number—or say nothing when it should speak. Several questions showing the same pattern raises the count; agreement across readings lowers it.
Reading AIO and hallucinations together
- Strong AIO, hallucinations detected: The bones and seals can look polished while the Ghost Test still finds disagreement between layers—prioritize making facts match everywhere they appear.
- Lower AIO, none detected: Those readings may agree yet the site is still hard to discover or parse—fix passage, structure, and metadata before chasing marginal ghost noise.
- Both look strong: A good sign for the sampled Witness, not a warranty for every page or engine.