Why a separate audit exists at all
A separate audit exists because AI assistants and Google draw from different pools. Across 1,000 consumer ranking queries, the domains GPT-4o cited had a median overlap of 0.0 percent with Google's top ten, in the University of Toronto analysis presented at the EDBT/ICDT 2026 workshops. A site can hold page one and be absent from every AI answer in its category. An SEO audit will not see that, because it is not looking.
The structure below is the one our own free analysis uses. We publish the weights so a reader can argue with them; a score nobody can question is a score nobody should trust.
1. AI visibility: are you named today? (30 points)
The first section asks the four main assistants whether they name the business for its category and market, and records what they cite instead. ChatGPT, Perplexity, Claude and Gemini are queried in the market's language, and Google's AI features are checked separately. Each engine that names the business earns its share; an engine that errors out costs nothing.
One honesty note belongs here. Only 34 to 42 percent of cited sources stay the same from one day to the next, and researchers at the University of St. Gallen recommend at least seven runs per prompt before a mention rate is reliable. A free scan runs each query once, so this section is a snapshot, and the report says so. The paid engagement re-measures on the method the research asks for.
2. Entity and authority: do you exist as a thing, not a page? (25 points)
The second section checks whether the business is described as an entity a machine can recognise. That means an Organization node in the structured data with a name, identifier, logo, address and phone; profile links that resolve rather than return 404; the brand name in the title and main heading; a Google Maps listing; and, with a small weight, domain rank.
The weight is high because the evidence is consistent. In Ahrefs' analysis of 75,000 brands, mentions of the brand across the web correlated more strongly with AI visibility than Domain Rating or branded search volume. Entity completeness alone carries 12 of the 25 points; domain rank carries 2, because the same analysis puts it well down the list. Our brand entity article covers the mechanism.
3. Evidence and content: what do your pages actually say? (25 points)
The third section reads up to eight pages, the homepage, three service pages, two commercial pages and three content pages, and scores what they contain. Five lines:
- Specificity (7): are prices, timescales, scope and conditions stated outright?
- Statistical evidence (6): numeric claims per thousand words.
- Source traceability (6): of the pages that make claims, how many link to an authoritative source. A business with nothing to cite is not penalised; the line is marked not applicable and the points are redistributed.
- Topical coverage (4): is the subject covered, including the secondary questions a buyer asks?
- Explanatory depth (2): are terms defined and alternatives compared?
Three of these need a reader, not a regular expression, so a language model grades them against fixed written anchors and must quote the sentence that justified each grade; a quote that does not appear on the page throws the grade away. The weights follow the research: in the founding experiments at KDD 2024, adding statistics and citations raised visibility in generative answers by up to 40 percent, and across 18,151 cited pages Zhang, He and Yao found pages with numbers, definitions and comparisons absorbed 55 to 62 percent more strongly.
4. Machine readability: can a parser use it? (12 points)
The fourth section scores the plumbing a machine reads: structured data types (5), FAQPage schema (1), content freshness (2), a second language with genuinely different content (2) and the accessibility tree (2).
Two weights deserve a word. FAQ carries one point, not five, because Zhang's data shows Q&A formatting on its own slightly below average for absorption; the value sits in what the answers contain, which section three already scores. Freshness earned its two points from the Toronto study: AI engines cite content two to three times younger than Google for the same queries, and they read the age from datePublished, dateModified and the time element. A page with no machine-readable date has no age at all.
5. Crawl access: can they get in? (8 points)
The fifth section tests whether AI crawlers actually receive the page. Half the points go to a live test: the page is fetched as a browser and as GPTBot, and the two responses are compared. A robots.txt that allows GPTBot while the server or CDN returns an error is the finding nobody else checks, and it is scored as a block because that is what it is. The remaining points cover robots.txt, sitemap, canonical tag and llms.txt at one each.
Google states that no special schema, llms.txt file or AI-specific format is required for AI Overviews; the llms.txt point is there for the assistants that do read it, and it is one point for that reason.
Search health: the same report, a different score
Search health is reported beside the AI score, not inside it. It covers on-page quality, mobile speed, Core Web Vitals from field data, crawl errors, SSL, layout shift, viewport and HTTP/2. These decide how Google ranks the site and whether a visitor stays, so they belong in the report. They do not decide whether an assistant names the business, so they do not belong in that score. Google is clear that technical eligibility is a precondition for its AI features, which is why crawl access sits in the AI score and speed does not.
What the total means
| Score | Band | What it usually means |
|---|---|---|
| 0–30 | Critical | Not named, weak or missing entity data, pages without evidence |
| 31–45 | Vulnerable | Some structure, little evidence, rarely named |
| 46–65 | Developing | Entity in place, evidence uneven, named by one engine at most |
| 66–80 | Solid | Named by some engines, sourced claims, dated pages |
| 81–90 | Strong | Named by most engines, complete entity, cited evidence |
| 91–100 | Rockstar | Consistently named; the audit finds little to fix |
Our own site scored in the developing band when we first ran the tool on it. The reason was section one: we ranked for informational queries and were not named for commercial ones. That is what an audit is for.
Frequently asked questions
Is an AI visibility audit the same as an SEO audit?
No. An SEO audit measures how Google ranks and renders the site. An AI visibility audit measures whether assistants name the business and whether its pages carry the evidence they absorb. The two share the crawl-access checks and little else, which is why we report them side by side.
How long does the free scan take?
Minutes. The engines are queried, up to eight pages are read and graded, and the report is compiled and sent. The reading step is the slowest because a model has to read the pages.
Why is one run not enough?
Because the engines are probabilistic. The same prompt minutes apart returns a different set of sources about as often as the same prompt on different days. One run tells you whether you can be named; several runs tell you how often you are.
What if my pages make no statistical claims?
Then the source-traceability line is marked not applicable and its points are shared among the other lines. A business with nothing to cite is not failing a citation check. Specificity, however, still applies: prices, scope and timescales are claims too.
Does structured data guarantee a mention?
No. Google says no special markup is required for its AI features, and no assistant promises a mention for a schema type. Structured data makes the entity recognisable; the evidence on the page decides whether it is worth citing.
Can I run the audit on a competitor?
Yes, and you should. The most useful line in the report is often who is named instead of you and what their pages contain that yours do not.
Run it on your own site
The free analysis produces exactly this report for your domain, your market and your language, with every line scored and every grade quoted. It takes a few minutes and tells you which of the five sections to start with.
