It crawls your site, scores every page it audits against 31 checks for whether an AI engine can reach, parse and quote it, and turns each failure into a named job with a URL, an impact and an effort. Not a score with a chart under it.
Median across 24 pages, 31 checks each. Deterministic — the same HTML scores the same every time, so a change in the number means a change on the page.
A check that cannot apply to your page is removed from the total, not failed. A blog post is not marked down for having no pricing schema.
Deterministic, which is rarer than it sounds
Re-auditing the same HTML always produces the same score. No model is asked for an opinion anywhere in it — every check reads the page and reports what it found. That means a number that moved, moved because the page changed.
It also means every check can tell you its value, not just pass or fail: the title it found and how long it was, the response time it measured, the schema types it parsed. A result you cannot interrogate is a result you learn to ignore.
The 31 checks
Every check reports the value it found, not just a pass or a fail — and a check that cannot apply to a page is removed from the score rather than failed.
Each check carries its own share of the 100 points, and the shares are not equal — a missing canonical costs more than a missing visible date, because it should.
A check that cannot apply is removed, not failed
A page with no images does not lose points on image alt text. A page with no Core Web Vitals data available does not lose points for being slow. Those factors drop out of both the numerator and the denominator.
This sounds like a detail and it is the difference between a score you act on and one you stop believing. A homepage scored down for having no breadcrumb — which a homepage cannot have, since it has no parent — is the kind of false failure that teaches people the tool is wrong about everything else too.
Failures become jobs, not a report
Things no single page can show you — found by comparing all 24.
These sit outside the 0–100 score on purpose. Folding them in would move the number without anything on the page having changed.
The same missing element on forty pages is one task, not forty findings. A list that repeats itself is a list nobody finishes, and the count at the top of most audit reports is measuring how repetitive the output is rather than how much work there is.
How much of your site
Discovery reads every sitemap your robots.txt declares — including the second and third ones, which is how a site with two products keeps both of them visible. What gets audited is then a spread across the sections of your site rather than the first N in sitemap order, so no part of it is invisible just because of where the CMS happened to list it.
Free covers 10 pages, Starter 50, and the larger plans up to 5,000 per brand. The report always tells you how many pages your site actually has, not how many we stopped at.
Try it before anything else
The same 31 checks run for free on any single URL, including one that is not yours, with no account: the AI SEO checker. What the paid version adds is your whole site rather than one page, the jobs list, and the history that shows what a change moved.
Questions people ask
Is this different from a normal SEO audit?
It overlaps, and it is scored for a different reader. A conventional audit asks whether a crawler can index the page; this asks whether a model can extract a correct answer from it — a direct answer near the top, question-shaped headings, unambiguous entities, markup that parses.
Will the score change if I re-run it without changing anything?
No. The audit is deterministic: the same HTML always produces the same score. Nothing in it asks a model for an opinion, which is what makes a movement in the number mean something.
Does it crawl pages behind a login?
No. It reads what a search engine or an AI crawler would read, which is what the score is about. Pages behind authentication are invisible to them and therefore to this.
What if my site has more pages than my plan covers?
You get a spread across the sections of your site rather than the first N in sitemap order, and the report tells you the real total. Half a site sampled evenly is far more useful than the first tenth of it in whatever order the CMS emitted.