Our data
What we have measured so far
Every number on this page comes from our own measurements — the same fifteen checks we run for you, taken with the same engine and the same published standard. We publish the aggregate, never a list of which sites did badly: whether a single report is public is the owner's decision, not ours.
The pattern in one paragraph
AI systems can almost always reach the sites we measure, but they struggle to understand and cite them. Crawler access (group A) fails on 20% of checks, while comprehension (B) fails on 66% and citation evidence (C) on 54%. The average score is 8.0 of 15 — about 53% of the checks pass.
The sample behind this page
- 122 domains measured against the fifteen checks
- 210 individual checks graded, each with the evidence we saw
- 8.0 / 15 average score (53% of checks pass)
- 8 rechecks run after fixes, each keeping its before/after evidence
Which checks fail most often
Ranked by how often each check failed across every report we have graded. A high number here is a market signal, not a criticism of any one site: it is what almost everyone gets wrong.
| Check | What it measures | Failed | Rate |
|---|---|---|---|
| C3 | There is original data or first-hand information | 13 / 14 | 93% |
| B2 | Structured data (JSON-LD) is present | 12 / 14 | 86% |
| C4 | There is a visible last-updated date | 12 / 14 | 86% |
| B4 | An llms.txt is present | 11 / 14 | 79% |
| B1 | Title and description are complete | 9 / 14 | 64% |
| B3 | Heading hierarchy is semantic | 7 / 14 | 50% |
| B5 | The operator behind the site is verifiable | 7 / 14 | 50% |
| C5 | There is third-party reference evidence | 6 / 14 | 43% |
| A3 | A sitemap exists | 5 / 14 | 36% |
| C1 | The conclusion comes first | 5 / 14 | 36% |
| A2 | Content is in the HTML, not only after JavaScript | 3 / 14 | 21% |
| A4 | No dead-link pollution | 3 / 14 | 21% |
How the same sites are graded
- 8 Invisible
- 4 Occasionally cited
- 1 Preferred source
- 1 Consistently cited
Grades are decided by the standard, not by us: one failed group A check caps a site at “Invisible” however well everything else scores. See the rules.
What this means if you run a site
The two most common failures are the two that are cheapest to fix and hardest for an AI to guess: connected JSON-LD (so a machine knows what the page is about) and first-hand numbers (so there is something worth quoting). Both are content work, not infrastructure work — which is why they are still broken almost everywhere while crawler access is mostly fine.
Start with the free check to see which of the fifteen your own site passes, then read the repair guide for the order that moves the score.
Method and limits
- Where the numbers come from. Our own engine at the published method, measuring public pages only — the same rules for every site, including our own.
- Sample size. 14 report(s) is a small sample. We would rather show a small honest number than a large estimated one; this page is updated as we measure more.
- What we do not publish. No per-domain results. A report becomes public only when its owner chooses to list it; the aggregate above cannot be traced back to a site.
- How to verify us. Run the free check on your own domain and compare it with the leaderboard — the same fifteen checks, the same evidence format.