Research
What we see across 236 sites
Live from each site's latest on-demand scan between Jun 22, 2026 and Oct 22, 2026, refreshed every hour.
90
median score
47%
of sites are not yet Excellent
19%
of sites answer AI crawlers differently than browsers
What each grade means
Grades come from the 0-100 score. The share next to each grade is live, from the sites in the stats above.
- Excellent90 to 100
AI crawlers can reach, read and cite the page. Remaining findings are polish.
53% of sites - Good75 to 89
Readable by AI crawlers, with gaps that make the page less likely to be quoted.
36% of sites - Moderate50 to 74
AI crawlers reach the page but miss important content or signals.
11% of sites - Poor0 to 49
AI crawlers are blocked or find little they can use.
0% of sites
Most common gaps
Checks that most sites do not fully pass. A site counts when the check failed or warned on its latest scan.
| Check | Sites affected | Failed | Warned |
|---|---|---|---|
| Title, description & canonicalstructured.meta-basics | 56%131 of 236 | 54 | 77 |
| Machine-readable freshness signalsgeo.freshness | 53%125 of 236 | 0 | 125 |
| Author & E-E-A-T signalsstructured.author-eeat | 51%120 of 236 | 1 | 119 |
| Content is structured for AI extractiongeo.extractability | 50%106 of 211 | 0 | 106 |
| Strict-Transport-Security header is presenttrust.hsts | 49%116 of 236 | 0 | 116 |
| Main content extractabilityrendering.main-content | 47%112 of 236 | 37 | 75 |
| Images carry alt textrendering.image-alt | 46%92 of 198 | 46 | 46 |
| Server responds quickly (TTFB)crawler.ttfb | 44%105 of 236 | 19 | 86 |
| Page uses a clear heading hierarchygeo.headings | 37%88 of 236 | 24 | 64 |
| Page has enough content to be cited by AI answer enginesgeo.content-depth | 36%86 of 236 | 40 | 46 |
Methods
- Each site counts once, by its latest on-demand scan from the web form, the GitHub Action or the MCP endpoint.
- Monitoring re-scans are excluded, because they re-scan the same few monitored sites on a schedule.
- The score is the headline score shown on each report: Agent Readability, averaged with Smart Agent when that check ran.
- Informational results such as llms.txt are never counted as failures.
- Scans span several scanner versions. Until October 2026 the server probe sent only a GPTBot user agent; newer scans send eight AI user agents.
- No site, page or customer is identified. Only counts are published.
Limits
- The sample is self-selected: these are sites whose owners chose to test them with an AI readiness scanner, which skews toward technical teams.
- A scan covers the page that was submitted, usually the homepage, unless a deep scan was run.
- Statistics appear only once at least 30 sites are in the sample, and a check is listed only once at least 50 sites were tested for it.