Open dataset · CC BY 4.0

AI crawler access, segment by segment

Every domain counted below went through the same audit: read its robots.txt, resolve the rules for the 48 tracked AI agents, then send live requests as four of them and record what came back. No segment gets a different test and nothing is self-reported, so a rate in one table can be set against a rate in another without adjustment.

A segment only appears once enough of its members have been measured for the rate to hold still. Below roughly 40 domains a single robots.txt edit moves the number by tens of points, which is noise wearing the clothes of a finding. 53 segments across 3 groupings currently clear that bar, drawn from 13,154 measured domains.

Top-level domain 34 segments, 11,810 domains

Grouped by the registry a domain sits under. This is the closest thing the corpus has to a geographic view, and it separates the widest: block rates differ several-fold between country-code registries, a larger spread than any other cut of the same data produces.

A top-level domain is a weak proxy for jurisdiction. Most of them are open to any registrant anywhere, so a domain under a country-code registry may be operated from outside that country by a company no local law reaches. Read these rows as differences between registration communities and the publishing cultures inside them, not as differences between legal regimes.

Segment Sites Mean score Blocks an answer engine Blocks a training crawler Refuses AI at the edge Publishes llms.txt
.com 6,762 71.6 8% 531/6,762 17% 1,141/6,762 21% 1,397/6,762 17% 1,120/6,762
.org 748 69.6 5% 39/748 15% 112/748 23% 175/748 8% 58/748
.net 593 68.1 6% 38/593 17% 99/593 21% 122/593 10% 58/593
.ru 468 66.8 3% 15/468 8% 36/468 9% 42/468 10% 46/468
.de 321 74.0 17% 55/321 29% 92/321 15% 48/321 7% 22/321
.io 308 72.7 2% 7/308 7% 21/308 8% 24/308 25% 78/308
.jp 236 68.1 7% 17/236 14% 33/236 9% 22/236 5% 12/236
.edu 216 74.3 2% 5/216 6% 12/216 9% 20/216 4% 9/216
.br 174 72.9 5% 9/174 13% 22/174 21% 37/174 12% 21/174
.uk 158 71.6 16% 25/158 28% 45/158 28% 44/158 7% 11/158
.in 151 64.5 2% 3/151 9% 14/151 18% 27/151 11% 16/151
.gov 150 68.8 3% 4/150 6% 9/150 23% 35/150 0.7% 1/150
.fr 142 72.4 25% 36/142 34% 48/142 15% 22/142 1% 2/142
.pl 118 72.4 10% 12/118 28% 33/118 14% 16/118 7% 8/118
.it 111 72.7 18% 20/111 28% 31/111 23% 26/111 4% 4/111
.tv 102 67.0 2% 2/102 21% 21/102 28% 29/102 4% 4/102
.cn 92 59.1 0% 0/92 0% 0/92 2% 2/92 2% 2/92
.co 92 72.9 4% 4/92 15% 14/92 22% 20/92 16% 15/92
.ai 86 74.0 1% 1/86 12% 10/86 15% 13/86 43% 37/86
.ca 80 72.7 9% 7/80 14% 11/80 26% 21/80 6% 5/80
.nl 71 72.1 10% 7/71 18% 13/71 15% 11/71 8% 6/71
.es 63 70.7 19% 12/63 32% 20/63 16% 10/63 13% 8/63
.me 60 70.7 3% 2/60 8% 5/60 18% 11/60 12% 7/60
.cz 55 69.6 9% 5/55 36% 20/55 18% 10/55 11% 6/55
.tr 54 67.4 13% 7/54 19% 10/54 4% 2/54 24% 13/54
.au 49 74.5 14% 7/49 18% 9/49 16% 8/49 10% 5/49
.ch 49 68.8 12% 6/49 27% 13/49 20% 10/49 6% 3/49
.app 46 72.7 0% 0/46 11% 5/46 11% 5/46 22% 10/46
.vn 46 73.2 4% 2/46 11% 5/46 11% 5/46 22% 10/46
.gr 45 77.8 11% 5/45 18% 8/45 22% 10/45 9% 4/45
.id 43 63.8 2% 1/43 7% 3/43 14% 6/43 5% 2/43
.eu 41 71.6 2% 1/41 7% 3/41 27% 11/41 10% 4/41
.ir 40 72.7 0% 0/40 0% 0/40 5% 2/40 10% 4/40
.se 40 70.2 18% 7/40 20% 8/40 18% 7/40 10% 4/40

Every percentage in this table is a share of that segment's own measured domains, shown beside it as a count over a denominator, never a share of the 13,154-domain corpus. Rows are ordered by size. A segment marked low n has fewer than 40 measured domains, which is too few for its rate to survive a single site's edit.

bar chart.fr (n=142)25%.es (n=63)19%.it (n=111)18%.se (n=40)18%.de (n=321)17%.uk (n=158)16%.au (n=49)14%.tr (n=54)13%.ch (n=49)12%.gr (n=45)11%.pl (n=118)10%.nl (n=71)10%

Share of each registry's measured domains whose robots.txt disallows at least one answer engine at the site root. The count beside each label is that registry's denominator; registries with fewer than 40 measured domains are left out of this chart. Click a bar for the segment.

Publishing platform 9 segments, 2,752 domains

Grouped by the system that produced the page, read from the generator meta tag it left in the HTML. A platform's shipped defaults decide a site's crawler posture more often than the site owner does, so these rows measure defaults as much as intent.

Platform detection relies on a generator meta tag, so coverage is partial and skewed. A site that strips the tag, or a system that never writes one, is invisible here. These rows therefore over-represent platforms that advertise themselves and say nothing at all about the ones that do not.

Segment Sites Mean score Blocks an answer engine Blocks a training crawler Refuses AI at the edge Publishes llms.txt
next.js 832 72.4 9% 78/832 16% 131/832 18% 148/832 16% 131/832
wordpress 828 80.0 7% 60/828 20% 164/828 29% 243/828 16% 129/828
drupal 377 73.6 4% 16/377 11% 41/377 20% 77/377 6% 21/377
client-rendered shell 267 52.1 3% 8/267 13% 35/267 12% 32/267 5% 14/267
nuxt 224 70.7 8% 17/224 17% 39/224 17% 39/224 19% 43/224
gatsby 76 74.4 1% 1/76 4% 3/76 9% 7/76 25% 19/76
astro 54 77.8 2% 1/54 4% 2/54 7% 4/54 31% 17/54
hugo 54 71.9 2% 1/54 6% 3/54 4% 2/54 9% 5/54
webflow 40 77.9 18% 7/40 10% 4/40 15% 6/40 28% 11/40

Every percentage in this table is a share of that segment's own measured domains, shown beside it as a count over a denominator, never a share of the 13,154-domain corpus. Rows are ordered by size. A segment marked low n has fewer than 40 measured domains, which is too few for its rate to survive a single site's edit.

Edge provider 10 segments, 8,827 domains

Grouped by the CDN or WAF in front of the origin, read from identifying response headers. This is where robots.txt and reality part company: a managed bot rule set can refuse an AI agent at the edge while the file at /robots.txt still says yes.

Only providers that expose an identifying response header can be detected. Sites behind an edge that stays quiet, or behind an origin that strips those headers, are absent from the count, so this sample is not representative of any provider's customer base.

Segment Sites Mean score Blocks an answer engine Blocks a training crawler Refuses AI at the edge Publishes llms.txt
Cloudflare 7,102 70.4 9% 640/7,102 18% 1,255/7,102 20% 1,407/7,102 13% 955/7,102
CloudFront 806 71.6 9% 70/806 14% 116/806 13% 108/806 14% 109/806
Fastly 320 73.6 13% 43/320 22% 70/320 15% 49/320 11% 36/320
Akamai 164 71.5 9% 15/164 14% 23/164 23% 38/164 12% 20/164
Vercel 141 76.1 4% 6/141 8% 11/141 12% 17/141 32% 45/141
LiteSpeed 96 78.6 0% 0/96 26% 25/96 41% 39/96 19% 18/96
Imperva 70 62.1 3% 2/70 4% 3/70 6% 4/70 4% 3/70
Azure Front Door 56 69.8 2% 1/56 11% 6/56 7% 4/56 7% 4/56
Netlify 45 77.7 2% 1/45 9% 4/45 4% 2/45 27% 12/45
Kinsta low n 27 87.5 4% 1/27 7% 2/27 19% 5/27 52% 14/27

Every percentage in this table is a share of that segment's own measured domains, shown beside it as a count over a denominator, never a share of the 13,154-domain corpus. Rows are ordered by size. A segment marked low n has fewer than 40 measured domains, which is too few for its rate to survive a single site's edit.

Measure a domain

A scan adds the domain to the corpus, and to whichever segments it belongs to. The full breakdown is in the open dataset.