The state of the walls

Headline numbers from the latest crawl of 1020 tracked sites (644 with readable policy). Snapshot: 2026-10-04

32%
block ≥1 AI crawler
53
run default-deny robots.txt (unnamed crawlers get nothing)
11.4%
publish llms.txt
Bytespider
most-blocked AI bot (25%)

Block rate by bot

Bytespider 25% CCBot 24.7% ClaudeBot 22% GPTBot 21.3% meta-externalagent 20.8% Diffbot 20.7% cohere-ai 20% omgili 20% Applebot-Extended 19.1% Google-Extended 18.8% PerplexityBot 18.5% Amazonbot 18.2% anthropic-ai 18.2% Claude-Web 18.2% YouBot 17.9% ImagesiftBot 17.9% omgilibot 17.7% Timpibot 17.5% Perplexity-User 16% Claude-User 15.5% cohere-training-data-crawler 15.4% AI2Bot 15.4% DuckAssistBot 14.9% MistralAI-User 14.9% PanguBot 14.8% ChatGPT-User 14.6% Claude-SearchBot 14.6% Meta-ExternalFetcher 14.6% DeepSeekBot 12.4% OAI-SearchBot 12% Gemini-Deep-Research 11.2% GrokBot 10.4%

Block rate by site category

CategorySitesBlocking ≥1 botRate
news333193.9%
food6583.3%
jobs4375%
social151173.3%
entertainment12866.7%
realestate3266.7%
sports9666.7%
health5360%
reference15960%
auto5240%
finance17529.4%
education7228.6%
ecommerce18527.8%
travel12325%
top1k44110724.3%
tech34411.8%
classifieds100%
government700%

Blocklists nobody wrote

Not every block is a decision. Some arrive as infrastructure: a CDN inserts a managed section, or a site copies a community roster of AI user agents. Either can be switched on or off in one move by someone who never reads the names inside — on 2026-09-01 roblox.com un-blocked eight AI crawlers at once by deleting a Cloudflare managed block, and on 2026-09-02 semafor.com un-blocked 31 by deleting a copied list. Each cohort is small enough to name in full, so we do, and we re-count them from the archived files every day.

BoilerplateSitesWhich onesWhat it is
Cloudflare Managed Content3gamespot.com (last read 2026-09-15) · nexusmods.com (last read 2026-09-01) · snopes.comA block Cloudflare generates and maintains, dropped into the file between marker comments. Removing it is a dashboard toggle, not an edit.
Content Signals Policy24amplitude.com · avast.com (last read 2026-09-24) · bitdefender.net · cloudflare.com · corriere.it · forter.com · gamespot.com (last read 2026-09-15) · hostinger.com · intercom.io · launchpad.net · linktr.ee · netlify.com · nexusmods.com (last read 2026-09-01) · nvidia.com · oxylabs.io · patreon.com · segment.io · sentry.io · snopes.com · theatlantic.com · trendmicro.com · vercel.com · visma.com · weibo.comCloudflare's machine-readable usage preamble (search / ai-input / ai-train). It carries no Disallow of its own — it states intent alongside the rules.
robotstxt.com/ai list1launchpad.netA community-maintained roster of AI user agents, copied in whole and credited in a comment. It grows when that list grows, without the site revisiting the question.

Counted by matching the archived robots.txt of every readable domain against the marker each cohort leaves behind, on 2026-10-04. A site can appear in more than one row. Membership is a fact about the file, not a judgement about the site: the Content Signals preamble states a preference and blocks nothing by itself. A member we could not read on this crawl is counted from its last archived file, dated beside its name.

Citing these numbers

Data is CC BY 4.0. Cite as "Canicrawl, AI crawler access census, 2026-10-04" with a link. Raw data: latest.json. Methodology: about. Every figure is reproducible from the committed daily snapshots.