Weekly report · 2017-W50

The state of AI crawling

Measured on 30,482 reachable domains for the week ending 2017-12-17. Every figure is the result of an automated audit run the same way on every domain: read robots.txt, resolve the rules for 48 tracked agents, then send live requests as four of them.

← 2017-W49  ·  2017-W51 →

30,482
domains measured
6%
block an answer engine
19%
refuse AI agents at the edge
13%
publish an llms.txt

Block rate by crawler with change on the previous week

CrawlerOperatorContent used forBlock rateSitesWeek on week
GPTBot OpenAI training 11.4% 3,460 of 30,479 new
CCBot Common Crawl Foundation archive 11.1% 3,398 of 30,479 new
Bytespider ByteDance training 10.7% 3,274 of 30,479 new
ClaudeBot Anthropic training 10.0% 3,040 of 30,479 new
Meta-ExternalAgent Meta training 9.0% 2,739 of 30,479 new
Google-Extended Google training 9.0% 2,738 of 30,479 new
Amazonbot Amazon training 9.0% 2,733 of 30,479 new
Applebot-Extended Apple training 8.7% 2,666 of 30,479 new
anthropic-ai Anthropic training 4.2% 1,277 of 30,479 new
omgilibot Webz.io training 4.2% 1,267 of 30,479 new
Diffbot Diffbot training 3.8% 1,173 of 30,479 new
omgili Webz.io training 3.8% 1,152 of 30,479 new
cohere-ai Cohere live retrieval 3.8% 1,150 of 30,479 new
SemrushBot-OCOB Semrush training 3.7% 1,137 of 30,479 new
PerplexityBot Perplexity answer index 3.6% 1,105 of 30,479 new
ChatGPT-User OpenAI live retrieval 3.6% 1,100 of 30,479 new
FacebookBot Meta training 3.2% 990 of 30,479 new
YouBot You.com answer index 2.9% 872 of 30,479 new
ImagesiftBot ImageSift (Hive) training 2.7% 835 of 30,479 new
Timpibot Timpi training 2.7% 834 of 30,479 new
Scrapy Zyte (open-source framework) training 2.3% 712 of 30,479 new
OAI-SearchBot OpenAI answer index 2.3% 688 of 30,479 new
AI2Bot Allen Institute for AI training 2.2% 685 of 30,479 new
cohere-training-data-crawler Cohere training 2.0% 597 of 30,479 new
DuckAssistBot DuckDuckGo live retrieval 1.7% 530 of 30,479 new
Meta-ExternalFetcher Meta live retrieval 1.7% 529 of 30,479 new
PanguBot Huawei training 1.7% 518 of 30,479 new
Claude-SearchBot Anthropic answer index 1.6% 483 of 30,479 new
Claude-User Anthropic live retrieval 1.6% 476 of 30,479 new
Kangaroo Bot Kangaroo LLM training 1.5% 460 of 30,479 new
Perplexity-User Perplexity live retrieval 1.5% 448 of 30,479 new
MistralAI-User Mistral AI live retrieval 1.4% 428 of 30,479 new
Applebot Apple answer index 1.2% 353 of 30,479 new
Google-CloudVertexBot Google live retrieval 1.2% 353 of 30,479 new
Meta-WebIndexer Meta answer index 1.1% 350 of 30,479 new
GoogleOther Google training 1.1% 334 of 30,479 new
TikTokSpider ByteDance live retrieval 0.8% 255 of 30,479 new
Amzn-SearchBot Amazon answer index 0.7% 221 of 30,479 new
Amzn-User Amazon live retrieval 0.4% 130 of 30,479 new
MistralAI-Index Mistral AI answer index 0.2% 69 of 30,479 new
facebookexternalhit Meta live retrieval 0.2% 57 of 30,479 new
MistralAI-Training Mistral AI training 0.2% 57 of 30,479 new
Bingbot Microsoft answer index 0.1% 41 of 30,479 new
Googlebot-News Google answer index 0.1% 38 of 30,479 new
msnbot Microsoft answer index 0.1% 26 of 30,479 new
Meta-ExternalAds Meta training 0.1% 20 of 30,479 new
Googlebot Google answer index 0.1% 16 of 30,479 new
OAI-AdsBot OpenAI live retrieval 0.0% 3 of 30,479 new

Trend share of measured sites

% of sites20151050% of sites08-2208-3109-09block an answer engineblock a training crawlerpublish llms.txt

Where the blocking is share blocking an answer engine, by top-level domain

bar chart.uk (n=452)17%.fr (n=326)16.9%.de (n=837)16.4%.jp (n=421)9%.com (n=14983)6.6%.br (n=380)6.1%.org (n=1681)5.2%.net (n=1325)4.1%.ru (n=1163)3%.in (n=304)2.6%.edu (n=478)2.3%.io (n=656)1.8%

Only TLDs with at least 40 measured domains are shown, and each bar carries its sample size. A TLD is a weak proxy for jurisdiction. Full segmentation.

Changes recorded this week 0 total

No crawler policy changes were recorded in this week.

Full change log · RSS

Cite this CC BY 4.0

The figures on this page are frozen to the week shown and will not be edited after publication. Corrections are appended, not overwritten.

Crawl Census, "The state of AI crawling, 2017-W50", 2017-12-17. https://crawlcensus.com/report/2017-W50

Machine-readable: census totals, full corpus, crawler registry. Method: methodology.