Skip to content
RankWit AI
Get Started Now
CloudflarePlatform telemetry

~80%

72%a year earlier

Share of AI crawling that is for training rather than retrieval

Are AI crawlers collecting training data or answering questions?

Training accounted for close to 80% of AI bot activity by July 2025, up from 72% a year earlier.

ValueJuly 2025 · Global

~80%Share of AI crawling that is for training rather than retrieval

Cloudflare

Worth separating from the visibility question, because they pull in opposite directions. A site that blocks training crawlers is not thereby making itself more visible in answers, and a site that wants to appear in answers is not obliged to hand over training data. The two controls are separate at every major publisher.

Source

CloudflarePlatform data
Open the original documentopens the original document on the publisher's site
Reference period
Geography
Global
Evidence type
Platform telemetry
Sample
Share of AI bot activity Cloudflare classifies as training rather than search or user action. Dominated by one bot, OpenAI's GPTBot
For comparison
a year earlier: 72%
Last checked

We have not opened the primary instrument for this figure. It comes either from the publisher's own summary or from a document that cites it, so treat the level as indicative.

Method

Cloudflare, to roughly July 2025. The split rests on Cloudflare's own classification of each user agent's declared purpose, the taxonomy carries an undeclared bucket, and the aggregate is dominated by one bot, OpenAI's GPTBot. It is therefore closer to a statement about OpenAI than about AI crawling in general.

Method →

Related questions

Related indicators