I have put hundreds of canary strings into my conversation history of all the big providers, and so far the 'hit rate' is very low.

Really looks like they aren't yet training on conversation histories effectively.

Alternatively, they might train in a different manner than what you expect.

They might have determined that your canaries (whether it is a unique string, URL or similar) have a low signal-to-noise ratio and not worth following up.

Unless you provide more details, it's difficult to ascertain whether your conclusion is correct.

Can you explain this a little more?