So, yeah, apparently part of wrapping my head around nostr will have to include understanding bot behavior. Hence the cat pictures bot thinks I want to zap it. Nah.

My initial assumption was that these bots are using natural language processing for targeting. ChatGPT thinks “…sometimes yes, but usually it’s much dumber than that.”

Further, it says:

When a bot follows you, probability-wise:

~70–80%: dumb follow-spray

~15–25%: keyword trigger

~5–10%: graph-based

<1%: real NLP targeting

So the m.o. for follow-spray purportedly is

Subscribe to recent kind:1 notes on 5–20 popular relays.

Extract pubkeys from:

• Anyone who posts

• Anyone who replies

• Auto-follow them

And the m.o. for keyword-trigger purportedly is

Subscribe to kind:1 notes with keyword filters

• Simple string matching or regex

• Follow authors whose posts contain keywords