How about:
1. Set up a system to analyze users reported for being suspicious of being a bot in a chat.
2. Said system allows the person reporting to select one or more recent chat messages from said user.
3. The chat message gets added to a staff-curated list of spam messages. The curated list runs a check behind the scenes (maybe via a JS that already does the "sending message" mechanics).
4. If the message matches the list, the account is flagged for investigation and the message is shadow-banned (the message appears for the sender, but anybody else won't see the message because it wasn't actually sent)