What AI can and can't do

You've reread his "ok." five times and still can't tell if he's upset or just driving. A fresh pair of eyes genuinely helps here. But those eyes have limits.

  • It can: notice that you apologize in every third message while he answers questions a day later.
  • It can: put a vague "something's off" feeling into words, like the fact that the conversation is always about his stuff.
  • It can: suggest a few possible replies for different goals.
  • It can't: diagnose your relationship.
  • It can't: tell whether someone is lying. Text isn't a lie detector.
  • It can't: replace a therapist. If the messages show control, threats or humiliation, you need a professional, not a prompt.

Step 1. Pick an excerpt and anonymize it

Don't upload six months of messages. Take 20–30 lines around the moment that got to you: a fight, a weird reply, a sudden silence. That way the AI doesn't drown in noise, and you don't hand your whole personal life over to a service.

  1. Replace names with "Me" and "Him" or "Her."
  2. Remove anything that could identify the person: addresses, workplaces, company names, phone numbers, usernames.