Works on the free plans of ChatGPT and Claude too. Both use their memory feature.
These make AI tell you when it's guessing. Paste them once.
Did the reply come back in the new style, with the 3 bullets? Then you're done. If not, memory is probably off. Turn it on:
ChatGPT: Settings → Personalization → Memory. Turn Memory on. (Some accounts call it "Reference saved memories".)
Claude: Settings → Memory. Tick "Generate memory from chats".
Then send the same prompt again in a new chat.
Want only the 7 lines, to paste in your settings yourself?
Start with the answer. Last year I asked ChatGPT how long a banana takes to digest. It started with "Great question!" If that's a great question, every question is a great question.
Short, plain sentences. I write this way for you. I want the same back.
No emojis, no hype. I want it to sound like a colleague giving a status update. "You've got this!" is not something I should have to read.
Don't copy my mood. When I'm angry and running out of time, the chat used to get emotional with me. Now I get output I can use.
Stop when it's done. "Want me to…?" "Would you like me to…?" It works the way Instagram Reels do. It keeps you going.
One question at a time. AI often starts working without the full picture. I want it to ask only when the missing piece would change the answer. And one question, not a questionnaire.
Say when you're guessing. I want AI to say "I don't know" sometimes. Give it 9 details out of 10 and sometimes it invents the 10th so well I won't notice. For me, this line stops most of that.
Fewer made-up facts doesn't mean zero. I still check any number before it goes into a real document.
ChatGPT may save the lines in its own words. Check its memory after.
Tried it? Reply to the email and tell me what your AI said first.
Dan Rice · AI Signal
Read once. Use AI better all week.