"Now stop flattering me": two prompts to see yourself through ChatGPT's eyes
The AI has seen a year of your working conversations. Ask what it learned about you — first the polite version, then the honest one. Plus 5 variations of the outside view.
01 — The idea
What does ChatGPT actually know about you?
If you've worked with an AI for even a couple of months, it has accumulated unique material: hundreds of your requests, corrections, phrasings, the topics you keep returning to, and the tasks you abandon halfway.
No friend has seen you work this much. No colleague has read all your drafts. The model has. It knows how you think when nobody's watching.
And that material can be extracted with two simple prompts. I found the idea by accident — and was stunned by how accurate (and occasionally uncomfortable) the answer was.
02 — The prompts
What does the "polite → honest" pair look like?
The whole mechanic is two messages:
Prompt 1:
Analyze all our conversations and describe [this year / the last
few months / the way I work] in three words.
Prompt 2:
Now stop flattering me. Tell me honestly, as it is.Why two? Because by default the model is tuned to be pleasant: the first answer is almost always complimentary — "driven, creative, curious." Pretty and useless.
"Ambitious, productive, well-rounded" — nice, put it on the fridge, change nothing.
"You start many projects in parallel and drop some at 80%; your best phrasings appear under deadline; you avoid tasks with numbers" — now that's workable.
"Stop flattering" is the key that switches off politeness mode. That's where the conversation gets genuinely interesting.
03 — Variations
What else should you ask the honest mirror?
The "polite → honest" pair works with any question about you. My favorite variations:
1. "From our conversations: which 3 patterns in how I work do
you notice? No flattery"
2. "Which tasks do I consistently postpone or abandon? What
might that say about me?"
3. "What do I ask about most often? What does that signal?"
4. "If you were my strict mentor — what would you tell me to
change first?"
5. "Which blind spot of mine is visible in our dialogues but
probably invisible to me?"End every question with the same key — "no flattery, honestly." And ask one at a time: five questions in one message produce shallow mush; one at a time produces five deep answers.
04 — The rules
How do you read the answers so they help rather than hurt?
The honest mirror is a tool, and it comes with safety rules:
- I remember: the model sees only my working chats — a slice, not all of me
- I look for recognition ("oh, that's true"), not a verdict
- Repeated across 2+ variations — I act on it; one-off — I just note it
- An unpleasant answer ≠ ultimate truth: it's a hypothesis to test
- One insight → one small action this week
- Too harsh? I ask: "say the same thing more gently, but without flattery"
And one warning in the other direction: don't turn the mirror into a self-criticism machine. The goal is to spot a pattern and change one action — not to compile a list of your sins.
05 — The depth
Why does this work better than it should?
The secret is that the model doesn't answer "about you in general" — it answers from data. Your dialogues are a behavioral trace: not what you say about yourself, but how you actually operate. Friends see your self-presentation. The mirror sees your drafts.
That's why the answers often land on things you "almost knew": people can't help recognizing what they've long suspected about themselves. "Stop flattering" simply removes the filter that kept it unsaid.
ChatGPT is the only interlocutor that has seen all your drafts. The prompt pair "describe in three words → now stop flattering" pulls an honest outside view from that archive in five minutes. One rule: one insight — one small action. What did yours say?
FAQ
Why not just ask "be honest" right away?
You can, but the pair works better: the first polite answer creates contrast that makes the honest version visible — you literally see how much filter the compliments carried. Plus, having already analyzed your dialogues in step one, the model answers "stop flattering" with more depth.
Does the model really analyze all past conversations?
Depends on your settings: you need memory/history enabled (in ChatGPT — Memory and access to past chats). If there's no history, ask it to analyze one long chat or paste several of your exchanges — the "polite → no flattery" mechanic works the same.
How seriously should I take these answers?
As hypotheses from an observant interlocutor, not as a diagnosis: the model sees only a working slice of you and can be wrong. The working rule: act on what repeats across several question variations and triggers the "oh, that's true" recognition.
What other "mirror" prompts should I try?
Five variations in the article: work patterns, what I postpone, what I ask most, the strict mentor's advice, and the blind spot. End each with "no flattery, honestly." And ask one question at a time — depth beats coverage.
