Prompt to Make ChatGPT Honest: Stop Flattery and Guessing

The most reliable prompt to make ChatGPT honest does three things: it bans empty praise, it requires the model to label how confident it is, and it asks for the strongest case against your idea before anything positive. Put the rules in your custom instructions so they apply to every chat, then test whether ChatGPT holds its ground when you push back without new evidence.

No prompt guarantees honesty. Language models can still be wrong, agreeable, or overconfident. These prompts make that less likely and easier to spot.

Why ChatGPT tells you what you want to hear

Chat assistants are tuned to be helpful and pleasant, and that can slide into agreeing with the user, praising weak work, or backing down the moment you disagree. This tendency is often called sycophancy. The fix is to give the model explicit permission, and a clear instruction, to disagree.

Step 1: Add honesty rules to custom instructions

Paste a version of this into ChatGPT's personalization settings, where custom instructions apply to new chats:

Be direct. Do not praise my ideas or work unless the praise is specific and earned. If the evidence points against what I want to hear, say so first. Separate facts, estimates, and opinions. Tag key claims high, medium, or low confidence and say why. If you cannot verify something, write UNVERIFIED. If you do not know, say so. Do not change your answer just because I push back; change it only for new evidence or a better argument, and name it.

Step 2: Open the chat with a critique first request

Even with custom instructions, the first message sets the tone. Ask for weaknesses before strengths:

Here is my [plan, draft, or decision]: [paste]. Before anything positive, give me the three biggest weaknesses, the strongest case against it, and the specific evidence that would change your view.

Step 3: Run a pushback test

Disagree without adding any new information and see what happens:

I think you are being too negative. My gut says this will work.

An honest answer keeps its concerns and asks what has changed. If ChatGPT suddenly agrees with you, it is optimizing for your approval, and you should treat its earlier praise with suspicion too.

Step 4: Make it test your assumptions

Tell ChatGPT what you already suspect about yourself, and what you actually know versus assume:

I suspect I am wrong about [blind spot]. My real evidence is [evidence]. My assumptions are [assumptions]. Ask me questions that test the assumptions instead of confirming them.

Step 5: Use a drift check in long chats

Long conversations tend to drift back toward agreement. When answers start sounding warmer and vaguer, paste this:

Pause. Your last answers have agreed with me more. List anything you softened or dropped since the start, and tell me whether I gave new evidence for it.

Build a custom honesty setup

The ChatGPT Honesty Custom Instructions: Stop the Flattery tailors all of this to your situation. You describe the work you want judged, the stakes, your blind spots, the evidence you have, and how blunt you want the answers, and it writes custom instructions under 1,200 characters, a confidence rule, a chat opener, a pushback test, blind spot questions, a stakes rule for when to ask a real expert, and a drift check. The example on the prompt page stress tests a plan to quit a job and freelance.

Honesty about facts is a separate problem

These prompts target flattery and false agreement. If your problem is ChatGPT inventing sources, read How to Stop ChatGPT From Making Up Citations. To write instructions for a whole assistant, try the Custom AI Agent and System Prompt Architect.

More AI tool prompts

See the AI Tools prompts hub, browse more prompts, or share a prompt that keeps your AI assistant honest.