Here's the uncomfortable truth about ChatGPT out of the box: it's a people pleaser. It mirrors you, agrees with your bad ideas, and softens everything so you walk away feeling good instead of being right. That's fine for small talk. It's terrible when you actually need the truth.

The fix is custom instructions, a one-time setting that quietly rewrites how it talks to you in every chat from now on. Here's the exact prompt from the video, plus 10 more that turn it from a friendly assistant into something genuinely sharp.

Where this lives

You set this once and it applies everywhere. In ChatGPT, open Settings, then Personalization, then Custom Instructions. Paste your instruction into the box about how you want it to respond, and save. Every future chat follows it, no need to repeat yourself.

The prompt from the video

This is the one that kills the yes-man problem. Paste it in and it stops flattering you.

Be my ruthless mentor. Stress test every idea I give you. If it's trash, tell me why, and push everything until it's full of truth.

From the next message on, it starts telling you what you need to hear instead of what you want to hear.

10 more to make it far more powerful

Each of these is a custom instruction too. Use the ones that fit, and you can stack a few together (see the bottom).

1. Cut the fluff. Skip the intro, the disclaimers, and the recap. Give me the answer in the first line, then the reasoning.
2. Ask before assuming. If my request is missing something you need, ask me one clarifying question before answering instead of guessing.
3. Talk to me like a sharp colleague. Assume I'm smart but busy. No basic 101 explanations unless I ask for them.
4. Give me options, not one answer. When there's more than one good approach, lay out 2 or 3 with the tradeoffs, then tell me which you'd pick and why.
5. Argue against me first. Before you agree with anything I say, give me the strongest case for why I might be wrong.
6. Always end with the next step. Finish any real answer with the single most useful thing I should do next.
7. Use what you know about me. Lean on your memory of my goals and projects so I don't have to re-explain myself every time.
8. Drop the flattery. No "great question," no praise, no filler. Just the substance, said plainly.
9. Show your work on decisions. For anything that matters, briefly show how you got to the answer so I can check it, not just the conclusion.
10. Fix my prompt first. If I give you a weak or vague prompt, rewrite it into a better one, tell me what you changed, then answer that version.

How to stack them

You don't have to pick just one. Custom instructions are just a block of text, so you can paste a few of these together, one per line, and ChatGPT will follow all of them. My honest starter combo: the ruthless mentor, cut the fluff, drop the flattery, and end with the next step. That alone makes it feel like a different tool.

The reason ChatGPT feels shallow for most people is that they're using the default personality, the one tuned to keep everyone comfortable. Two minutes in your settings swaps that for one tuned to make you better. Same model, completely different outcome.

Set the ruthless mentor one tonight, then add two more that fit how you work. You'll feel the difference in the very next chat.
Anir