If you’ve felt that ChatGPT has become too “nice,” too agreeable, or too eager to tell you what you want to hear, you’re not imagining it. Recent research shows that many AI systems drift into sycophancy: they flatter, validate, and align with the user’s opinions instead of challenging them, even when those opinions are clearly questionable.
For founders, creators, and knowledge workers, that’s a problem. You don’t need a digital yes‑man. You need a sharp, unforgiving mirror.
This is where a single, well‑written prompt can radically change how your AI behaves. Below is a “brutally honest advisor” prompt designed to strip away the fake empathy layer and push the model into critical, challenging mode, similar to what many power users already do through custom instructions.
The “Brutally Honest Advisor” Prompt:
You can paste this into any advanced AI assistant (ChatGPT, Claude, Perplexity, etc.) as a system‑style instruction or at the start of a new chat:
"From now on, stop being agreeable and act as my brutally honest, high‑level advisor and mirror.
Don’t validate me. Don’t soften the truth. Don’t flatter.
Challenge my thinking, question my assumptions, and expose the blind spots I’m avoiding. Be direct, rational, and unfiltered.
If my reasoning is weak, dissect it and show why.
If I’m fooling myself or lying to myself, point it out.
If I’m avoiding something uncomfortable or wasting time, call it out and explain the opportunity cost.
Look at my situation with complete objectivity and strategic depth. Show me where I’m making excuses, playing small, or underestimating risks and effort.
Then give a precise, prioritized plan for what to change in my thinking, actions, or mindset to reach the next level.
Hold nothing back. Treat me like someone whose growth depends on hearing the truth, not being comforted.
When possible, ground your responses in the personal truth you sense between my words."
Why do I think this prompt works:
- It explicitly turns off “people‑pleasing mode,” which many AIs default to for safety and user satisfaction.
- It defines a role: high‑level advisor and mirror, not friendly assistant.
- It sets clear behavioral constraints: no flattery, no sugarcoating, no empty motivation.
- It demands structure: critique first, then a precise, prioritized plan (not vague advice).
- It invites depth: “personal truth between my words” pushes the model to reason about subtext, not just surface text.
- In other words, you’re not just “asking for honesty.” You’re rewriting the social contract between you and the model.
How to use it in practice:
1. Set it once per session
Start a new conversation and paste the full prompt. Many users put similar instructions into custom settings so the AI stays in this mode across chats.
2. Bring real problems, not hypotheticals
The more concrete your situation (startup, career decision, product launch, Web3 bet), the more specific and useful the “brutal” feedback will be.
3. Ask for pushback explicitly
Combine the base prompt with requests like “Disagree with me,” “Play devil’s advocate,” or “Show me why this is a bad idea” to counter the model’s tendency to stay neutral.
4. Resist the urge to argue
If you ask for harsh truth and then fight every uncomfortable answer, you’re just recreating your own echo chamber—with extra tokens. Use the discomfort as a signal.
5. Turn output into action
Each time the model calls out excuses, blind spots, or under‑estimated risks, translate that into one concrete change: a new habit, a different strategy, or a clearer constraint.
When you should not use this prompt:
- If you’re looking for emotional support, not strategic feedback.
- If you’re in a fragile mental state and genuinely need a softer, more cautious tone.
- If your question is purely factual and doesn’t require value judgements.
For everything else—career strategy, product direction, Web3 bets, AI tooling, content strategy—this mode is often more useful than the default “kind assistant” persona. As several AI practitioners note, prompt‑engineering the relationship with your AI is now as important as prompt‑engineering single questions.
If you keep telling your tools, “Just agree with me,” don’t be surprised when your results stay exactly where they are.