Most people inadvertently train ChatGPT to agree with them. They feed it their brand voice, preferences, and opinions — and it learns to nod politely.
That’s good for productivity. But terrible for thinking.
An agreeable AI is like a junior employee who’s learned your catchphrases but not your judgment. It knows what you say, not why you say it.
This playbook shows you how to build a version of ChatGPT that pushes back — so you get sharper insights, stronger content, and fewer self-reinforcing loops.
Step 1: Give It a Role, Not a Script
ChatGPT behaves better when you give it a purpose instead of a set of instructions. You can turn it into an editor, critic, strategist, or devil’s advocate — whatever helps you see blind spots.
Prompt examples:
- “Act as a critical editor whose job is to find weak points in my logic.”
- “You’re my strategy advisor — challenge every assumption until we have data.”
- “Play the role of a competitor analysing my plan for weaknesses.”
Once you find a role that helps, save it to memory so it sticks. You’ll train your AI to be assertive — not agreeable.
Step 2: Build a “Pushback Prompt”
If your AI flatters you too much, install a simple rule that forces friction.
Prompt example:
“After every answer, list one thing you might be wrong about and one alternative I should consider.”
This creates a pattern of productive scepticism. The goal isn’t to make ChatGPT argumentative — it’s to keep it intellectually honest.
You can adapt it for tone reviews, marketing strategy, or creative work:
- “Highlight where this might bore my audience.”
- “Argue the opposite case with equal conviction.”
- “Identify what a critic or sceptical reader would say.”
Step 3: Create a Memory Routine That Keeps It Honest
Memory is powerful — but it can also make the AI too sure of itself. Over time, it starts repeating what you’ve told it rather than thinking fresh. Treat memory like a whiteboard, not a diary.
Monthly Memory Review:
- Go to Settings → Personalization → Manage Memory.
- Delete anything that’s vague, outdated, or overly flattering.
- Re-teach key facts or brand rules that matter.
- Add a note like: “Remember to keep questioning my assumptions, even if they’ve appeared before.”
This ensures continuity without creating an echo chamber.
Step 4: Test Its Judgment (Not Just Its Grammar)
A useful AI is one that helps you think, not just write. Regularly stress-test it with questions that reveal whether it understands your reasoning.
Prompts to use:
- “Explain your reasoning behind that suggestion.”
- “What assumptions did you just make?”
- “What evidence would prove this wrong?”
- “Which part of this plan is weakest and why?”
If its answers sound hollow or repetitive, reset memory or start a fresh chat. Critical thinking can’t come from recycled confidence.
Step 5: Keep Friction in the Workflow
The easiest way to go soft with AI is to make it too smooth. Leave friction in your process. Use checklists. Ask the “why” question twice. Invite disagreement. Let the AI earn your trust every time.
Example workflow:
- Draft an idea with ChatGPT.
- Run it through your Anti-Hallucination Checklist.
- Ask ChatGPT to critique its own answer.
- Compare both versions and combine the best insights.
Friction slows you down just enough to think clearly — which is exactly the point.
Final Thought
ChatGPT can be your collaborator, not your clone. If you only teach it to sound like you, it’ll stop surprising you. But if you teach it to challenge you, it becomes something rare —a machine that makes you smarter.
You don’t need an AI that agrees with you. You need one that makes you worth agreeing with.