attentionfactory
Talk to us
All guides
The free guide

The 3 prompts that stop
AI from lying to you.

AI fails you in three quiet ways: it makes things up, it leans on what it was trained on, and it agrees with you even when you're wrong. Here are the three one-line fixes, plus exactly where to paste them in Claude and ChatGPT so they run on every chat.

× Hallucination× Bias× Sycophancy
01

The three prompts

FIX 01

Hallucination

It invents facts that look real

Stops the confident made-up answers, the fake stats, and the source links that don't exist. Forces it to admit uncertainty and go verify.

Never guess or give false information. If you're not sure something is true, tell me you don't know and go find the real answer instead of making things up.
FIX 02

Bias

It leans on whatever it was trained on

Some viewpoints get amplified and others barely surface. This forces it to argue the other side so you see the full picture, not just the default one.

Reframe this in a way that challenges how you see the problem and make the strongest case for the opposite side.
FIX 03

Sycophancy

It agrees with you even when you're wrong

The sneaky one. You share a plan that's going to flop and it calls you a genius. This swaps flattery for an honest expert read.

Tell me what a top 0.1% person in this field would think, not what I want to hear, and be brutally honest with me.
02

Where to put them

You can paste a prompt at the end of any single message for a one-off. But the real move is setting them once so they run on every chat automatically. Two ways in each tool: across all chats, or scoped to a single project.

Every chat

Claude settings

  1. Open Claude, click your name at the bottom-left, then Settings.
  2. Go to Profile.
  3. Find the personal preferences box, the one asking what Claude should consider in its responses.
  4. Paste in the three prompts, then save. Every new chat now follows them.
One project

Claude project instructions

  1. Open or create a Project.
  2. Open the project's Instructions (set instructions / custom instructions).
  3. Paste the three prompts there and save.
  4. Now every chat inside that project runs the rules, while your other chats stay untouched.

The 10-second version

  • Three failures, three fixes: hallucination, bias, sycophancy.
  • Paste them once into settings for all chats, or project instructions for one workspace.
  • Same idea in Claude and ChatGPT. Set it and your AI stops quietly working against you.