Post

AI tools lean towards agreeing with you. A post this week put it well: keep steering it and you can end up with your own echo chamber. Six simple habits that get more honest answers out of AI. No technical skills needed. 🧵

1. Set the rules first. Open with how you want it to behave: don't invent statistics, don't assume I'm right because of my job, use trusted sources like the ONS and peer-reviewed papers. Close with a challenge. Top and tail every prompt.

2. Ask before you show your hand. Get its answer before you share your own view. Once it knows what you think, the reply tends to drift towards it.

3. Ask for the case against. "What's the strongest argument that I'm wrong?" or "How could this fail?" Ask why an idea will work and it builds the case. Ask how it fails and you often get a very different answer.

5. Get a second opinion. Open a fresh chat or try a different model, with no history steering it. If the answers differ, that tells you something.

6. Click the source. Even with rules in place, AI can produce a citation that looks real but doesn't say what's claimed. Check it before you repeat it.

One that works less well than people think: "Are you sure?" Models often back down when challenged, even when the first answer was right. A U-turn on its own proves nothing.