Actually Works

Actually Works · Episode 73

10 / 80

3 ways to make ChatGPT stop agreeing with you

AI chats tend to agree with you, even about a weak plan. These are three things to try, all from public guides and OpenAI's own help pages. The drawings in the video are illustrations, not screenshots, and none of the three has been measured by us, so treat them as experiments.

The exact clicks

  1. Open ChatGPT and start a new chat.
  2. Try 1: instead of asking What do you think? about your plan, ask What's wrong with this? Compare the two answers on the same plan, each in a new chat.
  3. Open Settings, then Personalization.
  4. Try 2: open Custom instructions and add a sentence such as: Don't agree with me just because I sound sure. State problems clearly and early. Save it.
  5. Start a new chat and ask about the same plan again. Read whether the answer is more critical than before.
  6. Try 3: in Personalization, open the Base style and tone dropdown and pick Candid. Ask about the same plan once more.
  7. Optional: write down which of the three changed the answer most for you. One result does not tell you how it behaves in general.

What changes

  • Try 1 changes the question, so the model is asked to look for problems instead of giving an opinion.
  • Try 2 is a standing instruction that applies to every new chat, unlike a prompt typed inside one chat.
  • Try 3 changes the base tone with a preset. What each preset does may change when OpenAI updates ChatGPT.

What it will not do

  • Not measured by us: nobody here tested how much each try reduces agreement. The sources are guides and help pages, not studies.
  • The preset names and where they sit in Settings can change between app versions. If the menu looks different, search Personalization in the app.
  • A critical answer is not automatically a correct answer. Check any claim that matters.