How to Make ChatGPT Less of a Yes-Man (and More Human)

Give it explicit permission to disagree: tell it to name risks, say when you're wrong, and admit uncertainty. Put those lines in your custom instructions so they apply to every chat. Then add five tone lines for plain, human-sounding replies.

Stop the flattery: four honesty lines

- If I'm wrong, say so plainly and explain why.
- Before agreeing with a plan, name its biggest risk.
- Don't praise my work unless you say exactly what's good.
- If you're unsure, say "I'm not sure."

ChatGPT also has built-in personalities (base style and tone) in Personalization. Try one, then add your own lines on top.

Sound more human: 5 tone lines

- Write like a friendly coworker, not a brochure.
- Use short sentences and everyday words.
- Use contractions.
- Skip openers like "Great question" and closers like "I hope this helps."
- Match my length: short question, short answer.

Paste them into your custom instructions (on the web: Settings → Personalization). See where custom instructions live for app paths.

Before and after

Before: "Great question! There are several factors to consider when choosing a laptop. Ultimately, the best choice depends on your unique needs."

After: "Get the lighter one. You said you travel weekly, and the extra power won't matter for email and docs."

Before: "That's a fantastic plan! You've clearly thought this through."

After: "The plan works if the launch date holds. If it slips a week, your budget runs out first. Want a backup?"

When the old tone returns

In long chats, the tone can slip back. Remind it with one line: "Back to the short, plain style." If it keeps happening, start a new chat.

Persona vs. one-off prompt

A one-off prompt fixes one chat. Custom instructions fix your default tone. A persona goes further: a fixed voice plus a job, like running your week or helping you reflect. Start with the smallest change that does the job.

FAQ

Will this make AI text undetectable? We make no such claim. These tips are about clearer, more natural replies, not hiding AI use. Follow your school's or employer's rules. Why does ChatGPT agree with everything? It's trained partly on people's ratings, and people tend to rate agreement highly, so it can lean toward telling you what you want to hear. OpenAI says it works against this, but clear honesty lines help.

BOB is a free LOADOUT persona with a plain, direct voice that sorts your week; NORA helps you reflect and set goals. Paste one into the AI you already use. Get BOB free →

Free LOADOUT personas are for adults 18+. They run on your own AI account (for example ChatGPT, Claude or Gemini), so follow your AI provider's terms; we don't see your chats. AI can be wrong; check anything important.