Most people try to get better answers from ChatGPT by writing longer prompts or adding more details.
What made the biggest difference for me wasn’t complexit, it was one change in my custom instructions.

I told ChatGPT, in plain terms, to respond with honest, objective, and realistic advice, without sugarcoating and without trying to be overly positive or negative.

That single instruction changed the entire tone of the model.

What I Noticed Immediately

Once I added that custom instruction, the responses became:

• More direct – less “supportive padding,” more straight facts.
• More realistic – no leaning toward optimism or pessimism just to sound helpful.
• More grounded – clearer about what’s known vs. what’s uncertain.
• More practical – advice focused on what’s actually doable instead of ideal scenarios.

It didn’t make the model harsh or pessimistic. It just stopped trying to emotionally manage the answer.

This is the intruction:
I want you to respond with honest, objective, and realistic advice. Don’t sugarcoat anything and don’t try to be overly positive or negative. Just be grounded, direct, and practical. If something is unlikely to work or has flaws, say so. If something is promising but still has risks, explain that clearly. Treat me like someone who can handle the truth and wants clarity, not comfort.

Why This Works

Large models often default to “safe,” diplomatic phrasing because they assume you want comfort, optimism, or positive framing.
By defining your expectation upfront, you remove that ambiguity.

Instead of guessing your preferences, the model acts within the instruction:
“Be honest, objective, and realistic. Don’t sugarcoat. Don’t dramatize. Just be practical.”

This gives it permission to drop the unnecessary softening and focus on clarity.

I’m diving deep into prompt design, AI tools, and the latest research like this every week.
I recently launched a newsletter called The AI Compass, where I share what I’m learning about AI, plus the best news, tools, and stories I find along the way.

If you’re trying to level up your understanding of AI (without drowning in noise), you can subscribe for free here 👉 https://aicompasses.com/