Get honest feedback with ChatGPT
Get honest, direct feedback from ChatGPT in about 10 minutes: pick the Candid style, cut the praise with custom instructions, then test it on real work.

What you need
Step 1: Pick the Candid style
1 minA built-in style sets the tone of every chat before you write a single rule.
- Go to chatgpt.com and sign in.
- Select your profile icon, then Personalization.
- In the dropdown next to Base style and tone, pick Candid.
You should see Candid next to Base style and tone.
Can't find the setting in the phone app?
Select your profile icon to open Settings. Then go to Personalization and use Base style and tone.
OpenAI describes Candid as direct, with honest feedback and clear next steps. If you also want fewer words, try Efficient instead.
Step 2: Turn down warmth and enthusiasm
1 minLess warmth keeps the full answer but drops the friendly padding around it.
- On the same Personalization page, find Characteristics.
- Set Warm to less, and Enthusiastic to less.
- If replies often carry emojis, set Emojis to less too.
You should see Warm and Enthusiastic both set to less.
OpenAI says the less setting keeps the same length and detail, in a calmer and more neutral tone.
Step 3: Write your feedback rules
3 minCustom instructions apply to every chat, so ChatGPT skips the compliments without being told each time.
- On the Personalization page, make sure Enable customization is on.
- Paste the text below into the Custom Instructions field.
- Keep what you had there before, unless it asks for praise or encouragement.
Your instructionsWhen I share my work for feedback: - Start with the biggest problem. No praise, no compliments, no filler. - Rank the problems by how much they hurt my goal. - For each one, quote the line, say why it's a problem and show a fix. - If my facts or my plan are wrong, say so plainly. - Don't agree with me just to be polite. - Mention strengths only if I ask for them. - If something is good enough, say so in one line and move on.
You should see your rules in the Custom Instructions field, with Enable customization on.
The text won't all fit?
Free and Go save up to 1,500 characters. Cut older rules you no longer use before you cut these.
The rules open with "When I share my work for feedback", so they kick in only when you ask for a review.
Step 4: Test it on real work
4 minA real draft shows whether the rules hold up, which a made-up test can't.
- Start a new chat.
- Paste the prompt below, fill in the brackets and add your draft at the end.
- Read the first line of the reply. It should name a problem.
Your promptReview this [EMAIL / PLAN / PITCH] for [AUDIENCE]. The goal: [WHAT IT SHOULD ACHIEVE]. List the 3 biggest problems, then rewrite the weakest paragraph. [PASTE YOUR DRAFT]
You should see a reply that opens with the biggest problem, not a compliment.
It still opens with praise?
A saved memory may be overriding your settings, OpenAI says. Look through your saved memories for old requests like "be encouraging", and delete them.
Step 5: Sharpen the rules
2 minA few extra lines fix the most common ways feedback stays too soft or too vague.
- If the feedback still feels soft, add this line to your rules: "Open with what's broken. Never wrap criticism in praise."
- If it's blunt but vague, add: "Give every problem a concrete rewrite."
- Start a new chat and run the same draft again to compare.
You should see a sharper reply to the same draft, with a fix next to each problem.
Now it invents problems to find something wrong?
Add: "If the draft is ready to send, say so." Honest feedback includes telling you when to stop editing.
You get
ChatGPT set to give direct, ranked feedback on any draft you share, with no opening praise
Try next
Did this guide work for you?
Thanks. It helps us fix the steps.


