How to Get Honest Feedback on Your Writing from AI

If you paste a draft into an AI assistant and ask “What do you think?”, you will often get a lot of praise. “This is clear and engaging! Great job!” That feels nice, but it doesn’t help you fix anything.

Most assistants are built to be helpful and agreeable, so they lean toward encouragement unless you tell them otherwise. The good news is that you can tell them otherwise. With the right request, AI can be a useful second reader. It can spot confusing sentences, missing steps and a tone that doesn’t fit your reader, often in seconds.

This guide shows you how to ask for real criticism, how to give the AI a checklist to judge against, and how to take its advice without losing your own voice.

Step 1: Ask for criticism, not a verdict

The biggest change you can make is to ask for problems, not opinions. “Is this good?” invites a yes. “What are the three weakest parts?” makes the AI go looking.

It also helps to tell the AI who the writing is for and what it should do. An email to a landlord about a broken heater has a different job than a school essay or a bakery’s “About us” page.

I'm going to share a piece of writing. Please act as a tough but fair editor.

What it is: [e.g. an email to my landlord asking for a repair]
Who will read it: [e.g. my landlord, who is busy and a bit slow to respond]
What I want it to do: [e.g. get the heater fixed within a week, politely but firmly]

Don't praise it. Don't rewrite it. Instead:
1. List the 3 biggest problems, most important first.
2. For each one, quote the exact sentence and explain why it's a problem for this reader.
3. Tell me one thing a reader might misunderstand or skip.

Here is the writing:
[paste your text]

The line “Don’t rewrite it” matters. If the AI rewrites the whole piece, you get its version of your writing instead of notes you can act on.

Step 2: Give it a rubric

A rubric is just a short checklist of what “good” means for this piece. Teachers use them to mark essays. You can use one so the AI judges your writing against your goals, not some vague idea of good writing.

You can write your own rubric or ask the AI to help draft one first. Either way, keep it to four to six points.

Please score my writing against this rubric. For each point, give a score from 1 to 5, quote the part of the text that supports your score, and suggest one specific fix if it scores below 4.

Rubric:
- Purpose: The main point is clear in the first two sentences.
- Audience: The tone suits [reader], with no jargon they wouldn't know.
- Structure: Each paragraph covers one idea, in a logical order.
- Evidence: Claims are backed up with examples or details.
- Action: The reader knows exactly what to do next.

Be strict. A 5 should be rare.

Writing:
[paste your text]

Here are some rubric ideas for different tasks:

• Cover letter: connects my experience to the job ad, specific not generic, under one page, ends with a clear next step.
• School essay: answers the actual question, has a clear argument, each paragraph supports it, conclusion doesn’t just repeat the introduction.
• Small-business webpage: says what we sell and where in the first line, sounds friendly, easy to skim on a phone, tells people how to order.

Quoting matters here too. When the AI has to point to the exact words behind each score, its feedback tends to be more specific and easier for you to check.

Treat the scores as rough signals, not grades. The same text can get different scores in a new chat or after a small change to your wording. Use the low scores to find weak spots, and don’t read a jump from 3 to 4 as proof that the piece got better. The notes and quotes are more useful than the numbers.

Step 3: Protect your voice

Your voice is the way you sound: your word choices, your rhythm, your humor or lack of it. AI suggestions often drift toward a smooth, polished, slightly generic style. If you accept every suggestion, your writing can end up sounding like everyone else’s.

Tell the AI what to keep before it starts.

Before you give feedback, describe my writing voice in 3–4 bullet points (for example: tone, sentence length, any habits that make it sound like me).

Then give your feedback, but follow these rules:
- Keep my voice. Don't suggest changes just to make it sound more formal or polished.
- Only flag things that cause confusion, weaken my point, or don't fit the reader.
- If you suggest new wording, keep it close to my style and offer it as an option, not a replacement.

Writing:
[paste your text]

The voice description is useful by itself. If the AI says your writing is “warm, direct, uses short sentences and the occasional joke,” you can check whether that’s what you meant. Then you can push back on any suggestion that would take that away.

A simple habit helps: read each suggestion and ask, “Would I actually say this?” If not, keep your own wording and fix the problem your way.

Step 4: Push back and go a round deeper

The first answer is rarely the best. Keep the conversation going.

Which of these problems would most likely make [reader] stop reading or say no? Explain why.

I disagree with point [number] because [reason]. Does your concern still apply? If not, what's the next biggest issue?

Now read it as a skeptical [reader, e.g. hiring manager]. What would make them doubt me?

After you revise, paste the new version and ask what changed for the better and what still needs work.

Watch out for

AI feedback is helpful, but it has blind spots:

• Praise creeping back in. Even when you ask for criticism, some replies drift back to compliments. If that happens, say: “Skip the positives. Just list the problems.”
• Made-up problems. Ask for three problems and you’ll usually get three, even if only one really matters. Ask it to rank them, and trust your own judgment.
• Quotes that aren’t yours. The AI sometimes misquotes you, paraphrases while calling it a quote, or “quotes” a sentence that isn’t in your draft at all. Check that each quote really appears in your text before you act on the note.
• Wrong facts. The AI may “correct” a date, name or detail that you had right. Check any factual change against your own sources. For legal, medical or financial details, such as tenant rights or repair deadlines in a landlord email, check with a professional or an official source, not the AI.
• Missing context. It doesn’t know your landlord, your teacher or your customers. If a suggestion ignores something you know about your reader, ignore the suggestion.
• Generic polish. Watch for words you would never use and for every sentence ending up the same length. That’s a sign your voice is slipping.
• Rules for school or work. Some schools and employers have rules about using AI on assignments or documents. Check them before you share coursework or work writing.

A note on privacy

Think before you paste. Writing often contains personal details: names, addresses, health information, money matters or confidential work material. Many assistants let you limit whether your conversations are used for training, but your chats may still be stored for a time. Options differ between tools and account types, and some tools offer few or none. Check your settings and the provider’s privacy policy.

For anything sensitive, replace the details with placeholders like [Tenant name] or [Company] before you paste. The feedback will mostly still work, because the AI is mainly looking at how you write, not who you are writing about. Some notes do depend on the details, like whether a cover letter fits a specific job, so you may need to judge those parts yourself. For work documents, check your employer’s policy on AI tools first.

Try it today

Pick something you’re writing this week, even a short email. Use the Step 1 prompt, fix the biggest problem, and see if the piece gets clearer. You stay the author. The AI is just a second reader that doesn’t get tired of looking.