Mark Your Own Work First

Read your draft back once, with the AI closed, before you ask AI anything.

Most people work the other way round. They finish, paste the draft in, and ask what's wrong.

Back comes the answer, kindly, in a numbered list. You fix three things and the piece improves. But you've learned nothing about your own writing, because you never found out what you would have spotted yourself.

AI feedback comes so fast, and is so often good, that it arrives before your own judgement has had a go. Over a year, your judgement goes unused, and unused judgement gets worse.

Dr Philippa Hardman is a learning scientist who writes about how AI changes learning, and she makes this a rule. Judge your own work before AI comments on it, and prove you can do that with the AI off.

The rule is hers, while the thinking behind the rule comes from how anyone learns to judge work.

To get better at spotting a weak argument, you try to spot it and then find out what you missed. When the finding-out comes first, you never try, so the only skill you build is agreeing with feedback.

She also reports a study in which learners got exactly the same feedback. Some were told a person wrote the feedback and some were told an AI did. The ones who believed a person wrote it worked harder on fixing their work.

The words were the same, and what changed was who they thought was paying attention.

And who pays closest attention to your draft? You.

Two lists

1. Read the draft as the person who'll receive it. Keep AI closed, and write down the three weakest things by hand. Be exact, because "paragraph four says the same as paragraph two" beats "the middle is flabby". If you can't find three, read it again, because they're there.
2. Let AI mark the draft without seeing your notes. Keep your notes to yourself and use the prompt below.
3. Compare your list with AI's list. Where you both caught something, fix it. Where AI caught something you missed, ask yourself why you didn't see it, because that miss is the lesson. Where you caught something AI didn't, trust yourself. AI reads for the average reader, and you know the real one.
4. Fix things in your own order. Start with your list and then AI's extras. The draft improves either way, but the order decides whose judgement was in charge.

Here's a draft. Give me the three weakest things about it, in order of how much they'd bother a careful reader. Quote the line for each one. Don't rewrite anything. Don't soften it.

This adds five minutes. Once a month, you can skip the second step and send something with only your own marks on it.

That's what Hardman means by "prove it with the AI off". It's the check that you still can.

For a two-line reply, skip all of this, and just let AI fix the reply and send it.

Save this for writing where your judgement is part of what you're paid for, or where you want your judgement still working in five years.

"AI's feedback is better than mine anyway." Often it is, especially early on, and going first can feel like pride.

Sometimes it is pride. But comparing is how you find out how much better AI is, and at what.

Skip that and you're guessing, and over time the guess drifts towards "AI is always right".

AI gives you a second opinion. You give the first.

Credit goes to Dr Philippa Hardman for the phrases "judge your own work before AI critiques it" and "prove it with the AI off". They come from her piece "How to Design Around Cognitive Offloading" (August 2026). The study about feedback labelled as coming from a person is one she reports in "The hidden cost of AI-generated feedback" (February 2026). Read her work for the original sources.