Fix the draft, don't start over

Here’s the most common way people waste time with these tools.

They ask for something. They get back a draft that’s mostly right, with a bad opening and one section that misses the point. They think “this isn’t working”, rewrite the whole request, and get a different draft that’s mostly right, with a different bad opening.

Round three, they give up and write it themselves. Which was always going to happen, because they never once said what was wrong.

The move

Say what to keep, say what to change, and change one thing at a time.

Keep sections 1, 2 and 4 exactly as they are. Rewrite only section 3. At the moment it describes what we did. I need it to state what changed for participants, using only the data in the table I gave you.

Compare that with “make it better”, which is the least useful instruction available and by a wide margin the most used. “Better” gets you different. Usually longer, more emphatic, and with your one good line quietly deleted.

Five worth remembering

“Keep everything except X.” Protects the parts that work. Without it, every revision is a fresh roll of the dice on the whole document.

“Give me three options for the opening paragraph only.” Choosing between options is faster than critiquing one draft and gets you something better. Keep the scope tight: three versions of one paragraph, not three versions of everything.

“This is 40% too long. Cut the weakest third and tell me what you cut and why.” Asking for the reasoning matters, because it forces a decision about what’s load-bearing. Sometimes you’ll disagree with the cut and put it back. Asking only for “shorter” tends to strip out the specifics and keep the padding.

“Check this against these criteria, then revise it.” Two steps in one go, and much stronger than asking for a revision straight off. Give it your real criteria: does every claim trace back to the data, is the tone right for an accreditation file, has any outcome level been overstated. Having it critique before revising genuinely improves the result.

“What did I not ask for that I should have?” Worth asking at the end of a long job. Every so often it catches a missing section you’d otherwise have noticed on Thursday.

It will agree with you

Ask “is this good?” and you’ll be told it’s good. Ask “what’s weakest here?” and you’ll get something you can use.

This isn’t the tool being dishonest. There’s a strong pull toward agreement built into how these things are trained, and you have to work around it. The fix is to make criticism the only available answer:

Name the three weakest things about this draft. Don't tell me what works. Assume a doubtful reviewer who thinks this document overstates its case. What do they go after first?

Know when to stop

Two or three focused revisions gets you most of what’s available. After that you’re into diminishing returns, and past about the fifth go things start getting worse. The draft drifts, good phrasing from earlier gets smoothed away, and the length creeps back up.

When you’re close, stop and finish it yourself. The last ten percent is quicker by hand, and it’s the part that makes the document sound like a person wrote it.

Try this

Take a draft you gave up on. Instead of a fresh request, write one sentence: what to keep, and the single thing to change. See how close two goes gets you.