How to Use AI as a Writing Partner Without Letting It Write For You
/Five jobs AI does well for a writer (research, pushback, headlines, tone, summaries), where it fails, and the rules that keep the thinking yours.

TL;DR: Let AI poke holes, dig up leads and throw options at you. Keep the draft for yourself, because the thinking is the part worth keeping and you can’t check what you never thought through.
Half the internet thinks AI is the second coming. The other half thinks one prompt turns you into a fraud for life. Everybody else is stuck in the middle wondering what the hell to actually do with it.
I’d use it like this.
What does AI actually do to your writing if you let it draft?
It eats the part you needed. You ask for a draft, it hands you one, you paste it in, and you just skipped the fight where your brain connects two things nobody else connected and you finally get something you didn’t get five minutes ago.
You got output. You paid for it with learning.
AI is a power tool. Build a house with it or take your hand off with it, the tool doesn’t care either way. So the real question is where in the process you let it touch the work.
Can AI replace a writer?
Not one with something to say. Interchangeable work was always going to lose to something cheaper, and that was true long before chatbots. If you’ve got an actual take and your own way of saying it, there’s nothing there for a model to copy.
And if your boss thinks people and algorithms are swappable parts, he’s cheap. Leave. If you’ve got the talent, you don’t need his leftovers.
How can AI help with research without making things up?
Use it to find leads, then check every single one yourself. Give it your topic and ask for the angles, the best arguments on each side, and what kind of source would settle the question. That beats hunting for a study you only sort of remember.
It still makes stuff up, though. A 2026 study in the Annals of the Royal College of Surgeons of England checked over a thousand references that chatbots handed out for medical questions. The worst models had about a third of them fabricated or unverifiable, some with convincing titles and fake links pinned on real institutions. A few models did fine. You won’t know which kind you’ve got until you check.
So:
- Every citation is a lead until you’ve opened the real page.
- Quote the source, never the chatbot’s memory of the source.
- Can’t find it in two minutes? It doesn’t go in.
- Don’t let the same model fact-check its own work.
How do you use AI to stress-test an argument?
Paste in your thesis and tell it to come at you. Ask for the three strongest objections, then ask what a sharp reader who disagrees would say first.
Some of it will be junk. Usually there’s one hit you were standing too close to see. Fix that hole yourself, in your own words. Pointing at it was the easy part.
If it just nods along, tell it to get meaner. These things lean toward kissing your ass, so your prompt has to push back on that.
Is AI any good at headlines?
Good for volume, bad for taste. Ask for fifty and you’ll have them in thirty seconds. Forty-nine will be trash.
That’s still faster than staring at a blank screen trying to write something that doesn’t sound like every other clickbait headline out there. Grab the closest one and rework it until it sounds like you and it’s true. Overpromise in the headline and the reader bails on the second click.
Can AI check tone and structure?
Yeah, and it’s one of the safer jobs to hand it, because you’re still the one deciding.
- Tone. Paste a paragraph and ask if it reads formal or casual, technical or plain. It’ll tell you. Whether that’s right for your readers is your call.
- Summary. Have it squeeze your 3,000-word piece into three sentences. If your point isn’t in there, you buried it. Sometimes it’s sitting on page four and the whole damn thing needs to be rebuilt around it.
It doesn’t write a word of the piece for either one. You just get to see what a cold reader walks away with.
What rules keep AI from taking over the draft?
Three.
- Don’t paste raw output. Rewrite whatever you keep until it’s yours. If you can’t explain why every choice in the final draft is there, a robot wrote it and you watched.
- Check every source. It’ll invent a study, put words in an expert’s mouth, and link you to a page that never existed, all with a straight face.
- Decide what you’re after. Chase speed and you get garbage fast. Chase learning and you get good slowly. The tool won’t make that call for you.
What’s the one question that decides whether you’re cheating?
Who did the thinking, you or the machine?
Ask yourself if you fought with the idea until it made sense, or took whatever smart-sounding thing came out and called it done. Ask if you came out the other end knowing something new, or just holding more content.
The purists act like writing was clean before this. It never was. Writers have always lifted from everybody who came before them and mashed it into something new, and AI is one more pile to lift from. The fans think putting the tool in everybody’s hands means everybody will be good with it. Hand a guy a chainsaw and he still isn’t a carpenter.
Frequently asked questions
Is it cheating to use AI when you write?
It comes down to who did the thinking. If AI found holes in your argument, pitched headlines you rewrote, or checked your tone, you used a tool and the work is yours. If it wrote the draft and you pasted it in, you handed off the part that counts. If your publisher or your readers expect you to say you used it, say so.
Can I trust the sources AI gives me?
No. Treat every one as a lead. Chatbots still produce convincing citations to papers, quotes and links that don’t exist, and how often depends on the model and the subject. Open the real page, make sure it says what the model claims, and quote from that.
What is the best way to use AI for brainstorming?
Ask for a lot, then throw most of it out. Get a pile of headline, angle or outline options, pick the closest one, and rewrite it in your own voice. Tossing the bad ones is half the value, because it shows you what you actually want to say.
How do I stop my writing from sounding like AI?
Don’t paste raw output, and rewrite anything you keep until you could defend every word of it. Have AI react to your draft instead of producing it. If it started in your head it’ll usually sound like you, and if it started as a prompt it’ll sound like everybody else’s prompt too.
Are AI creators welcome here?
Yes, and we want them. People who build with AI are exactly who this is for. What I care about is whether a human did the thinking, not what was running on the desk.
I think AI is incredible, and it scares me, mostly because I can see a lot of people quietly forgetting how to think and calling it progress. Robots taking over is way down my list. Use it, don’t let it run you, and if you think I’ve got this wrong, tell me. I might.
Comments are open to members. Sign up free →