Learn the basics / Part 5 of 10 / 8 min read
Writing with AI without sounding like AI
Everyone can now spot the default voice: smooth, confident, empty. Here is what gives it away, and the working method that keeps the writing yours.
You have already read AI writing today and known it. Not because of a single word, but because of a shape: an opening that restates the question, three balanced points, a closing paragraph that resolves nothing.
If your writing starts arriving in that shape, people stop reading it. Here is how to keep that from happening.
What gives it away
- The throat-clearing opening. "In today's fast-paced business environment." Delete every opening sentence that could sit on top of any article.
- Balance where you should have a view. It will give you both sides when your reader wanted your recommendation.
- Adjectives standing in for facts. "Significant improvement" instead of "twelve days down to three".
- Tidy triples. Everything arrives in threes, at equal length, because that pattern is everywhere in the training data.
- No specifics. No names, no numbers, no Tuesday afternoon when the thing actually happened. Specifics are the one thing the model cannot invent for you truthfully.
The jobs worth handing over
Structuring a mess. Twenty scattered thoughts into four themes. This is the highest-value use and the output is almost always usable.
The second draft, not the first. Write your rough version, then ask for a tighter one. The ideas stay yours and the polish is fast.
Cutting. "Take this from 400 words to 150 without losing the specifics" is a job it does better than most people, including me.
Being your first sceptical reader. "What would someone who disagrees say?" and "what did I not explain?" both produce better notes than a friendly colleague will give you.
Format changes. The same content as an email, as five bullets, as a slide title. Genuinely tedious, genuinely solved.
The jobs to keep
The opinion. The story only you have. The numbers from your own work. Anything where being wrong has consequences.
A useful test: if the reader would feel cheated to learn a machine wrote it, you should be writing it.
The method that works
- Write your own rough version first, however bad. Ten minutes, no editing. This fixes the voice problem before it starts, because the ideas are already yours.
- Paste it in with the context: who reads it, what you want them to do, how long it should be.
- Ask for a tightened version, not a rewrite. "Keep my phrasing where it works" changes the output a lot.
- Take back every sentence that sounds better in your words. Usually a third of it.
- Read it aloud. Anything you stumble over gets rewritten. This single step removes most of the machine texture.
Three prompts I keep
- "Cut this by half. Keep every specific number and name. Lose the throat-clearing."
- "Rewrite this as if you were explaining it to a smart friend who does not work here."
- "What is the weakest claim in this, and what would make it stronger?"
Those three, on almost anything, do more than a folder of clever prompts.
The honest limitation
AI will not make dull work interesting. If the underlying update is that nothing happened this week, no amount of rewriting fixes it, and the polished version is worse than the honest short one because it wastes the reader's time more elegantly.
The newsletter
Get the next playbook by email
Each issue is one task from real work, step by step, with the prompt ready to copy. Free, and one click to leave. Reply to any issue with the task you are stuck on.
No spam. Unsubscribe in one click.