Writing with AI without sounding like a robot

The complaint about AI writing is always the same: it reads like AI. That is not a limitation of the model, it is a consequence of how people prompt it and what they accept back. Both are fixable.
Why the default output sounds wrong
Left alone, these tools produce prose with recognisable tics. Sentences of uniform length. Balanced constructions where every point gets a counterpoint. Openings that restate the question. Closings that summarise what was just said. Adjective pairs where one would do. Transitional phrases carrying no information — moreover, in today's fast-paced world, it is important to note that.
None of that is wrong exactly. It is just the texture of writing produced by something optimising for inoffensiveness, and readers detect it immediately even when they cannot name what they are detecting.
What actually changes the output
- Give it your own writing. Paste three or four paragraphs you wrote and ask it to match the rhythm and vocabulary. This single step does more than every other technique combined.
- Constrain the shape. Specify a length in sentences, ban a list of words, forbid summary paragraphs. Constraints break the default patterns because the defaults require freedom to appear.
- Ask for a draft, not a piece. Framing matters. Requesting rough material you will rewrite produces looser, more useful text than requesting a finished article.
- Work in fragments. One paragraph at a time, with you writing the connective tissue. The seams between AI paragraphs are where the tell is strongest.
- Make it argue. Ask for the strongest version of one position rather than a balanced overview. The both-sides reflex is the most recognisable habit in the whole category.
The editing pass that fixes most of it
Read it aloud. Anything you stumble over gets cut. Then do three specific things: delete every sentence that only restates the previous one, replace every abstract noun with the concrete thing it stands for, and vary the sentence lengths deliberately — a short one after two long ones resets the reader's attention.
Then cut the first paragraph. It is almost always a warm-up that says what the piece will be about instead of starting it. The second paragraph is usually the real opening.
The line worth holding
Anything factual needs checking, because these tools produce wrong information in exactly the same confident register as right information. And anything carrying your name should contain at least one thing a model could not have written — a specific number, an opinion with a cost attached, something that happened to you. That is what makes writing worth reading, and it is the one part that cannot be delegated.
Disclosure and the practical question
Whether to say a piece was drafted with AI depends on context, and the honest test is whether the reader would feel misled to learn it. Academic and journalistic work generally has explicit rules; internal documents generally do not. For anything in between, the safer position is that AI-assisted drafting is unremarkable and AI-generated opinion presented as your own judgement is not.