Using AI without sounding like everyone else

There are two ways to use AI as a writer. The first is to ask it for an article and publish what comes back. The second is to use it as a research assistant, a sparring partner and a first-draft machine while keeping every decision that matters firmly in your own hands. The first pays nothing. The second is the most useful skill a working writer can pick up right now.

Why AI writing sounds like AI writing

Language models produce the most probable next word. Averaged across the internet, the most probable sentence is the blandest one. That is why unedited AI copy has a recognisable texture: balanced, hedged, smooth, and entirely free of anything a specific human being would actually say.

It also has tells. Empty openings that restate the title. "In today's fast-paced world." Triads of adjectives. Symmetrical paragraphs of exactly the same length. A conclusion that summarises what you just read instead of landing a point. Readers may not name these things, but they feel them, and increasingly so do search engines.

What AI is genuinely good at

• Research scaffolding. Summarising a field, listing the arguments, surfacing terms you did not know to search for.

• Structure. "Here are my twelve messy notes — propose three possible article structures." It is very good at this.

• Interrogation. "What is the strongest objection to this argument?" Far more useful than asking it to write.

• Tedious transformations. Turning a transcript into notes, a long piece into a summary, a list into a table.

• Second-pass editing. "Where is this repetitive? Which sentences say nothing?"

What it is bad at, and will stay bad at

It cannot have been in the room. It does not know what a client said on the call, what went wrong on the job, what you believe and would argue about at a dinner table. Everything valuable in commercial writing — specificity, judgement, opinion, lived detail — is exactly what the model does not have.

That is the whole opportunity. As the cost of generic text falls to zero, the value of specific, experienced, opinionated writing goes up.

A process that works

1. Decide the point before you open anything. One sentence: what do I want the reader to think differently after reading this? If the model supplies the point, the piece has nothing in it.

2. Dump your own material first. Voice-note or type everything you know: stories, numbers, objections, the thing you always end up saying to clients. Ten messy minutes of this is the difference between your article and everyone's article.

3. Use AI to organise, not originate. Feed it your dump and ask for structures, gaps and counterarguments.

4. Write the parts that matter yourself. The opening, the arguments, the stories, the conclusion. Let it draft the connective tissue if you like.

5. Strip the tells. Cut every empty opener, every hedge, every sentence you would not say aloud. Vary your paragraph lengths. Put a real example in place of every abstraction.

6. Fact-check everything. Models invent statistics, dates and quotes with total confidence. One fabricated figure in a client deliverable costs you the client.

Prompts that actually help

• "Here are my notes. What am I missing that a sceptical reader would want answered?"

• "Give me five ways to structure this argument, from most conventional to most unusual."

• "Rewrite nothing. Just list the sentences in this draft that say nothing."

• "Argue against my central claim as strongly as you can."

Notice that none of them say "write me an article."

What to tell clients

Be straight about it. Most clients do not object to AI in the research and drafting process; they object to paying professional rates for something they could have generated themselves in thirty seconds. Explain where you use it and where you do not, and make clear that the judgement, the research and the final words are yours.

Some clients will ask for AI-free work, usually for legal, medical or brand-voice reasons. Take them at their word and price accordingly.

The uncomfortable part

If your writing service was "I turn a brief into competent, unremarkable copy", AI has already taken that job. The work that remains — and it pays better than the work that went — belongs to writers who bring a point of view, ask better questions than the brief, and know a business well enough to tell a client when the brief is wrong.

Use the machine for the typing. Keep the thinking.