Use AI as an editor, not a writer
Your client can tell. The tells that mark a report as machine-written, why they cost you the work, and the way round that keeps the judgement yours.
Here is an uncomfortable truth about the report you're about to send: your client can tell. Not "might suspect" — can tell, increasingly at a glance, because everyone has now read enough AI-written text to recognise the flavour. And a proposal or report that reads as machine-written doesn't just land flat. It quietly answers the one question the client was actually asking — what do you think? — with "I didn't."
The tells, and why they give you away
The first is sheer size. AI writes generously because generating text costs it nothing, so the two-page answer arrives as ten pages in which everything is covered and nothing is weighed. But a client didn't ask for coverage — they can get coverage from a search engine. They asked for judgement, and judgement looks like the opposite: this matters most, these two things matter less, ignore the rest. Length without weighting reads as diligence to the person who generated it and as evasion to the person who has to read it.
The second is the register that no human ever produces. A person who spent a week inside your business writes "the invoices your team sent us". A machine that was handed some files writes "the documents supplied for this conversation" — because to the machine, that's precisely what they were. These phrases are fingerprints. The writer wasn't in the room, wasn't at the meeting, has never met the client — and the prose says so, in passing, to anyone paying attention.
The third is the suspicious tidiness: every section the same shape, every point arriving in threes, every paragraph opening with a sturdy topic sentence and closing with a small bow. Real thinking is lumpier than that. One issue deserves two pages and another deserves a sentence, and a document that gives everything equal, symmetrical treatment is telling you nobody decided what mattered.
What the reader concludes
A report is a sample of your thinking — often the only sample a potential client gets before deciding whether to pay for more of it. If it reads as machine-written, they draw one of two conclusions. Either you didn't read your own report carefully, which is careless. Or you did, and considered it a fair representation of your judgement, which is worse. Either way, the thing you were selling — that a capable person looked hard at their situation — is exactly the thing the document disproves. Nobody was ever persuaded to hire a firm by prose they suspect nobody at the firm has read.
The part people miss: writing is the thinking
The deeper problem isn't detection. It's that the struggle of writing — deciding what goes first, what gets cut, what you're actually recommending when forced to say it in one sentence — is not the packaging of the analysis. It is the analysis. The discipline of getting ten pages of material down to two is where you discover what you think. Hand the writing to a machine and you don't just risk sounding like a machine; you skip the very work the client is paying for. Delegate the typing and you've saved an hour. Delegate the writing and you've delegated the judgement.
The right way round
None of which means keeping AI away from your reports. It means giving it the other chair — the editor's, where it is genuinely excellent. Write the draft yourself, badly and fast; ugly first drafts are fine because nobody will see them. Then put the AI to work on what it's good at: which sentences are unclear? Where do my figures contradict each other? What would a sceptical reader push back on? What did the brief ask for that I haven't covered? Cut a quarter without losing meaning. Ask it to critique and question, not to rewrite — and accept its suggestions one at a time, on your judgement, rather than pasting its version over yours. Used this way it will catch the inconsistency on page four, the claim you never supported and the paragraph doing nothing, and the voice on every page remains recognisably the person the client thought they were hiring. One caution from our governance guide bears repeating: a client-identifiable draft belongs on a business-grade AI account with the proper commitments, never on someone's personal chatbot login.
Our own cards on the table
This site practises the arrangement it's describing. Our news articles are drafted by an AI agent — we say so on each one — and not a word of them is published until a person has read it, edited it and put their name on the byline. The rule we run, and the one we'd suggest for your reports, is not "no AI in the writing". It's simpler and harder: a human owns every sentence that goes out with their name on it. The machine can sharpen your thinking. The moment it replaces it, the reader can tell — and the reader is the one with the budget.
How this was made
Drafted by our research agent from the sources above, then edited, fact-checked and approved by a human before publishing. We publish this note on every article because we think you should know.
More articles
What should we automate first?
A scoring method you can run in a meeting: three questions per task, and the honest list of things that should stay human.
Your staff are already using AI. Now what?
Somebody in your business pasted work into a chatbot this week. Banning it won't stop it — but three practical moves turn a quiet risk into your automation roadmap.
The first 90 days of an agent project
You've said yes — or you're deciding whether to. Here is what actually happens, week by week, including the fortnight where it feels slower than the old way.