A 1,100-word piece lands in your queue. The sentences are clean. The structure is tidy, the subheadings are sensible, and the ending wraps everything up without leaving a loose thread. You read it twice and realise you cannot remember a single specific thing it said. Nothing is wrong with it, exactly. That is the problem.
Most editors meet their first suspicious draft this way — not with a smoking gun, but with a faint sense of having just read a well-dressed summary of nothing. What follows is how to turn that instinct into a decision you can defend.
Start With a Better Question
Chasing proof of machine authorship is a losing game. Detector tools return probabilities, not verdicts, and they misfire in both directions: they flag formulaic human writing, and they miss lightly edited machine output. Treat any score as one weak signal, never as evidence on its own.
The useful question is not "did a machine write this?" It is: can the person who submitted it stand behind it, and does it meet our standard? That framing protects you. It also produces a decision you can explain to a writer, a publisher, or a lawyer if it ever comes to that.
Style Clues That Genuinely Help
No single tell proves anything. Clusters do. When four or five of the following appear in the same draft, slow down.
Rhythm that never stumbles
Human drafts are lumpy. A long sentence gets followed by a three-word one. A paragraph runs nine lines and the next is a single sentence because it landed. Machine prose tends toward a steady, medium-length cadence, paragraphs of three to five sentences, each arriving at a tidy closing thought. Read the piece aloud. If your voice never has to change gear, that is worth noting.
Fluency without specificity
"Organisations must balance innovation with operational discipline." True. Also empty. Human writers, even careful ones, tend to smuggle in something oddly precise: the client who insisted on Tuesday deadlines, the spreadsheet with fourteen tabs, the mistake that cost a week. Vague authority — "studies suggest", "many experts agree", "it is widely accepted" — with nothing checkable behind it is a stronger signal than any punctuation habit.
Familiar scaffolding
Watch for openings that announce the essay before writing it, sections that promise to cover three things and then cover exactly those three, and constructions like "It's not just X, it's Y." Lists of exactly three items. A closing paragraph that begins by restating the introduction. Any one of these is ordinary web writing. Five of them in 900 words is a pattern.
Mechanical quirks
- Leftover formatting artefacts: stray ** or ## characters, bullet points that use bold lead-ins on every line.
- Inconsistent quotation marks or apostrophes — curly in one paragraph, straight in the next — often a sign the text was pasted from somewhere else.
- Em dashes used at a rate no human editor would allow through.
- A first draft with no typos, no repeated words, and no half-finished thought in the margin.
Check the Trail, Not Just the Text
Text is only half the evidence. Documents carry a history. Open the revision record and look at how the piece was built: one enormous paste followed by twenty minutes of trimming looks very different from nine sessions across three days, with comments and deleted paragraphs in between.
Check the file's creation date against the deadline. Check whether the draft arrived as a link with visible history or as a flat attachment with none. Look for research artefacts — a notes file, a folder of screenshots, a browser history of source reading — if the writer's process normally produces them.
Be fair about this. Plenty of honest writers draft in a notes app, in email, or in a tool that keeps no history. A missing trail is a reason to ask a question, not a reason to accuse anyone.
Ask for Proof You Can Verify
Specific requests are hard to fake and easy to answer honestly. Ask the writer to:
- Paste the exact sentence from the source document that supports their second paragraph.
- Explain the origin of the one number in the piece that does not appear anywhere else.
- Name the person behind the anonymous anecdote in the opening.
- Say which paragraph they cut, and why.
Someone who wrote the piece answers in a sentence, usually with a small correction along the way — "that number came from the finance deck, but I think it was the 2023 version." Someone who did not will paraphrase their own paragraph back at you, fluently and uselessly. That contrast is more informative than any detector dashboard.
The Two-Minute Conversation
You rarely need a formal meeting. Three questions in a chat window will do:
How did this start — what were you actually trying to say? Then: which part gave you the most trouble? Then: take the opening paragraph and say it out loud in a more casual voice for me.
Writers know their own text from the inside. They can rephrase it on demand, defend a word choice, tell you what they were arguing against. A confident writer will also tell you when a section is weak, which is exactly what someone who assembled a draft from a prompt cannot do.
Set the Policy Before You Need It
Individual judgment calls are exhausting and inconsistent. A short policy solves most of it. Require that submitted work be the writer's own and that they can verify every claim. Make sourcing expectations explicit. State that accountability for accuracy sits with the byline, not with whatever tools were used along the way. Describe what happens when a piece fails verification — usually a revision request and a conversation, not an immediate termination.
If you are considering formal action against a contributor, involve your legal or HR team first. Allegations of this kind carry real consequences, and a detector score is not a finding of fact. Treat process and documentation as part of the job, not an afterthought.
A Practical Checklist for the Next Suspicious Draft
- Read it aloud. Note whether the rhythm ever changes.
- Highlight every claim you cannot check in sixty seconds. Count them.
- Look for the cluster: uniform cadence, generic authority, tidy structure, mechanical quirks.
- Open the document history and compare the draft trail with the deadline.
- Ask two specific, verifiable questions about sources and choices.
- Request a live rephrase of one paragraph, in a different tone.
- Decide on the text, not on the theory: publish, revise, or reject — and write down why.
Do this consistently and you will stop worrying about whether you can prove anything. You will simply be doing what good editors have always done: asking where the details came from, and expecting an answer.
Photo: Ron Lach / Pexels
