How to tell if a text was written by AI: the signs and why detectors lie

In short: reliably determining whether a text was written by AI is impossible today — neither by eye nor by detector. What gets taken for signs of machine style (smoothness, formulaic structure, no specifics) actually separates formulaic text from living text, not a machine from a human. AI detectors produce false positives, especially on texts by people writing in a non-native language, so an accusation can't be built on their "percentage".

Why this is hard in the first place

A language model was trained on human text and is optimised to produce statistically "normal" text. Good machine text is, by definition, text indistinguishable from average human writing. The task "tell AI from human" runs into the fact that the real boundary isn't machine versus human — it's distinctive style versus averaged style. And people write in an averaged way too, especially when tired, cautious, or working in a second language.

Add that almost nobody publishes raw model output: text gets edited, extended, blended with your own. Pure "human" and pure "AI" cases barely exist — there's a continuum.

What people mistake for traces of AI — and why they're wrong

Which leads to the thing worth holding in mind before any list of signs: what usually gets called "machine style" is a description of averaged text, not machine text. Here's what typically gets presented as evidence:

• Suspicious smoothness. Every sentence roughly the same length and complexity, the rhythm never varies.
• Everything in threes. Lists of exactly three items, symmetrical paragraphs, sections of equal size.
• The "not X, but Y" construction. "This isn't just a tool, it's a way of thinking" — five times a page.
• Zero specifics. No names, dates, numbers or lived examples. True of anything and therefore useful for nothing.
• Marker vocabulary. "Landscape", "unlock the potential", "in today's digital era", "it's important to note".

Not one of these separates a machine from a human. They separate formulaic writing from living writing — and humans write to a template too: when tired, cautious, revising from a study guide, or working in a second language. This list will happily "convict" a tidy person. And conversely: a model's output under a good prompt shows none of them — ask for specifics, a real example and a stance, and the "machineness" evaporates (you can see how in the formula for a strong prompt). The list is useful for editing your own text and useless as forensics on someone else's.

Why detectors lie

An AI detector doesn't "recognise" text. It measures statistical properties — how predictable each next word is (perplexity) and how evenly sentence complexity is distributed (burstiness). Machine text is, on average, more predictable. Every problem flows from that:

• False positives on humans. Someone who writes plainly and predictably looks statistically like a model. A Stanford study (Liang et al., published in Patterns, 2023) found detectors systematically flagged essays by non-native English writers as machine-generated, while native speakers' texts passed. The reason is exactly this: a narrower vocabulary yields low perplexity.
• False negatives. Ask the model to write with more variety, or lightly rewrite by hand, and the detector goes quiet.
• The developer gave up. OpenAI shut down its own AI Text Classifier in 2023, publicly citing low accuracy. That's the company that builds the model — if the problem were solvable, they'd have solved it.
• No calibration. "87% likely AI" looks scientific, but the number is unverifiable: detectors don't publish a transparent method, and accuracy drops on text unlike their training data.
• Short texts are guesswork. A single paragraph simply doesn't carry enough statistics.

The conclusion is blunt but necessary: a detector's percentage cannot be used as grounds for an accusation — not at a university, not in an editorial office, not in hiring. It's a tool with an unknown error rate, unevenly distributed, that lands hardest on people already in a weaker position.

What about watermarks?

It's technically possible to embed an imperceptible statistical mark during generation and check for it later. Google DeepMind released such an approach (SynthID) and opened parts of the tooling. But the method has a ceiling: it only works for models that implemented it, the mark doesn't survive every kind of editing, and nobody is obliged to apply it. It's a useful technology for platforms, not a universal detector for you. It remains contested: there's no single standard.

What to do instead of using a detector

If you're a teacher, editor or manager, "did AI write this" is almost always the wrong question. The real ones are "does the author understand what they submitted" and "does the text do its job". Both are testable:

• A conversation. Two follow-up questions about the content answer everything no detector can.
• The process. Drafts, version history, interim submissions. You see the path, not just the result.
• A task AI can't close alone. Local context, personal experience, data from the session, a tie to a specific case.
• Fact-checking. The most practical marker: if the text has links and figures, open the sources. A link that doesn't exist is far harder evidence than "87%", and it's also a sign of an AI hallucination.
• Open rules. Say in advance what's allowed: "AI is fine for drafting and checking, not for the final conclusions." A transparent rule removes half the problem.

If you're the one accused

It's a real and unpleasant situation: a detector produced a percentage, and the text is yours. What helps:

• Version history. Google Docs, Word, any editor with autosave. It's the best evidence there is — the writing process is harder to fake than the result.
• Drafts and notes. Keeping them is tedious; they're also what saves you.
• Readiness to explain. If you understand your own text, a conversation closes the matter.
• Citing the detectors' limits. Calmly point out that the tool produces false positives, particularly for people writing in a second language, and that the model's own developer discontinued its detector for low accuracy.

How to write with AI without sounding like a machine

And the most practical takeaway. If you use AI honestly — as a draft — "machineness" is removed not by trickery but by substance:

Here's my draft: [text].
Task: make it more concrete without making it longer.
Rules: cut the generalities; every claim gets either an example,
or a figure with a source, or gets deleted.
Break the monotone rhythm: alternate long sentences with short ones.
As a separate list, flag the places that need my personal experience —
I'll fill those in myself.

That last line is the key. A text stops being machine-like exactly when it contains what the model doesn't have: your specific case, your number, your stance. More techniques for working with drafts are in AI at work and in our set of ready-made prompts.

The main thing

AI detectors aren't a truth sensor — they're a probabilistic guess with an unknown error rate. And the "signs of machine style" separate formulaic writing from living writing, not a machine from a human — which is why they can't serve as proof. The only reliable way to know whether a person stands behind a text is to talk to them about it.