How to check text for AI signals, and why no detector is reliable

Deze inhoud is in het Engels.

People often want to know whether a piece of writing came from an AI writing tool. It is a fair question, but the honest answer is that nobody can tell for certain. This guide explains what detectors measure, why they get it wrong, which signs are worth noticing and how to handle a doubtful text fairly. The most important point comes first: no AI detector is reliable, and a detector result must never be used to accuse anyone of anything.

Door Valtrinity. Bijgewerkt op 18 oktober 2026.

No detector is reliable

Every tool that claims to spot AI writing gives an estimate, not a fact. These tools look for patterns such as how predictable the wording is and how evenly sentences are built. People write in those patterns too, especially when they write formally, follow a template or are writing in their second language.

That means detectors make two kinds of mistake. They flag human writing as machine written, and they miss machine writing that has been edited or rewritten. Neither mistake is rare, and a score gives no way of knowing which one you are looking at. A high or low percentage is a hint at best.

• Detectors give estimates, not proof.
• Human writing is often flagged, and edited machine writing often passes.
• Formal, simple or translated writing is flagged more often.
• The same text can score differently from one run or one tool to the next.

Never use a result to accuse anyone

A detector result is not evidence of misconduct. Accusing a student, a colleague, a freelancer or a job applicant on the strength of a score is unfair and can do real harm, because the person cannot prove a negative and the tool cannot explain its reasoning.

If you have a genuine concern, treat it as a reason to start a conversation, not as a verdict. Ask the writer about their process, ask them to talk through the argument or the sources, and look at drafts and notes. Follow your own organisation's policy, and keep the decision with people.

Signs worth noticing

Some features of a text are worth a second look. None of them proves anything alone, and each has an ordinary human explanation.

• Smooth, even sentences with little variety in length or rhythm.
• Confident claims with no specific sources, names, dates or places.
• References or quotations that do not exist when you search for them.
• Repeated stock phrases and tidy lists of three.
• A summary that sounds right but says little that is specific to the topic.
• A sudden change in voice, vocabulary or quality compared with the writer's earlier work.

Checks that tell you more

Checking the facts is far more reliable than checking the style. Look up the sources, quotations and figures that the text gives. Invented references are one of the few signs that point clearly to careless use of a writing tool, though even they show only that the text was not checked, not who wrote it.

Compare the text with the writer's known work, ask for earlier drafts and version history, and talk to the writer. A plagiarism checker can show copied passages, which is a different question from whether a text was machine written.

Using the AI content detector sensibly

If you use a detector, use it the way you would use a smoke alarm that sometimes goes off for toast: as a prompt to look more closely, never as the final answer. Run the same text more than once, read the highlighted passages with your own judgement and write down what you checked.

Never share a score as if it were a finding. Do not use it alone to grade, discipline, reject or dismiss anyone, and tell the person what you are looking at if you raise a concern.

• Treat a result as a prompt to look closer.
• Check facts and sources yourself.
• Talk to the writer before you decide anything.
• Keep a record of what you actually verified.

Tools die in deze gids worden genoemd

Veelgestelde vragen