When not to run a detector: cases where the result settles nothing
Detection makes no sense where you cannot act on the result: very short texts, passages with fixed wording, material in a language the tool barely covers, and anywhere you cannot ask the author. All the number produces then is a suspicion you can do nothing with.
Short and formal texts
A two-sentence answer, a form, a structured list — there is nothing to measure. The detector returns a number, but it rests on too small a sample.
Passages with prescribed wording end up the same way: legal clauses, methods written to a standard, step-by-step descriptions.
When there is nobody to ask
Detection is input for a conversation. When no conversation is possible — an anonymous submission, a text by an unknown author — you are left with a number and nowhere to go.
In that case judge the content: facts, sources, how concrete it is. That says more than a score.
When the decision is already made
Running a detector to back up a conclusion you reached some other way is the worst use of one. The tool turns into a rubber stamp.
If the result cannot change your decision either way, do not run it at all.
FAQ
So detection does not work on short texts?
It works worse and the result swings. The shorter the text, the more weight a single sentence carries.
What about texts in other languages?
Most detectors were built mainly for English. In other languages the spread between tools widens, and with translations especially so.
What should I use instead?
A conversation about the content, check questions, version history. With outside suppliers, a clear brief and accountability for the text.
Try it on your own text
Paste a text or upload a file and look at the score and at the specific sentences that came out suspicious.
Run a detection