One score is not a verdict
A 62% label means the text resembles that detector's AI examples strongly enough to score 0.62, not that 62% of the sentences are machine-made.
AI detector guide
This is a guide, not the product: the review itself runs in Vortixy Chat, at /chat. AI detectors return a similarity score, not a verdict: the same text can score high in one tool and low in another because each one trained on different data. Short texts score less reliably, heavy editing shifts scores both ways, and false positives are common.
A 62% label means the text resembles that detector's AI examples strongly enough to score 0.62, not that 62% of the sentences are machine-made.
Scales differ between detectors, so a 40% in one tool is not the same evidence as a 40% in another. Never mix scores across tools.
Formulaic human writing can score high and edited AI drafts can score low. Never publish, reject or accuse based on one score alone.
They are moderately useful signals, not verdicts. Accuracy depends on language, length, genre and how much the draft was edited.
Each detector trained on different data and weighs different patterns. Style, length, edits and the source model all shift the score.
Yes. Formulaic phrasing, short length and predictable structure raise scores even for fully human writing.
No. Treat the score as one weak signal inside a human review: read the work, talk to the author and check the process.
Paste your text, keep the context together and export to Word or PDF when it is ready. Free plan available, no install needed.