How to Read an AI Detection Score
Updated
Knowing how to read an AI detection score stops a number from being treated as a verdict. On this site, the AI text, image and video checks each return a percentage and a label. Here is exactly how those labels are set, what the highlighted sentences add, and how to act on a result.
Limitations: Every result is a probability that can be wrong, not proof of how something was written. How results can be wrong →
3 free scans in total, shared across all five tools (not per day or per month). No signup needed to start.
How to Read an AI Detection Score on This Site
- 55% and above: labelled an AI-generated signal: the text reads strongly like machine writing to the provider's model.
- 30% to 54%: labelled mixed or inconclusive: parts read as AI-like, parts as human.
- Below 30%: labelled a human or authentic signal.
- Every band: is a probability from a third-party model, not a measurement of how the text was written.
What a Percentage Is Not
A 70% score does not mean 70% of the text was written by AI, and it does not mean there is a 70% chance someone cheated. It is the model's confidence that the text as a whole looks AI-generated.
Scores near a band edge are the least stable. Small edits can move a 52% to a 58%, which changes the label without changing anything important about the text.
Use the Highlighted Sentences
Text results highlight the sentences the provider flagged. Read them: are they the generic, summary-style sentences, or a distinctive passage? A high score driven by a few stock phrases means something different from one where every sentence is flagged.
Results on texts near the 300-word minimum are less reliable than on longer texts, because there is less evidence either way.
Acting on a Result
For your own writing, use the score to find passages worth rewriting in your voice. For someone else's, use it as a reason to look closer: compare with their other work, ask about their process and look at drafts. Never treat a single score as proof.
Frequently asked questions
On this site it falls in the AI-generated signal band (55% and above): the model finds the text strongly AI-like. It is still a probability, not proof.
No. It is the model's confidence about the whole text, and 40% falls in the mixed or inconclusive band.
Scores reflect the whole text, and results near a band edge can move with small edits.
Image results use the same bands: 55% and above is labelled likely AI, 30-54% uncertain, below 30% likely authentic. Video results use the same bands for the badge, alongside the provider's own result label.