The AI likelihood score estimates how likely a piece of writing is to have been generated by a language model. It is an estimate about the text, not a fact about you.
What it measures
Language models tend to produce writing with recognizable patterns: even sentence rhythm, predictable word choices, low variation in structure. The score reads those patterns and reports a likelihood for the document and for each sentence.
What it cannot do
- It cannot prove anything. A high score means the writing looks like model output. Plain, formal or formulaic writing by a person, translated text, and heavily edited drafts can score high. Model text that has been reworked can score low.
- It is not a detector of tools. Grammar checkers, translation and dictation can nudge the score without any AI writing.
- It is not what your institution uses. Other detectors use other methods and give other numbers. Silvertext is a private pre-check; nothing is shared with your school or employer.
How to use it
Treat a high sentence as a prompt to look at that sentence. If it is your own writing, leave it; if it reads flat or generic, rewriting it in your own voice is a good edit regardless of the score. Do not rewrite a document just to lower a number.
Sentence-level findings
Sentences above a likelihood threshold are shown as cards so you can see which parts drive the score. See the article on sentence-level AI findings.