0 · Intuition

“Is this AI?” Anyone: a teacher, a parent, a reader, a recruiter.

Evidence standard: A calibrated likelihood with a plain-language explanation, offered as information and never as a verdict.
Cost of a wrong call: Embarrassment, or a broken relationship if the hunch is voiced as an accusation.

What it can and cannot claim

Can claim

There are signals worth a second look, and here is what they are.
Here is how confident a calibrated screen would be, with the range, if one has been run.

Cannot claim

That a named person used AI.
That a piece of work is dishonest, plagiarized or in breach of a rule.
Anything at all about a person who has not had a chance to explain.

Error rates

Silvector publishes no measured detection error rate for this rung yet. Any number you see elsewhere on this site is labeled with its source, and none of them are ours. Untrained people are a weak instrument too: in one controlled study, evaluators told GPT-3 text from human text at about chance level.

A valid error-rate statement at this rung includes:

The content type and language the rate was measured on.
False positive rate on human-written work from the same population as the person in question.

Bias risks

Text detectors misclassify writing by non-native English speakers as AI-generated far more often than writing by native speakers.
Short or formulaic writing (forms, lab reports, translated text) carries few signals either way.

What Silvector has at this rung

Capability	State	Where
A consumer "check this" experience for text, images, audio or video	not built	`No content detector ships in this repository.`

How to escalate

To reach Rung 1 · Screening:

Run a screening signal that has a stated error rate for this kind of content (rung 1).
Keep the original file or link, unedited, so a later check sees what you saw.

Sources

Next: Rung 1 · Screening · The whole continuum