SynthID Detector for Teachers: What It Can and Can't Prove

By Alex | Last Updated: October 2026

Independent information site. Not affiliated with or endorsed by Google. The official detector is at synthid.com.

No. The public SynthID Detector can't tell you whether a student used AI to write an essay. It checks images, video and audio, and it has no text option. It can help with suspicious image, audio or video submissions, within clear limits, and it should never be the only basis for an accusation.

What it can and can't check

Student work	Can the public detector check it?	Notes
Essay, report or code	No	The public site has no text option
A photo, artwork or diagram	Yes, if a supported tool made it	A clean result proves nothing
An audio recording	Yes, if a supported tool made it	Same limit
A video	Yes, if a supported tool made it	Gemini can say which parts carried the mark
A screenshot of text	Not useful	It is an image, and written text isn't marked in the pixels
Work from a tool outside SynthID	It comes back clean	Not a SynthID partner

What a positive result tells you

A positive result means a SynthID partner's AI tool made or edited at least part of the file, and the file may have been edited since. It does not say which tool, how much of the file is AI, whether the assignment allowed AI, or what the student intended. Treat it as a reason to ask questions, not as a verdict.

What a clean result tells you

Nothing about authorship. The student may have used a tool that doesn't add SynthID, or the mark may have been weakened by editing. A clean result is never evidence that a student did the work themselves.

Why AI text detectors mislead

Many teachers hope a tool can check essays. The evidence is discouraging. OpenAI withdrew its own AI text classifier on July 20, 2023, citing its low accuracy. Studies cited at the time found that detectors often flagged human writing as AI-generated, especially writing by non-native English speakers, which makes false accusations a fairness problem as well as an accuracy one.

SynthID does have a text watermark, but it is narrow:

It only marks text produced by Gemini, not text from other chatbots.
Google's developer documentation says detection is probabilistic and returns watermarked, not watermarked or uncertain.
It is less effective on factual responses, where there is little room to vary the wording.
Confidence can drop substantially if the text is thoroughly rewritten or translated.
It isn't designed to stop determined bad actors.

So even a public text checker built on it could neither clear nor convict a student. We did not evaluate specific commercial detectors here. The evidence above is about the general approach and OpenAI's withdrawn tool.

Safer practice

These are general good practice, not findings from the sources above.

State your AI policy. Say clearly what is allowed on each assignment.
Collect process evidence. Drafts, outlines, notes and version history show how work developed.
Talk to the student. A short conversation about their choices or their argument is far more informative than a score.
Keep a baseline. A short in-class writing sample shows a student's usual voice.
Assess the process. Oral defenses, annotated sources and reflections are harder to outsource.
Use detectors as prompts, not proof. Whatever a tool says, follow your school's academic integrity procedure.

If an image, audio or video comes back positive

Read the exact wording on the result screen.
Check the assignment rules on AI use.
Ask the student how they made it and request the original and project files.
Remember a positive can mean AI edited part of the file, such as a retouch or a filled background.
Follow your school's procedure rather than deciding alone.

Questions teachers ask

Can SynthID detect ChatGPT essays?

No. The public detector doesn't check text, and the announcements we read about OpenAI cover images, and per The Next Web audio, not written output.

Can I paste an essay into synthid.com?

The public site has no text option.

Does a clean image result prove the student made it?

No. It only means no supported SynthID mark was found.

Is a positive result enough to fail a student?

No. It shows AI involvement at some level, not misconduct. Check your policy and talk to the student first.