How we test
Understand the checks
behind your result.
How we test the detectorWe compare results across image sources and common edits.Last reviewed 6 Sep 2026
We keep development, tuning and final test images separate. We check performance for different image sources and common changes such as resizing or compression, so one overall number does not hide a weak group. The final test set is not used to choose a model or adjust its cutoff.
How to read your resultA higher score means stronger signs of AI generation.Understanding results
The image check looks for signs of AI generation. File details and available source records are shown separately. A result can be wrong, even when the signs look strong. Check the image’s source and context before making a decision.
Known limitationsScreenshots, edits and compression can change the clues.Limits published
Screenshots, repeated compression, subtle AI edits and new image tools can weaken detection. When the clues are too weak or disagree, the report says there is not enough information.
Why results can changeImage checks can change as detection improves.About updates
New image tools and editing methods can change the signs we look for. Updates to our checks may change the result of a later scan. An earlier report describes the check made at that time.
What is public today
What can you check for yourself?
Here is what we currently explain or publish, and what still needs more testing.
| Item | Current public status | Interpretation |
|---|---|---|
| What was checked | In your report | The result applies to the image, page, frame or audio clip you selected. |
| What the score means | How it works | A higher score means stronger signs of AI generation. It does not tell you how certain the result is. |
| Source information | Shown separately | File details and signed records can add context. Missing details do not mean an image was made with AI. |
| How your file is handled | Privacy details | Read what is uploaded, who receives it and when our copies are removed. |
| When results can be wrong | Known limits | Screenshots, repeated sharing, edits and newer image tools can affect the result. |
Current answer
What is available in the public service now?
What Made It? checks a selected image for signs of AI generation and shows the file details and Content Credentials available with it. These checks answer different questions, so the report explains each separately.
This result may be wrong. Check the image's source and context before making a decision. A low score does not prove that an image is original or unedited, and a high score does not prove who made it.
Our team maintains these explanations. Privacy explains how your file is handled, and How it works explains how to read your result.
Reviewed by the What Made It? team · Last updated