The app moderates more than text. Policies can evaluate images, audio, video, and PDFs, alongside text, using the same policies, tests, and datasets you already use.
Media is added by URL: include a column in your dataset CSV with a link to each file, and upload the CSV the way you would any other dataset. See Manage Datasets for the upload flow.
You’ll want to make sure that the schema for your media is mapped correctly so that it interprets the media URLS as inputs for your policies.
In the dataset viewer when you click “expand”, images render as previews and audio plays directly in the table, so you can review media rows without leaving the app.
Once media is in a dataset, testing works the way it does for text: run the dataset against a policy version, review the results, and refine from there. See Evaluations.
A policy evaluating an image, audio clip, video, or PDF uses the same category structure as a text policy. The guidance in Creating Policies applies: clear definitions, positive phrasing, and examples matter just as much for media as for text.
You’ll want to make sure that you configure the correct media inputs for a policy if you want to test it against a dataset with multiple media types. You can do this by clicking the three dots (…) at the top of a policy, then selecting “inputs and outputs”.
