How we use AI

We use AI tools across our work, including for research, drafting, checking and illustration. The final judgement is always human: our editor reads, edits and approves every piece, makes the final call on what we publish, and is responsible for it. AI never holds a byline or speaks for our standpoint, and it never invents a quote, a figure, a source or a voice: if we can't source something, we say so.

What AI does here

• Research and drafting: finding and reading sources, and writing first drafts that our editor then rewrites and checks. We build each piece from the public record, with every claim checked against the source it rests on.
• Fact-checking: before we publish, AI tools help us check each draft: that every fact matches its source, every quote is exact, and nothing is unfair to the people named. They point out problems; the editor checks each one and decides. Which AI services we use, and what we send them, is on our Privacy page.
• Illustration: our Frames, explained below.
• Listening: some pieces can be played aloud. The voice is AI-generated, and the player always says so.
• Reader messages: an AI service helps us sort and summarise what you send us. Our editors read your messages themselves.

Why we use AI-generated images

Many of our stories have no photograph we can use: the people at the centre may be survivors who must not be shown, the moment may never have been photographed, or the only pictures can't be licensed. Even when a photo exists, showing a person's caste can put them at risk, and most stock images of Dalits show only poverty and suffering. So we made our own art form, Frames: stills from an imagined film, drawing on the visual language of the anti-caste movement. The person at the centre acts (claiming, reading, organising, working) in deep Ambedkarite blue for dignity and assertion, and neighbourhoods, fields and workplaces are shown as lived and dignified, never as squalor.

The people in Frames are fictional. They never stand in for a real person, a survivor, an accused or an actual event, never show violence, and never signal caste through skin colour or features.

Where a real photograph is right for the story, we use it, credited to its photographer and/or source.

How we label it

• Images: every AI-made image is labelled "AI-generated image, The Dalit Project" on the page and in the image file itself.
• Audio: the player says "Read by an AI-generated voice", and the label can't be turned off.