A record eight Pulitzer winners and finalists disclosed using AI, and what they used it for is the interesting part
Eight Pulitzer awardees declared AI use this year, the most since disclosure became mandatory in 2024. In almost every case the tool was used to search or translate large document piles, not to write.
Eight of this year’s Pulitzer Prize awardees, five winners and three finalists, told the jury they had used AI somewhere in their reporting. That is the highest number since the prize made disclosure mandatory in 2024, and according to Nieman Lab the pattern in what they used it for is remarkably consistent: almost all of it was searching, sorting or translating large piles of documents, and almost none of it was writing.
The examples are specific. The Wall Street Journal ran an internal language model over thousands of public records about the Texas floods to summarise them. The Associated Press used one to search tens of thousands of leaked documents on Chinese surveillance technology. The New York Times used GPT-5 to check a classification of SEC crypto cases that its journalists had already done by hand. The Minnesota Star Tribune used ChatGPT to translate a diary written in faux Cyrillic, a style of writing that substitutes Cyrillic letters for similar looking Latin ones, and then had human language experts verify the result. Pulitzer administrator Marjorie Miller told Nieman Lab the industry now accepts that AI “is here to stay” and has a clearer sense of where it does not belong, “such as in writing and editing stories in any format that might be considered for a Pulitzer Prize.”
What is behind this. Two things are worth separating here. The first is that finding a needle in 40,000 documents is a task where a language model is genuinely strong and where a mistake is cheap, because a human then opens the document and reads it. Writing prose is the opposite: errors are invisible, expensive and travel. Newsrooms have arrived at that split empirically rather than ideologically. The second is that the disclosure requirement is doing quiet work. Because entrants must declare AI use, we get an unusually honest sample of how professionals actually use these tools, instead of vendor case studies. Note also what every example has in common: a human checked the output. The Star Tribune hired language experts. The Times used the model to verify manual work, not to replace it. The rule from next year extends to book entries too.
What this means for you: this is the most useful template going for anyone deciding where to point an AI tool. Use it where you have too much material and a way to verify what comes back, so long email threads, meeting notes, contracts, a folder of PDFs, research for a course. Be much more careful where the output is the finished thing and nobody will check it. If you have been unsure whether using AI at work is somehow cheating, the Pulitzer answer is clear and worth borrowing: it depends entirely on which part of the job you hand over, and you say so out loud.
Sources
MiniMax released the weights for its H3 video model, and an open model now tops a video ranking for the first time
MiniMax published the weights for H3, a 33 billion parameter video model that handles text, images, video and audio together. Artificial Analysis ranks it first in video editing. Two components stayed closed, and the licence has a revenue cap.
