The Times Said Nothing Changed. 3,242 Articles Disagree.

Around 2022, the New York Times changed the way it frames trans people. The subjects stayed the same; the lens moved — who carries a story, what the paper treats as settled, what it treats as doubtful. For years that shift was something trans readers could feel in the paper and never prove. Alejandra Caraballo, the civil rights lawyer, scored 3,242 Times articles against one fixed rubric, from January 2014 through this year, and the change turns up on every dimension she measured. Protective framing fell and kept falling. Anti-trans voices grew more prominent until they rivaled trans people in coverage nominally about trans people. Conflict language crowded out the language of ordinary life. The medical beat — the stories about trans health care — moved hardest of all.

The Times has heard versions of this accusation for years, and it has always had the same defense: name the article. Any single story can be defended on its own terms, the paper’s editors defend each one, and the argument resets the next morning. A shift in framing doesn’t live in any single article. It lives in the pattern across thousands of them, in the one place no reader — and no editor, honestly — can see from a day’s paper. So Caraballo read the thousands.

The method is what the paper will attack, so the method is built to be attacked. Caraballo pulled the corpus from the Times’ own public data feeds, including every article the paper’s own editors tagged as trans coverage — 2,266 pieces carrying the Times’ own subject labels, 3,242 after de-duplication. Her rubric never asks whether a story is good for trans people, only how it is framed: who carries it, what it treats as settled, what it treats as doubtful. Three language models from three different companies scored every article against that rubric independently, a word-counting sentiment pass with no AI in it ran as a control, and one of the three models is open-weights and runs on a home computer, so anyone can rerun the whole thing without her cooperation, her servers, or her permission — and every one of those measures, run by different code on the same twelve years, landed on the same year. 2022. The data is going up on GitHub.

She is candid about where she stands — “I don’t think the existence of transgender people should be up for debate,” she writes — and candid that the findings can’t lean on it. A neutral report on a pro-trans court ruling scores neutral under her rubric, not positive. “If you think I got it wrong, the data is right there,” she writes. “I’d rather be corrected with it than agreed with on faith.”

Three years ago I measured one narrow slice of the same question. In 2024, with Media Matters and GLAAD, I counted how often the Times quoted an actual trans person in its stories about anti-trans legislation, in the year after hundreds of the paper’s own contributors had publicly criticized its coverage. About a third of the stories did. Two-thirds carried no trans voice at all. For one three-week stretch in the spring of 2023, the paper covered anti-trans bills nine times — Florida, Texas, Louisiana — and quoted no trans person in any of them.

That study measured who got quoted. Caraballo’s measures who carries the story, what gets treated as settled, and how the tone moves — across twelve years instead of one — and the two instruments, built by different people and run on different data, found the same thing. Her charts show the trans-voice and anti-trans-voice lines converging year over year until, early this year, they crossed: in the paper’s coverage of trans people, the people arguing against trans rights now carry more of the story than trans people do.

Her data can also date the turn, which is something mine couldn’t do. It concentrates in 2022, and she lays the paper’s own personnel record beside it without claiming more than the record shows. Joseph Kahn became executive editor in June 2022. Jennifer Finney Boylan — for years the paper’s most prominent trans voice — lost her contributing-writer contract early that same year, and Pamela Paul, who became the opinion pages’ most reliable skeptic of trans rights, was given her column that March. A new reporter took over the gender-medicine beat and became its most persistent critic. None of that proves why the coverage turned — Caraballo says so herself — but it puts names and dates on the season the numbers moved.

The Times has responded, through senior communications executive Danielle Rhoades Ha: “We reject any suggestion that our coverage is biased or anti-trans; we report on gender identity and related issues fully and fairly and approach our coverage with curiosity and sensitivity, as we do with every story.” The analysis, she said, “shows a fundamental misunderstanding of our journalism and how it works,” and she noted that the newsroom publishes more than 2,000 pieces of journalism each week. The defense is a claim about framing — fully and fairly — and framing is the exact thing that 3,242 scored articles, three models, and a no-AI baseline all measured moving in one direction. The statement never touches the measurement. It restates the claim the measurement contradicts, and offers the paper’s own volume as the rebuttal to a study built from the paper’s own volume.

“I don’t misunderstand anything,” Caraballo told me when I sent her the paper’s statement. “I analyzed what they wrote, every word of it. If there’s extra context they want people to know, they should write about it, not just mention it in a request for comment.” The context she left out of the study, she said, cuts the other way: “how they put their criticized coverage up for a Pulitzer or how they put star reporters to push critical trans narratives in major pieces. That’s the context I left out so their words can speak for themselves.”

The method deserves scrutiny, and Caraballo built it to take some: the rubric is published, one model is downloadable, the corpus comes from the Times’ own feeds, and the full dataset is being released for anyone to break. Skepticism that stops at “an AI said so” without opening the dashboard is the single-article defense wearing a new coat. The honest critique would rerun her numbers and show where they fail. Nobody has yet. “To my knowledge,” she told me, “no one else has contradicted my results.”

I asked her what the dashboard would show if the paper actually changed. The line on skeptical and restrictive coverage would turn back around, she said; more trans voices would be centered in the coverage; the hyper-focus on trans youth care through a critical lens would end. “They would need to stop treating trans existence as exceptional and instead ordinary.”

Any of this matters beyond media-criticism scorekeeping because the Times still sets the boundaries of respectable opinion in this country. What it frames as settled, courts and statehouses treat as settled; what it frames as an open question turns up in attorneys general’s legal filings, in school-board meetings, and in the minds of parents deciding whether to believe their own kids. The 2023 open letters — one signed by more than 100 organizations and public figures, one by hundreds of the paper’s own contributors — warned that the framing had already escaped the page. The paper dismissed them and said nothing had changed.

Three years later, the critics came back with an instrument.