A QueerAF investigation found that Claude, Gemini and ChatGPT invoked legal or medical language in 94% of 504 responses to questions about trans people. The finding points to a less obvious problem than outright hostile answers: even a seemingly neutral reply can make trans life look like a legal case or a medical question.
Watch Desk analysis
What happened
QueerAF and trans researcher Chloe Turner asked three AI models 168 questions, drawing on themes and concerns submitted by the publication’s readers. The questions covered topics including history, safety, healthcare, policy and personal experience. The investigation says 94% of the resulting responses used at least one legal or medical term.
In its report on the investigation, QueerAF gives examples of models bringing up UK court rulings, the Equality and Human Rights Commission and the Cass Review, including in answers to questions that were not primarily about law or medicine. It says more than a quarter of responses mentioned the For Women Scotland ruling, while the EHRC appeared in nearly a third.
Why it matters
A reply can be factually plausible in its details and still narrow the picture it gives. If systems repeatedly route questions about trans people through courts, clinical language and policy disputes, readers may get a debate about trans existence where they asked for something else. The framing is part of the answer, not decorative trim.
The investigation describes a particular set of prompts and three models, not every possible question or every version of those systems. Its findings are QueerAF’s analysis; the figures should be read in that context, rather than as a universal scorecard for AI.
Our read
This is a useful question for anyone judging whether an AI answer is “balanced”: balanced between which sources, and on whose terms? The striking point is not simply that models mention legal or medical material, but how often that frame appears across a broad set of questions. The sample gives readers something concrete to debate, while leaving room to ask how the prompts were selected and how the results might change with different questions.
What to watch
- Whether QueerAF publishes further parts of its investigation or more detail on its method.
- Whether the model makers respond to the findings or change how their systems answer these questions.
- Whether independent researchers test the same pattern with different prompts and model versions.
Discussion spark: When an AI answer is factually plausible but frames trans life mainly through law or medicine, is that neutral context or a distortion of the question?
Sources and evidence
- Source update (31 October 2026, 05:59 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.