Researchers tested whether describing AI in human terms actually warps how people think about it — and mostly found it doesn't, at least not right away.
A study comparing how 815 participants responded to passages about large language models and recommendation systems found that swapping in anthropomorphic language — words that attribute human traits to software — produced no substantial change in perceptions across several measured dimensions. A separate condition, where participants read text explicitly warning about AI dangers, did move the needle, suggesting people's views are responsive to framing in general. The researchers designed the passages to mirror realistic public-facing AI discourse, not academic abstractions.
The finding cuts against a widely held assumption in AI ethics circles: that calling a chatbot "intelligent" or saying it "understands" you is actively distorting public understanding and warping policy debates. If single-exposure effects are this modest, the concern may be more about cumulative media diet than any one headline. That has real implications for how regulators and journalists think about disclosure requirements around AI language.
The study leaves the door open for gradual, repeated exposure to do more damage over time — which is exactly how most people actually encounter AI coverage, through a drip of news articles, product pages, and social posts rather than a single controlled reading.