Does AI writing cluster in Medium's tech posts because of the topic, or because of the platform itself?

This explores why AI-written posts seem to pile up in Medium's tech and software categories more than in its other topics. The short answer is that the collection doesn't break Medium's numbers down by category, so it can't tell you why those two lead. What it does have is the bigger picture around that question, and a strong hint about what drives AI use on a platform.

The platform-level story comes from a detector-based study of 2.4 million posts Is AI-generated content rising faster on some platforms?. Between January 2022 and October 2024, the share of posts flagged as AI-written rose from about 2% to about 38% on Medium and Quora. On Reddit over the same period it barely moved, from 1.31% to 2.45%. The rise started in December 2022, right after ChatGPT launched. So the useful unit of analysis may be the platform rather than the topic. Medium and Quora reward polished, explanatory, standalone articles. Reddit runs on quick replies inside a conversation. AI is good at producing the first kind of writing and awkward at the second.

A second study points in the same direction from a different angle. Pangram Labs found that LinkedIn accounted for 62% of all flagged AI content in its sample, with more than 40% of its longform posts rated fully AI-generated Why does LinkedIn have the most AI-generated posts?. The researchers' reading is that people turn to AI most readily when writing is tied to their professional identity. Here the motive is to look competent and active, not to talk with others. Tech and software writing on Medium plausibly fits that pattern well. Much of it is career-building content like tutorials, explainers and "how I built X" posts. It's also written by the people most likely to already use AI tools every day, on subjects where the models write fluently.

That last part is an inference, not a finding. The collection has no category-level data, no surveys of authors' motives, and no check on whether detectors misfire more often on technical prose. That last gap matters because formulaic, jargon-heavy writing may look "AI-like" to a classifier even when a person wrote it. The question worth carrying forward is this: when a platform rewards polished, identity-building content over conversation, does AI-written text tend to fill it? The Medium/Reddit gap and the LinkedIn result both suggest it does.