AI-Generated Junk Posts Flood Social Media: Study Shows Scale and Solutions

Pangram’s research, based on more than one million data points collected via its Chrome extension (priced at $20 / month), defines “long‑form content” as any post exceeding 250 words. The analysis covers LinkedIn, Medium, Substack, X (formerly Twitter) and Reddit.

Platform‑level AI‑generated content rates

LinkedIn: 41 % of long‑form posts are fully AI‑generated. Originality.ai reports 54 % when using a >100‑word definition.

X: 25 % of posts are fully AI‑generated; an additional 23.2 % are AI‑assisted, leaving 52.7 % human‑written.

Medium: roughly one‑third of articles are either fully AI‑written or AI‑assisted.

Substack: 21.9 % of analyzed articles are fully or partially AI‑generated, the lowest rate among the platforms.

Reddit: 11.6 % of posts are AI‑generated or assisted, while 98.1 % of comments are human‑written.

Short posts (50–250 words) across platforms show a 30 % fully AI‑generated rate.

Detection performance

Pangram’s detector maintains a false‑positive‑rate (FPR) ceiling of 0.0001 and an approximate false‑negative‑rate (FNR) of 0.01 on most large language models. Under the same strict FPR constraint, GPTZero and Originality.AI exhibit higher false‑positive or false‑negative rates.

Behavior of thought‑leader authors on LinkedIn

Only 4.3 % of LinkedIn long‑form content is AI‑assisted; the remaining AI‑generated portion is either fully automated or absent. Consequently, 55.2 % of LinkedIn long‑form posts are human‑written.

Conclusion from the report

The study concludes that AI‑generated writing pervades social‑media feeds, especially long‑form posts, mirroring its spread on websites and other online products. The authors argue that informing users about AI‑generated content is necessary to mitigate the issue.