Longer social media posts are increasingly likely to include AI-generated writing, according to a Pangram analysis covering five platforms. The clearest signal came from LinkedIn, where long-form posts were flagged at a higher rate than any other platform named in the source.
LinkedIn stands apart in long-form AI writing
Pangram’s analysis found that one in four social media posts over 250 words is AI-generated. That threshold matters because the study focused on longer posts, not every caption, reply, or short update a user might publish.
LinkedIn was the strongest example in the data. The platform had 41 percent of long-form posts flagged as AI-written. It made up only a third of all posts scanned, yet it accounted for nearly two-thirds of all detected AI content.
That gap is the central finding. LinkedIn was not just one of several platforms where AI writing appeared. In Pangram’s scan, it was disproportionately represented among the posts identified as AI-generated.
How the platforms compared
The source names several platform-level differences. X/Twitter showed a high rate in one specific format: close to half of long-form articles were AI-generated or AI-assisted. Substack sat at the other end of the comparison, with the lowest long-form AI rate at around 10 percent.
Reddit showed a more split pattern. Reddit replies were 98 percent human-written, but standalone posts contained AI text far more often. That distinction is important because it suggests that different post types can produce very different signals on the same platform.
The named platform findings can be summarized simply:
- LinkedIn: 41 percent of long-form posts were flagged as AI-written.
- X/Twitter: close to half of long-form articles were AI-generated or AI-assisted.
- Substack: the lowest long-form AI rate, at around 10 percent.
- Reddit: replies were 98 percent human-written, while standalone posts used AI text more often.
The source says the analysis covered five platforms, though not every platform is described in the provided article excerpt. Based only on the named results, LinkedIn is the most prominent case, while Substack appears to have the lowest long-form AI rate among the named platforms.
What Pangram measured
The data came from Pangram’s Chrome extension, which scanned over one million posts across five platforms between April and June 2026. The analysis focused on posts over 250 words, making it a study of long-form social media content rather than social media activity as a whole.
Pangram says its Pangram 3 detection model has a false positive rate of 0.01 percent. The source also notes that the model is likely better at identifying human-written content than AI-generated content. If that is the case, the real AI rate could be higher than the detected rate.
That caveat matters for interpreting the results. Detection tools do not turn a platform into a complete map of authorship. They provide a measured signal under a specific method, and the source frames Pangram’s results with that limitation in mind.
What the study does not claim
The study makes no claims about content quality. A post being AI-generated does not, from the source’s facts alone, prove that it is inaccurate, useful, low-effort, or harmful. The analysis is about whether text was flagged as AI-written, not whether the content was good.
Still, the platform patterns carry practical meaning. If a high share of long-form LinkedIn posts is AI-written, readers may become more aware of formulaic writing and more skeptical of professional-sounding posts. Writers and publishers may also face more pressure to make their work visibly specific, grounded, and human-edited.
The source also says LinkedIn itself seems to feel the pressure and has already started cracking down on AI-generated posts. No further details are provided in the source excerpt, so the strongest supported conclusion is narrower: LinkedIn is aware enough of the issue to act against at least some AI-generated posting.
Why this matters for readers
The Pangram analysis points to a broader shift in how long-form social media is produced and consumed. For readers, the key issue is not simply whether AI tools exist. It is whether posts that look personal, professional, or expert-led are actually written by people in the way they appear to be.
For platforms, the challenge is different. They need to decide how much AI-generated writing is acceptable, how to detect it, and how to respond without confusing AI assistance with low-quality content. The source does not answer those policy questions, but the LinkedIn figures show why the debate is not abstract.
For now, the clearest takeaway is factual: in Pangram’s scan of over one million posts between April and June 2026, long-form AI-generated social media was common, and LinkedIn led the named platforms by a wide margin.