LinkedIn and X are full of AI spam
From a sample of over a million posts across various platforms, a significant portion is AI-generated. Long LinkedIn posts are most affected.
Over a third of the content on LinkedIn is said to be AI-generated. The proportion is also continuously growing on other platforms. Longer posts, in particular, are apparently less and less frequently written by human hands. This is reported by the technology company Pangram, which specializes in the detection of AI content.
The data comes from Pangram's browser extension for Chrome. This scans social media feeds as you scroll through them and flags content that is likely AI-generated. Users can anonymously share their statistics with Pangram via opt-in. The company has now published the figures from the first few months. Just over one million posts with at least 50 words on LinkedIn, X, Reddit, Substack, and Medium were analyzed.
The accuracy of AI detectors is controversial because they sometimes mistakenly identify human texts as AI-generated (false positive). However, the Pangram 3.3 software used achieves a false-positive rate of only 0.01 percent, according to the company. Even if it were many times higher, it would only make a small difference for such a quantitative analysis.
The most important findings:
- The longer a post, the higher the probability that it was written with artificial intelligence.
- One in four long posts (over 250 words) is completely AI-generated. On LinkedIn, the rate is even over 40 percent.
- Almost half of all posts on X are at least partially written with AI.
- Replies are less frequently generated with AI than top-level posts.
A system that goes in circles
It is astonishing that so many people are willing to let an AI speak for them on LinkedIn in a professional context and under their real name. One might expect the hurdles to be higher here than on anonymous platforms. But the opposite is true. One reason for this is likely LinkedIn's built-in AI writing assistants, which make the use of language models very convenient.
It makes sense that the proportion of completely AI-generated posts is lower for short-form content. A short thought is formulated more quickly than a complex argument. The AI proportion is lowest on Reddit and Substack. Even there, however, it reaches over ten percent for long posts.
This development is not only problematic because the smoothed-out AI slop is a nuisance to read. Platforms like Reddit, on the other hand, serve as data sources for training AI models – a system that is increasingly going in circles. The hallucinations inherent in language models could therefore become more frequent. According to studies, 35 percent of all newly created websites are also AI-generated.
My fingerprint often changes so drastically that my MacBook doesn't recognise it anymore. The reason? If I'm not clinging to a monitor or camera, I'm probably clinging to a rockface by the tips of my fingers.
From the latest iPhone to the return of 80s fashion. The editorial team will help you make sense of it all.
Show all
