Those LinkedIn thought leadership posts? Probably AI-generated
But AI detectors aren't infallible either.
Image by Cybernews.
- Pangram study found 40% of LinkedIn's long-form posts flagged as fully machine-written, accounting for nearly two-thirds of AI content across major platforms.
- LinkedIn's "Enhance post" feature and other AI writing tools have normalized AI-assisted publishing, making synthetic text mainstream.
- AI detection tools produce probabilistic estimates with high false-positive rates that disproportionately misclassify non-native English speakers as AI users.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
A new Pangram study has found AI text saturating social media, with LinkedIn as the biggest hotspot. However, critics warn that these findings must be taken with a pinch of salt, noting their high false-positive rates, especially when it comes to text written by non-native English speakers.
Pangram published the findings after analyzing more than a million posts across major social media platforms, including LinkedIn, X, Reddit, Substack, and Medium. LinkedIn accounted for nearly two-thirds of all posts Pangram classified as AI-generated, and more than 40% of its long-form posts were flagged as fully machine-written.
Pangram argued that LinkedIn's own AI writing tools, including its "Enhance post" feature, have helped normalize AI-assisted publishing on the platform.
The report also found AI-generated content across every platform it analyzed, suggesting that synthetic text has become a mainstream feature of online communication rather than a niche phenomenon.
AI detectors still struggle with accuracy
Pangram’s findings have revived a longstanding debate over the reliability of AI detection tools.
Researchers have consistently warned that AI detectors produce probabilistic estimates rather than definitive conclusions. Depending on the writing style, topic, and AI model involved, the systems can generate both false positives by incorrectly flagging human-written text as AI-generated and false negatives that miss AI-created content altogether.
Even AI companies have acknowledged the technology's limitations. OpenAI discontinued its own AI text classifier in 2023, citing low accuracy rates.
Stay updated with our latest stories and follow us on social media
Be the first to discover new stories, ideas, and updates from our team.
Several academic studies have also found that many AI detectors disproportionately misclassify writing by non-native English speakers. One widely cited study found that several leading AI detectors were significantly more likely to classify essays written by non-native English speakers as AI-generated than essays written by native speakers, even when both were entirely human-written.
The findings have raised concerns among educators and legal scholars because many schools and employers now rely on AI detection software when evaluating written work. Critics argue that false positives could unfairly penalize legitimate authors.
The debate extends beyond text detection.
A recent Reuters analysis of Meta's new image-generation model, Muse Image, exposed major gaps in visual verification. While the tool successfully identified 40 original AI-generated images, reducing the images to between one-third and one-half of their original size made Meta’s detector miss 55% of its own AI-generated pictures.
Pangram's report adds to growing evidence that generative AI is reshaping online communication. But as AI-generated content becomes harder to distinguish from human writing, researchers continue to stress that detection scores should be treated as indicators rather than proof.
For now, AI detectors can estimate the likelihood that content was machine-generated, but they cannot definitively answer the question.
