Over the past few months, Linkedin has been filled with posts that seem to be written by the same voice. Inflated reflections, repeated structures, supposedly inspirational messages and comments that reformulate the original post without contributing anything new. The professional platform owned by Microsoft has ended up recognizing the problem and is now announcing measures to reduce the visibility of the so-called “AI slop”: content generated with artificial intelligence perceived as generic, automated or lacking original criteria.
Linkedin claims to have developed systems capable of detecting this type of publications and comments using an approach defined internally as “AI solving AI”, that is, artificial intelligence trained to identify patterns typical of other content generated by AI. According to Laura Lorenzetti, the company’s Vice President of Product, the objective is not to prohibit the use of generative tools, but to limit the expansion of publications that “they do not provide perspective, experience or original thinking.”
The platform claims that in its first internal tests the system was able to identify generic content with an accuracy of 94%. However, it has not shared data on false positives or technical details about how the model exactly works. Detected posts will not be deleted either: they will simply stop being algorithmically boosted in recommendations and amplified feeds. They will remain visible to direct contacts and followers of the author.
Linkedin especially targets three types of content:
- Auto-generated posts with repeating formulas
- Automated comments created to artificially increase engagement
- Videos designed solely to capture attention through patterns of engagement bait
Among the examples cited by the platform itself are type structures “It’s not X, it’s Y.”increasingly frequent in publications generated or assisted by AI. This is a rhetorical formula known as emphatic epanortosis, based on correcting or reformulating a first statement to intensify the second (“It’s not networking, it’s community building”; “It’s not marketing, it’s culture”). Although it is a legitimate language resource, Linkedin considers that its massive and automated repetition has become one of the most recognizable patterns of empty content optimized for engagement.
Editors and moderators are labeling thousands of posts as original or generic to train systems
The company is also incorporating human oversight into the process. Editors and moderators are labeling thousands of posts as original or generic to train detection systems. In addition, Linkedin analyzes behavioral patterns, publication speed, linguistic structures and comment volume to identify mass automations or artificial interaction networks.
The decision reflects the extent to which the proliferation of AI-generated content is beginning to be perceived as a problem for social platforms. According to a Graphite report corresponding to the first quarter of 2026, the volume of articles generated by artificial intelligence published on the internet already equals that of those written by humans. In parallel, various technological platforms are developing identification and traceability systems, such as SynthID watermarks or C2PA standards driven by OpenAI.
In the case of Linkedin, the challenge is especially delicate due to the very nature of the platform. Its algorithm prioritizes familiarity, interaction and perception of professional authority. And that has turned the feed into fertile ground for publications optimized to generate rapid reach through emotional or pseudo-expert formulas easily replicable by generative models.
The paradox is that Linkedin belongs to Microsoft, the main investor in OpenAI and one of the companies that are most intensively integrating generative AI into professional products. The platform itself offers writing assistants to write posts, improve profiles, generate comments or help candidates and advertisers. In practice, Linkedin is simultaneously building the tools that facilitate this type of content and the systems intended to limit its expansion.
The company also recognizes that the rollout will be progressive and could take several months to be clearly perceived in the global feed. Meanwhile, the measure also functions as an implicit recognition of an increasingly evident phenomenon: the massive automation of professional discourse begins to deteriorate the differential value of platforms based on criteria and experience.
