LinkedIn has rolled out a major overhaul of its feed algorithm, replacing its older engagement-driven ranking system with one powered by large language models that can understand what posts actually mean, not just which keywords they contain.
As explained by LinkedIn:
While the Feed has long been AI-powered, recent LLM advances gave us the opportunity to rethink what’s possible. That’s why we’re rolling out a new advanced ranking system, powered by LLMs and GPUs, that better understands what a post is actually about and how it relates to a member’s evolving interests and career goals.
In practical terms, if you happen to be interested in ‘electrical engineering’ but engage heavily with posts about ‘small modular reactors,’ traditional keyword-based systems might miss the connection. Our LLM-based retrieval understands these topics are semantically related because the underlying language model brings world knowledge learned from its massive pre-training corpus — it knows that electrical engineers often work on power grid optimization, renewable energy integration, and the infrastructure challenges.
The new system, built on a 150-billion-parameter foundation model, replaces what engineers previously described as a “feature factory” of separate task-specific models. Instead of calculating the likelihood of clicks and likes in isolation, the platform now reasons about user intent, matching content to professional interests through semantic understanding rather than keyword matching.
When industry news breaks, related posts now surface within minutes rather than hours. When a user signals a new professional interest, their feed reflects it almost immediately. And the system no longer evaluates posts in isolation. It cross-references content against the author’s profile, rewarding consistency between stated expertise and subject matter while suppressing mismatches.
The shift is already showing up in the data. Industry research indicates that overall views are down roughly 50% and engagement has dropped about 25% year on year. But this is redistribution, not decline. Visibility for recognized expert creators has risen from 15% to 31% since 2022, while visibility from other creators has fallen from 57% to 28%. The platform is deliberately concentrating reach among people who demonstrate genuine authority,
LinkedIn is also introducing a “Depth Score” that measures how long users actually engage with content, factoring in reading time, comment quality, saves, and private shares. Engagement bait is being actively suppressed. Posts asking for generic comments, videos unrelated to their text, and recycled thought leadership that lacks substance will all be downranked. Internal data reportedly showed that 60% of high-engagement posts in late 2025 relied on at least one manipulation tactic, even as user satisfaction was declining.
LinkedIn has announced that its revamped ranking system will give creators a better chance to connect with audiences. This new approach will keep up with the latest news, rather than just recycling older updates.
Document carousels are now the top-performing format at around 6.6% engagement, compared to roughly 2% for standard text. Video creation is growing twice as fast as other formats.
All in all, the message is clear: you now have to post less, say more, and prove you know what you’re talking about.
