SEO

LinkedIn's Profile Relevance Uses LLM Embeddings to Surface New Voices

LinkedIn has described a system called Profile Relevance, which uses LLM-generated embeddings to show users relevant posts from people they do not follow. It is a retrieval source for the feed, not a ranking tweak, and it means a good post can now reach a reader who has never heard of its author.

How Profile Relevance works

LinkedIn engineer Shiyong Lin explained the system in a post on the LinkedIn Engineering Blog, published on November 9, 2026. The process starts by summarising each member's profile into a prompt, taking into account profile information and posting history. Using those generated summaries, LinkedIn built a recommendation system that measures how well each post matches a member's professional context, across factors including topic, location, job title, company, industry, seniority and education.

The summaries are turned into embeddings, which are numerical representations that let the system compare a member and a post in the same space. When the embedding for a post sits close to the embedding for a member, the post is a candidate for that member's feed, even if the two have no connection. LinkedIn assessed the results manually before building the system that now shows users more relevant posts from members they do not follow.

Why embeddings beat keywords here

The advantage of an LLM embedding is that it understands professional context rather than matching words. A traditional system might notice that a member lists electrical engineering and then show them posts containing that phrase. An LLM-based system understands the deeper relationships, such as that an electrical engineer who engages with posts about grid optimisation may also care about renewable energy infrastructure and small modular reactors, even if no engagement history yet confirms it.

That matters most for cold-start members, people with limited activity history on the platform. A new member with only a profile headline and a job title gives a keyword system almost nothing to work with. The LLM can still build a useful picture from the profile alone, because its training gives it world knowledge about what people in a given role tend to care about. For members with limited historical activity, LinkedIn says the system delivered a 17.7 percent increase in viral engagement and a 1.8 percent lift in daily active users.

What the experiments measured

LinkedIn ran online experiments before rolling the system out, and the numbers are modest but consistent. Feed Updates Viewed rose 0.54 percent, viral rate rose 0.94 percent, and the hide rate fell 18.8 percent, which the company reads as a more relevant and less noisy member experience. A falling hide rate is the most telling of the three, because hiding a post is the clearest signal a user can send that the feed got it wrong.

There is one important condition on how posts get picked up. A post recommended through Profile Relevance must already have received a like, comment or share from an unconnected member. That first outside interaction is the gate: it proves the post has travelled at least one step beyond its author's network before the system will carry it further. One outside interaction does not guarantee wider distribution, and the post still has to pass the ranking stage that decides what a member actually sees.

Retrieval is not ranking

LinkedIn's broader feed explanation, published in March 2026, separates the process into two stages. Retrieval finds candidate posts, and ranking determines what a member actually sees, using professional context and patterns in their interactions over time. Profile Relevance is a retrieval source. It decides which posts are eligible to be considered, not which ones win the feed. A post that is retrieved but poorly ranked still will not be seen.

That distinction is worth keeping in mind when reading coverage of the announcement. The November 9 post is a technical account of one discovery system, not an announcement that the entire feed has just changed. The feed has been rebuilt around LLMs and sequential transformers over the course of 2026, and Profile Relevance is one piece of that rebuild, the piece that finds posts from beyond a member's network.

What it means for SEOs and link builders

For creators, the message is that reach on LinkedIn is no longer capped by follower count in the way it used to be. A post that earns one genuine interaction from outside your network can now be retrieved for strangers whose profiles match its subject, and the system is explicitly designed to surface high-quality content beyond a member's immediate network. That rewards posts with a clear professional angle, a specific audience, and language that matches how that audience describes its own work.

For link builders, LinkedIn is becoming a discovery channel for expertise, not just a distribution channel for links. A sharp, original post that earns outside engagement can put its author in front of exactly the people who cite, hire and link to experts in that field. The practical move is to write posts that a stranger in your target profession would find useful on their own, with the link as a secondary step, and to make sure your profile itself is complete and specific, because the profile is now part of the matching system that decides who sees your work.

Sources

Sources: Keywords Everywhere, November 9, 2026.

Related reading on this site: AI Assistants Hit 76 Billion Monthly Sessions, Graphite Estimates, SEO Split Testing in 2027: How to Prove a Change Actually Worked, Cloudflare CEO: AI Agent Optimization Is the New SEO, ChatGPT Sends Just 0.62% of Google's Organic Traffic, Google's Search Volume Is About 5x ChatGPT's, OpenAI Says, and OpenAI Launches Businesses in ChatGPT With Free Profiles and AI Agents.

More from Linkbaits

Turn this playbook into links

Linkbaits helps you create content that earns citations and backlinks. Start with the toolkit and build your first linkable asset this week.

Start free trial