On November 9, 2026, LinkedIn described in its Engineering Blog how the feed selects posts from people you don't follow. A language model converts your professional profile and the content of each post into numerical vectors, so-called embeddings. LinkedIn recommends posts whose embedding matches yours.
The method is called Profile Relevance and is added as another source alongside the existing signals. LinkedIn does not say for which countries and from when the change applies, not even for Germany.
What does the language model read from your profile?
For each member, LinkedIn builds a structured text from the professional information the member shares: headline, summary and skills, plus career stages and education. Name and gender are not included. For posts, LinkedIn analyzes text, titles and transcripts. In addition, LinkedIn appends the country of the member and the author to the texts, because profiles and posts often contain no location information.
There is also a threshold: According to the Engineering Blog, LinkedIn only serves a post via Profile Relevance once at least one unconnected member has liked, commented on or shared it.
LinkedIn only recalculates the profiles whose information has changed. As a result, the computing load is 24 percent of what recalculating all profiles would cost. Changes to your profile are included in the next calculation. LinkedIn does not say how quickly that happens.
What has improved according to LinkedIn's measurements?
Compared with the previous method MemberEBR, LinkedIn reports from online experiments 0.54 percent more feed updates displayed and a 0.94 percent higher viral rate. The hide rate, i.e. the share of hidden posts, fell by 18.8 percent.
Profile Relevance in Numbers
Measurements from LinkedIn's online experiments, compared with the previous method MemberEBR
- +0.54 %more feed updates displayed
- +0.94 %higher viral rate
- −18.8 %Hide rate: share of hidden posts, decreased
- +17.7 %viral interactions among cold-start members
Figures from LinkedIn itself, LinkedIn does not state absolute numbers
Source: LinkedIn Engineering Blog
Among cold-start members with little activity so far, viral interactions rose by 17.7 percent and the number of daily active members by 1.8 percent. All figures come from LinkedIn itself, and LinkedIn does not state absolute numbers. Whether a 0.54 percent increase in feed updates displayed will be noticeable for individual authors remains open.
What this means for you
Check the information in your profile that the language model analyzes: headline, summary, skills, career stages and education. Our assessment: the more precisely your field of expertise is stated there, the clearer your embedding is. LinkedIn does not make a corresponding recommendation.
Write posts so that the topic and industry are recognizable in the text. LinkedIn evaluates the quality of the matching based on topic, location, job title, company, industry, career level and education.
Before changing your profile, note the impressions of your recent posts and compare after two weeks. The comparison does not prove cause and effect, but you can see whether anything changes among people outside your network.
LinkedIn published the blog article on November 9, 2026.
Sources
