LinkedIn uses LLM profile summaries to surface posts from outside your network.

LinkedIn says a new AI retrieval system, Profile Relevance, now helps choose which posts from people outside a member's network appear in their feed. It summarises each member's public profile and matches it against posts using LLM embeddings.
The system is described in a LinkedIn Engineering blog post published on 9 November, written by Shiyong Lin and 12 colleagues. Each member's summary draws on their headline, profile summary, skills, work history and education; LinkedIn says name and gender are left out. Posts are turned into embeddings from their text, titles and transcripts, plus author details such as location where available. A fine-tuned large open-source model pulls the top 10 topics from profiles, and a fine-tuned small open-source embedding model places profiles and posts in the same space.
Two changes made it work in the feed. LinkedIn added the member's and author's country, taken from the location on their profile, so members are more likely to see posts about their local professional context. It also set a minimum engagement threshold: a post from this source needs at least one like, comment or share from an unconnected member before it is recommended. In online experiments LinkedIn recorded a 0.54% lift in feed updates viewed, a 0.94% rise in viral rate and an 18.8% fall in hide rate. For members with little activity history, viral engagement rose 17.7% and daily active users 1.8%.
The figures are LinkedIn's own. The post does not say what share of the feed this source supplies, when it reached all members, or whether it affects ads or sponsored creator content.
Our read: for B2B marketers and executives in Saudi Arabia and the UAE, the profile now does more of the work. A clear headline, skills and role history help decide which unconnected posts a member is shown, and the country signal favours regional content. For brand and employee posts hoping to travel beyond followers, an early reaction from an unconnected member now counts before LinkedIn widens the audience.
Why it matters
Reach beyond followers on LinkedIn now depends partly on how well a post matches members' profiles, location included, and on early engagement from an unconnected member.
Related trends.

YouTube starts adding estimated TV co-viewers to public view counts
YouTube is rolling out a change that adds an estimate of people watching together on a TV to the public view count under its videos. One playback on a television can now register as more than one view.
Why it mattersThe most visible number on YouTube now mixes counted plays with a modelled TV audience, which changes how creator rates and campaign reach should be compared.

Google tracking code now flags AI Max clicks on exact keywords
Google Ads has added a ValueTrack parameter, {search_term_matchtype}, that records whether a search reached an ad through the advertiser's own keyword or through AI Max. It closes a gap in which AI Max expansions carried the keyword's match type.
Why it mattersWithout it, AI Max expansions looked like ordinary exact or phrase matches in click data, hiding how much traffic Google's AI was adding beyond advertisers' keywords.

LinkedIn demotes AI-generated 'slop' comments with new classifiers
LinkedIn is rolling out classifiers that push suspected AI-generated comments down its "most relevant" comment ranking. Chief product officer Hari Srinivasan says tests cut views of such comments by more than 20%, with a wider rollout coming soon.
Why it mattersAutomated comments have been a cheap way to look active on LinkedIn; ranking them lower makes genuine engagement by people and brand teams count for more.