Justine NamourCTO & Co-founderIn 2026, the LinkedIn algorithm retrieves posts with a language model that matches members and posts by meaning, then ranks them with a Generative Recommender that predicts clicks, skips, long dwells, likes, comments and shares. LinkedIn publishes no company-page penalty. Creators tend to travel further because their profile and posts match a clear topic and hold attention.
This post explains how the feed ranks content according to LinkedIn's own engineering publications, and what B2B teams should do about it. For the full mechanics of both the LinkedIn and X feeds, with every source dated, read Naano's report The LinkedIn and X algorithms in 2026, decoded.
How does the LinkedIn algorithm work in 2026?
LinkedIn's feed now works in four steps built around two models, described by its engineering team in March 2026 [LinkedIn Engineering, 12 March 2026]:
- Candidate posts. Millions of posts from the member's network and from outside it, refreshed within minutes.
- Language-model retrieval. A fine-tuned large language model, used as a dual encoder, turns each member and each post into an embedding and matches them by meaning. It replaced several separate retrieval sources, from a chronological network index to trending and collaborative-filtering systems.
- Generative Recommender. A transformer reads more than 1,000 of the member's past interactions as an ordered sequence, each post followed by what the member did with it, and predicts what the member will do next.
- The feed. Posts are ranked on those predicted actions.
What the retrieval model reads is text. For a post: its format, its author, its engagement counts and the post text. For a member: profile, skills, work history, education and the posts they engaged with. Engagement counts are converted into percentile buckets first; LinkedIn reports that this made the correlation between popularity and the model's similarity scores jump 30-fold [LinkedIn Engineering, 12 March 2026].
Posts a member saw but ignored are used as hard negatives, to teach retrieval what not to bring back. The approach is also described in LinkedIn's research paper on feed retrieval with causal language models [arXiv 2510.14223, October 2025].
What does the LinkedIn algorithm predict?
The ranker does not optimise for likes alone. It predicts two kinds of actions [LinkedIn Engineering, 12 March 2026]:
| Type | Actions predicted | What it means for a post |
|---|---|---|
| Passive | Click, skip, long dwell | The first line decides whether the reader stays |
| Active | Like, comment, share | Posts that invite a reply or a reshare score higher |
The skip matters most. Dwell time has been a ranking signal since at least 2020: LinkedIn defines a skip threshold below which a view counts as skipped, and lowers a post's score in proportion to its predicted chance of being skipped [LinkedIn Engineering, 12 May 2020]. In LinkedIn's words, time well spent is better than more time spent. LinkedIn dwell time for B2B creators covers what that means for writing.
Does the LinkedIn algorithm favour creators over company pages?
Not by a published rule. LinkedIn has not published a multiplier, cap or penalty that applies to company pages as such. Articles that quote a fixed reach ratio between personal profiles and pages are not citing LinkedIn's engineering publications.
What LinkedIn has published explains why personal creators usually do better in practice. This is Naano's reading of those sources, not a LinkedIn statement:
- The author is part of the match. Retrieval reads the post's author and the member's history. A creator who has posted about one topic for years gives the model a clear signal about who should see the post. A company page that posts about hiring, product and events gives a weaker one.
- The first line decides the skip. A practitioner opening with a scene or a claim holds attention. A brand announcement is easier to scroll past, and a predicted skip lowers the score.
- Active engagement is scored. Comments and shares count next to likes. Readers argue with people more readily than with logos.
LinkedIn also states that demographic attributes are not used to rank, and that it uses demographic data only to test that no group of creators or members is systematically ranked lower [LinkedIn Engineering, 20 November 2025].
What changed in the LinkedIn algorithm between 2020 and 2026?
| Date | What LinkedIn published | Why it matters |
|---|---|---|
| 12 May 2020 | Dwell time used to improve feed ranking, with a skip threshold | Attention, not just clicks and likes, enters the score |
| October 2025 | Research paper on large-scale feed retrieval using causal language models | Posts retrieved by meaning rather than by network alone |
| 20 November 2025 | How LinkedIn tests and measures what appears in the feed | Demographics not used to rank; used only for fairness testing |
| 12 March 2026 | Rebuilt feed: one language-model retrieval and a Generative Recommender | Several retrieval sources replaced; ranker reads 1,000+ past interactions |
The direction is consistent: the feed matches content to people by topic and ranks on predicted attention as well as engagement.
How does the X algorithm compare?
X takes a different route, and it publishes its For You feed code on GitHub under the Apache 2.0 licence [xai-org/x-algorithm, read October 2026].
| X | ||
|---|---|---|
| Where candidates come from | One language-model retrieval, in and out of network | Followed accounts, plus Phoenix retrieval, SimClusters and popular posts |
| Ranking model | Generative Recommender over 1,000+ past interactions | Phoenix, a transformer built on xAI's Grok |
| Negative signals | A predicted skip lowers the score | Not interested, mute, block, report and not dwelling carry negative weights |
| Freshness | Prompts and embeddings refreshed within minutes | Posts older than 48 hours are filtered out |
| Repeat authors | Not detailed in the 2026 publication | An author's later posts are discounted |
For brands running creators on both networks, the practical difference is time: on X, a post has 48 hours in For You, and one creator posting twice is discounted.
What should B2B brands do with company pages?
Keep the page, but stop treating it as the main distribution channel. A company page is what prospects, candidates and investors check before they engage, it is what appears when someone searches for the brand, and it is the publishing entity for LinkedIn's ad formats. A sensible cadence is one or two posts a week on announcements: launches, hires, milestones, customer news.
Route distribution through people instead. Three layers work together:
- Founders. A founder posting from a personal account about the problem the company solves gives the feed a clear topic signal. The limit is founder time. See founder-led distribution for B2B SaaS.
- Employees. Employees posting in their own voice about their work reach their own networks. Employee advocacy vs creator-led growth compares the two.
- Sponsored creators. Independent creators extend distribution into audiences your team does not reach. On Naano, each creator sets a fixed price per post, visible before booking; transacted medians ran from €84 per post under 5,000 followers to €588 at 25,000–50,000 [Naano Index snapshot, n=300, 11 August 2026]. Thought Leader Ads vs creator posts covers paying to boost personal posts instead.
How do you write a post the 2026 algorithm distributes?
Naano's report turns the published mechanics and its own data on sponsored creator posts into a short playbook [Naano, The LinkedIn and X algorithms in 2026, October 2026]:
- Cast for topic, not reach. Pick creators whose profile and past posts match your buyers' interests: that is what retrieval matches on.
- Write the first line for the skip. One idea, under 80 characters, no brand name.
- Tell a scene before the product. Let the brand arrive after the story, once the reader has decided to stay.
- Write in short lines. Short lines keep the eye moving, and that is what dwell measures.
- State, don't ask. Give readers something to agree or disagree with; comments are scored.
- One creator, one angle. Spread a launch across several voices rather than asking one creator to post twice.
- Publish on a working day and stay in the comments. The first replies are part of the distribution.
- Keep the post up and measure per creator. Naano creators keep sponsored posts live for at least six months, and every creator gets a tracked link.
See how to write a B2B sponsored post for the writing, and LinkedIn engagement rate benchmarks for what good engagement looks like.
How do you measure whether it worked?
Measure per poster, not per post. The same content performs differently from different authors, because the author is part of what the feed matches.
On Naano, each creator gets a tracked link, and Naano reports clicks, identified companies and leads per creator, to connect with your own conversion events. Naano does not publish a network cost per lead, cost per click or click-through rate: tracked-link coverage is too partial, with 62 of 89 completed bookings in the latest snapshot carrying any click event [Naano Index, August 2026]. How to measure ROI on B2B creator marketing gives the method.
FAQ
How does the LinkedIn algorithm work in 2026?
LinkedIn's 2026 feed works in two steps. A fine-tuned language model turns each member and each post into an embedding and retrieves posts by meaning, in and out of the member's network. Then a Generative Recommender, a transformer that reads more than 1,000 of the member's past interactions, predicts clicks, skips, long dwells, likes, comments and shares to rank them [LinkedIn Engineering, 12 March 2026].
Does the LinkedIn algorithm favour personal profiles over company pages?
LinkedIn has not published a rule or multiplier that favours personal profiles over company pages. What it has published is that posts are matched to members by meaning, using the post text, its author and the member's history, and ranked on predicted actions such as skips and long dwells. A practitioner writing about one topic tends to fit that model better than a brand announcement.
Does dwell time matter on LinkedIn in 2026?
Yes. Dwell time has been a LinkedIn ranking signal since at least 2020. LinkedIn defines a skip threshold below which a view counts as skipped, and lowers a post's score in proportion to its predicted chance of being skipped. The 2026 ranker predicts long dwell alongside clicks, likes, comments and shares [LinkedIn Engineering, 2020 and 2026].
Does LinkedIn use demographics to rank posts?
No. LinkedIn states that demographic attributes are not used to rank the feed. It uses demographic data only to test that no group of creators or members is systematically ranked lower [LinkedIn Engineering, 20 November 2025]. Ranking relies on the member's profile, skills, work history and the posts they engaged with, read as text.
What kind of post does the LinkedIn algorithm distribute in 2026?
Posts that earn attention before engagement. Because a predicted skip lowers the score, the first line matters most. In Naano's data on sponsored creator posts, the most-engaged posts open with a short statement, speak in the first person, are written in short lines and name the brand after the story [Naano, October 2026].
Work with the algorithm, not against it
If your company page has hit a distribution ceiling, route the message through people your buyers already read. Browse 3,300+ B2B creators on LinkedIn and X with their prices visible on the LinkedIn creator marketplace, or let the Naano team cast, brief and report with Managed Campaigns.
Related reading
- The LinkedIn and X algorithms in 2026, decoded
- Creator-led growth for B2B: the complete 2026 guide
- Why nano-creators outperform macro-creators in B2B
- LinkedIn Ads vs creator-led growth
Sources cited
- LinkedIn Engineering, "Engineering the next generation of LinkedIn's Feed", Hristo Danchev, 12 March 2026.
- LinkedIn Engineering, "Understanding dwell time to improve LinkedIn feed ranking", Siddharth Dangi, 12 May 2020.
- LinkedIn Engineering, "Putting members first: testing and measuring how content appears in your feed", Sakshi Jain, 20 November 2025.
- LinkedIn, "Large Scale Retrieval for the LinkedIn Feed using Causal Language Models", arXiv 2510.14223, October 2025.
- xai-org/x-algorithm, X's For You feed source code, Apache 2.0, read October 2026.
- Naano, The LinkedIn and X algorithms in 2026, decoded, October 2026.
- Naano Index snapshot, 11 August 2026.
CTO & Co-founder
Justine Namour, CTO and co-founder of Naano, builds the marketplace: booking, creator payments and tracked links reporting clicks, companies and leads.
Ready to try it
Run a creator-led growth campaign on Naano.
Book fixed-price LinkedIn creator offers without a platform retainer on Self-Serve. Compare eligible creators and track campaign outcomes.