The LinkedIn algorithm in 2026: what actually changed under the hood, and how to adapt your B2B strategy

Why we're writing this now

The signals were there since late 2025, but two March 2026 publications made the scale of the change clear. First: an arXiv paper from LinkedIn's team describing 360Brew, a 150-billion-parameter language model built to replace the 14-15 historical ranking systems. Second: an official post by Tim Jurka, VP Engineering at LinkedIn, who spent 13 years building the feed, explaining that the rebuild is done and the 'current' feed already runs on it.

Meanwhile, the reach numbers creators see are brutal. Van der Blom measures -60% cumulative reach over two years for active creators, with -50% views YoY, -25% engagement YoY, and -59% follower growth YoY. This isn't a bug, it's what a semantic-match-first ranker does to generic content.

The question we hear every week from B2B clients: 'should we abandon LinkedIn?' The honest answer is no, because as average reach collapses, the quality of the reach you do get is skyrocketing. Sabrina Chevannes puts it better than we can: if you reach 100,000 people and nobody buys, that's not success. If you reach 20 and every one becomes a client, that's a business. That's the real thesis of the new LinkedIn.

What changed under the hood: 360Brew

The old feed worked as a feature factory: each content type (chronological, trending, hashtags, second-degree connections, followed topics) went through its own ranker, then the outputs were merged. Result: what a post 'had to do' to break through varied depending on which channel distributed it.

360Brew fuses all of that. A single model, trained on Meta Llama 3 then fine-tuned on LinkedIn's own feed data, reads content and profiles the way a human would. It no longer matches keywords, it matches semantic embeddings. It no longer sorts by recency, it sorts by relevance to each viewer individually.

What that changes concretely, per the technical breakdown on LinkedIn Engineering's blog:

1. Your profile becomes a prompt. The system reads your name, headline, company, industry, and title as five semantic fields that define 'what you speak on with authority.' Other fields (skills, work history, certifications, languages) are used on the viewer side to understand what the person wants to see.
2. Your activity history is treated as a time-ordered sequence. The model looks at what you've posted, liked, commented on, and saved over the last 90 days to understand your trajectory.
3. Your post gets embedded. Text, images, format are all transformed into a semantic vector.
4. Matching runs on embedding similarity, not keywords or hashtags. A viewer who has shown interest in 'operational efficiency in mid-market manufacturers' will see your post even if you never used those exact words.

The measurable result in LinkedIn's own A/B test: smaller accounts (fewer connections) gain +3.29% revenue and +1.17% pro interactions. Large accounts leaning on brute-force reach lose ground. Practical translation: your Quebec or US SMB with 800 followers is structurally less disadvantaged than it was in 2024.

The first 60 minutes decide everything

Every 2026 source converges on this: LinkedIn's decision window is now about 60 minutes post-publish. Practitioners call it the 'Golden Hour.'

The mechanism, per Digital Applied's decomposition of LinkedIn's Depth Score: your post is first tested on 2-5% of your network. If the signals in the first 60 minutes are strong (dwell time high, substantive comments, saves, reposts with commentary), the post gets amplified to second- and third-degree connections. If not, it dies quietly. Dataslayer notes that only 5% of underperforming first-hour posts recover.

What to do concretely in that window, aggregated from what Matt Gray, Lara Acosta and Tommy Clark do on their clients:

Your profile counts for 50.1% of ranking

The most counterintuitive number in van der Blom's 2026 report: more than half of a given post's ranking comes from who's posting, not what's posted. Post performance itself counts for 29.5%. The remainder is viewer context.

And Diandra Escobar clarifies the mechanism: only 5 profile fields travel with your content and serve the semantic match. Your name, headline, company, industry, and title. That's it. If your headline says 'thought leader and change maker,' your content has no semantic anchor. The system can't match you to anyone because it has no idea what you do.

What we rewrite on every B2B client before any LinkedIn campaign in 2026:

• Headline: the words your ideal client would use to describe what they're looking for. Not your internal job description. If you sell cybersecurity consulting to mid-market manufacturers, your headline says 'cybersecurity consultant for mid-market manufacturers,' not 'Senior Security Advisor.' The system matches on the words your prospects type, search, and read.
• Consistent company + industry. If you're a freelancer/consultant, your 'company' should reflect your niche, not just your name. If you work in a multi-service SMB, pick the industry that matches your primary content.
• Job title with words a searcher (or an LLM) would understand. The 150B-parameter model was trained on natural language, not internal jargon. 'Head of Business Development' is better understood than 'BDL2.'
• An About section that describes a specific client problem + how you solve it. This field isn't in the 5 that travel with content, but it's massively used on the viewer side when someone clicks your profile after seeing your post.

The first 45-50 words are your algorithm audition

This point is new in 2026 and not yet widely understood. Diandra Escobar pulled it from LinkedIn Engineering's own breakdown: the retrieval system (the step that decides whether your post enters the candidate pool before being ranked at all) truncates your post text at the first 60 tokens. In English, that's roughly 45 to 50 words.

In other words: if your hook (the first 2-3 sentences) has no strong semantic signal matching a viewer's interest, your post doesn't even enter the race for the Golden Hour. Anything past those 50 words is only used at the ranking stage, once you're already admitted to the pool.

What that changes concretely about how to write B2B post openings:

• The hook must contain the words your ICP uses. Not vague words like 'innovation' or 'leadership.' Real words like 'stalled pipeline,' 'rising CAC,' 'junior sales onboarding,' 'ERP migration,' depending on your actual ICP.
• The hook must pose a resolvable tension in 45-50 words. 'I lost 42% of my pipeline in 2025. Here's the attribution mistake we fixed in 30 days' beats 'This week I've been thinking about how important measurement is in B2B marketing where attribution is hard.'
• Avoid emojis in position 1-2. The tokenizer counts them as wasted tokens that eat your precious 60.
• Avoid 'drum roll' openings like 'Yesterday, something crazy happened to me...' that have zero semantic signal in the first words.

Personal vs Company Page: the gap is 5-8x

The 'personal profile vs company page' debate is settled for 2026, and not with the nuance we'd have hoped. Digital Applied consolidates 30+ studies: personal profiles outperform company pages by 5-8x on engagement. Van der Blom (via CMI) measures that company page content appears in 2% of feeds, versus 31% for top personal creators. ZoomSphere corroborates the 2%. LinkedIn acknowledges it in its own Business Blog: 'employees' combined networks are approximately 12x larger than a company's direct following.'

The nuance that matters for an SMB: the gap narrows on video format. Video published from a company page reaches about 2-3x less than personal video, not 5-8x. So if you're committed to publishing from the page for brand reasons, video is the least-penalized format.

The real strategy that works for our clients: stop pushing the company page as your primary channel, and build an ecosystem of 3-5 employees who each publish 1-2x per week from their personal profile. Tommy Clark has an interesting rule: if you get 3-5 active people from your company posting on LinkedIn, you're in the top 0.1% of companies. We validated this with three B2B clients in Q1 2026: total organic reach for the brand (via employees) went from ~2% of network to 12-15%.

What we stopped doing at Market Edge in 2026

Two tactics we still ran in 2025 are now actively counterproductive, and we've cut them across every client:

Engagement pods. LinkedIn made an official announcement on March 12, 2026: 'we are working to make engagement pods ineffective and curb comment automation.' Sabrina Chevannes confirms that pod users are 'actively being punished' by 360Brew. Mechanism: the model detects inauthentic engagement patterns (same 15 people commenting on every post within 5 minutes, without ever commenting on other posts by the same author) and demotes the account instead of amplifying it.

Hashtag-heavy strategy. Van der Blom measures that posts without hashtags now outperform posts with hashtags by 5 to 10%. Beyond 10 hashtags, the visibility penalty climbs to 30-50%. Mechanism: 360Brew reads content semantically, it doesn't need hashtags to classify. Hashtags add tokenized noise without useful signal. We keep 0-2 hashtags maximum, only if they're truly thematic (not generic tags like #marketing #b2b #growth).

What we're still testing:

• LinkedIn's active detection of generative AI. ZoomSphere says 360Brew doesn't technically detect AI - it detects nobody finishing the read (dwell time). Rewarx (an anti-detection tool vendor) claims there are active classifiers on 'sentence uniformity, vocabulary diversity, punctuation patterns.' Our stance: we don't take the risk. Use AI to brainstorm and structure, never publish without human rewriting.

What the real practitioners are saying

We pulled these posts from creators we read every week at Market Edge. None of them sell a miracle course; they operate accounts and publish results. Swipe to browse.

The underused long-term format: native LinkedIn newsletter

A puzzle piece we don't see discussed enough. LinkedIn lets you launch a native newsletter from your personal profile. Each issue sends a notification to subscribers (~40-60% open rate on our clients, comparable to a well-run SendGrid).

Two structural advantages regular posts don't have:

1. Reinforcement of the author embedding. Diandra Escobar: 'LinkedIn newsletters create a recurring engagement pattern that can strengthen your member embedding over time.' Every opened and read issue enters the 90-day signal the system uses to match you.
2. Owned distribution, not rented. Van der Blom: 'my Sunday newsletter has the highest conversion rate of anything I publish.' Chris Donnelly talks about 'deplatforming your audience' - the newsletter is your insurance if LinkedIn changes cadence again in 6 months.

We recommend monthly newsletters for clients starting out (sustainable cadence), then bi-weekly if it fits. What works: one dense theme, 800-1200 words, a proprietary number, a lesson, a CTA to your next step (usually the company newsletter, not a sales page).

Verdict and where to start

If you only do three things this week, do these, in this order:

1. Rewrite your LinkedIn headline with the words your ideal client would use to describe their problem. Not your internal title, not 'passionate about...', not 'helping companies...'. The exact words someone would type into a search bar. This field is one of the 5 that travel with your content and weighs into the 50.1% profile-based ranking share.

2. Take your 3 best posts from the last 12 months and look at the saves/likes ratio. If your saves are under 5% of your likes, your content is too motivational and not tactical enough. Saves are worth ~5x likes in amplification signal. Shift to operational content: numbers, breakdowns, process, specific use cases.

3. Identify 2-4 people in your company who will each publish once a week. The 3-5 employee ecosystem is what puts you in the top 0.1% of companies on LinkedIn (Tommy Clark). Help them rewrite their headlines. Combined organic reach is 5-8x that of your company page and 12x your direct audience (LinkedIn Business Blog, official number).

The bigger frame: LinkedIn in 2026 is no longer a reach platform, it's a semantic-matching platform. The question to ask before every post is no longer 'how many people will see this,' it's 'does 360Brew have enough signal to understand who this is for.' If yes, reach follows. If no, no viral hook will compensate.