📝LinkedIn's feed in 2026 is no longer ranked by your network — it's ranked by 📝360Brew, a 150-billion-parameter LLM that infers what each member actually wants to see. The signals that worked in 2025 have been re-weighted, demoted, or replaced. Here are the seven shifts that matter most.
1. Network logic became interest logic
LinkedIn replaced its multi-source retrieval system with 360Brew, a decoder-only foundation model built on Llama 3 that represents every member and every post as a vector in a shared semantic space. The feed distributes content by inferred topical interest, not first-degree proximity. A relevant post from a stranger now outranks an irrelevant post from a connection. Personalization works from day one without engagement history because the model reads meaning, not behavior.
Sources: Search Engine Land — LinkedIn updates feed algorithm with LLM-powered ranking · Falia — 360Brew Explained 2026 · ALM Corp — LinkedIn Feed Algorithm Update 2026
2. Dwell time is a top-three signal
Posts that hold attention for 61+ seconds achieve roughly 15.6% engagement; posts scanned in under three seconds drop to 1.2%. Dwell time now correlates with engagement at more than ten times the rate of likes. Long-form, scannable, value-dense content beats short hot takes — not because length is virtuous, but because attention is the metric.
Sources: meet-lea — LinkedIn Algorithm Explained 2026: Dwell Time, Comments & Engagement
3. Document carousels lead, video reach is collapsing
Document/PDF carousels average 6.60% engagement, the highest of any format. Native video averages 5.60% but reach is down 36% year-over-year as LinkedIn quietly throttles autoplay video in the feed. Text-only posts average around 2%. Posts containing external links lose 18 to 60% of their reach; placing the link in a comment now reduces visibility up to 80%.
Sources: Dataslayer — LinkedIn Algorithm 2026: What Works Now · The State of Brand — LinkedIn Is Quietly Killing Video Reach · CarouselMaker — Carousels vs Text Posts vs Videos 2026
4. Comments are scored on length and substance, not count
Comments of 15+ words carry 2.5x the algorithmic weight of short ones. Generic replies like "Great post!" or "So insightful!" are now classified as engagement noise and can actively suppress reach. The 2025 shortcut of farming low-effort comments has reversed into a penalty.
Sources: meet-lea — LinkedIn Algorithm Explained 2026 · HypergrowthAI — 5 Steps to 10x Your LinkedIn Reach in 2026
5. Hashtags are deprecated as a primary signal
Under 360Brew, semantic embeddings already infer topic. Hashtags now play a minor amplification role — zero to three is optimal, and hashtag stuffing is penalized as spam-adjacent behavior. Topical keywords inside the post body matter more than hashtags appended to the bottom.
Sources: Sprout Social — LinkedIn Hashtags: How to Use Them in 2026
6. The first 60 minutes are the test window
LinkedIn shows new posts to 2–5% of your eligible interest graph in the first hour. Engagement velocity inside that window determines whether the post is expanded or buried. The 2025 "self-comment within 10 minutes" trick still works, but the real gate is the full first hour — dwell time, substantive comments, and saves inside that window decide everything.
Sources: Growleads — LinkedIn Algorithm 2026: Why Your First 60 Min Decide Everything
7. Topic consistency builds creator authority
360Brew reads your post history as a chronological sequence and infers topical authority across it. Creators who post consistently within a defined cluster of topics get amplified within that interest graph. Bouncing across unrelated topics dilutes the model's confidence in your expertise — every off-topic post measurably costs reach on your next on-topic post.
Sources: Essey Marketing — What Is LinkedIn 360Brew? · pettauer.net — LinkedIn 360Brew and the New Physics of Visibility
What This Means In Practice
The 2025 playbook — post often, drop hashtags, ask for comments — is neutral at best and penalized at worst. The 2026 playbook is narrower: pick a topic lane, write content that holds attention for a full minute, invite long substantive comments, and treat the first hour as the entire test.