LinkedIn's algorithm reset in 2026: 10 changes that actually moved reach
Contents
- What actually changed, in one paragraph
- The replacement most people still describe wrong
- Dwell time now outranks everything you can fake
- Saves and private shares beat every public reaction
- Comments still carry weight, generic ones carry none
- The link penalty everyone quotes, and the number that held up
- Engagement pods: the detection is real, so is the 96 percent drop
- Hashtags stopped mattering, and LinkedIn said so itself
- Generic AI content gets suppressed, not AI use itself
- Document posts are the one format holding its ground
- Why the reach numbers contradict each other
- The four-week reset: rebuilding distribution from zero
- What I don't know
- Ten changes, one mechanism
- FAQ
- Sources
LinkedIn ran its feed on a new ranking engine for most of 2026, and the platform confirmed or enforced ten separate changes to how content gets seen between January and July. I tracked each one as it landed, cross-checked the numbers against three independent researchers and two of LinkedIn's own product leaders, and found that roughly half the figures circulating in marketing blogs don't agree with each other. I've been on LinkedIn since April 2004, watched three earlier algorithm overhauls get called the death of organic reach, and watched each settle into workable rules. This is that set of rules for the second half of 2026, sources attached, disagreements named rather than smoothed over.
In short: LinkedIn replaced its ranking engine with a 150-billion-parameter model called 360Brew in late 2025, and by mid-2026 the effects were measurable: dwell time and saves now outrank likes, engagement pods lose roughly 96% of their reach once flagged, generic AI content gets suppressed at a claimed 94% detection accuracy, and a link in a post costs somewhere between 19% and 60% of its distribution depending on whose dataset you check. The direction is consistent across every source that has published numbers. The size of the drop is not, and that gap matters more than any single statistic here.
What actually changed, in one paragraph
Ten changes shipped or became visible on LinkedIn between January and July 2026: a full replacement of the ranking engine, a shift to dwell time as the dominant quality signal, saves and private shares re-weighted above public reactions, a semantic read of comment quality, a harder penalty on outbound links, expanded detection of engagement pods, the quiet end of hashtags as a distribution lever, active suppression of generic AI-written posts, a widening gap between document-format performance and everything else, and a reach baseline that reset lower and stayed there. All ten came from the same engine, and reward the same behavior: content a real reader stops for, keeps, and forwards to someone specific.
The replacement most people still describe wrong
Most explainers call this "an algorithm update." It was not. LinkedIn retired its previous ranking infrastructure and replaced it with a single model, internally named 360Brew, reported at 150 billion parameters, deployed through the second half of 2025 and live by early 2026. An update tunes weights inside a fixed system. A replacement changes what the system can see.
The old stack scored posts largely on keywords, hashtags and early reaction counts. 360Brew reads a post the way an editor reads a submission: whether it matches the author's demonstrated expertise, whether the network being reached is likely to care, and whether the engagement it gets looks earned or manufactured. An industry breakdown of the rollout published in March 2026 described four alignment checks behind the score: profile coherence, network relevance, engagement pattern and content consistency over time.
For a founder or sales leader publishing two or three times a week, the model needs a consistent signal to calibrate against. A profile that jumps between recruiting posts and personal reflections gives 360Brew nothing stable, and that calibration decides whether your next post reaches anyone beyond existing followers.
Dwell time now outranks everything you can fake
Every 2026 analysis of the update agrees on one point: how long a real person spends on your post now matters more than whether they tapped a reaction button, a mechanic we go deeper on in our piece on dwell time distribution. LinkedIn measures this through two signals, time spent viewing a post on screen during a scroll, and additional time spent reading after someone taps "see more," neither of which can be gamed with a bot, precisely why LinkedIn built the update around it. Document posts benefit structurally: a reader swiping through slides registers as continued attention, the mechanical reason carousels now outperform every other format.
The lesson for a text-heavy post is blunt: a post skimmed in under three seconds contributes almost nothing to your distribution, no matter how many people saw it. A post that holds a reader past the first screen keeps compounding, because 360Brew treats that retention as evidence the content deserves a second look.
Saves and private shares beat every public reaction
Forbes contributor Jodie Cook, writing about the changes on July 23, 2026, put a number on what many operators had already noticed anecdotally: saves now drive roughly five times more reach than a like. A save tells the algorithm someone considered your post worth finding again; a public like costs the reader one tap and no thought.
Private shares, meaning a post forwarded by direct message, appear to carry even more weight, according to the same March 2026 industry breakdown of the 360Brew rollout, which places DM shares as the second most powerful signal in its hierarchy, behind only sustained dwell time. Neither figure is LinkedIn-confirmed; treat both as independent analysis, not official documentation. The full weighting hierarchy, and what to publish to earn both signals, is covered in our dedicated piece on saves and sends.
The design implication is specific: a post earns a save when it functions as a reference worth returning to, and earns a DM share when a reader can name the one colleague who needs to see it. Posts written to be liked and posts written to be kept are not the same posts.
Comments still carry weight, generic ones carry none
A comment of fifteen words or more, engaging with a post's actual substance, still moves distribution meaningfully. A one-word or templated comment contributes close to nothing, and 360Brew is built to tell the difference through semantic analysis of comment content, not comment count.
Timing compounds this. According to Richard van der Blom's 2026 Algorithm Insights research, cited in secondary analysis of the report, responding to comments within the first thirty minutes of posting earns roughly 64% more total comments and 2.3 times more views than responding later or not at all. An author who replies quickly keeps the thread active, itself a distribution signal, reason enough to block thirty minutes after a post goes live rather than checking back at day's end.
The link penalty everyone quotes, and the number that held up
This is where the sources stop agreeing, and I'd rather show you the disagreement than pretend it doesn't exist.
Jodie Cook's Forbes piece from July 30, 2026 states the penalty plainly: "Every time you put a link in a LinkedIn post, the algorithm cuts its reach by around 60%."
Richard van der Blom's research, built on a dataset of 1.3 million posts and reported by independent LinkedIn consultant Melanie Goodman in her analysis of the 2026 report, found something more specific and considerably smaller: one external link placed in the body of a post reduces median reach by 18.8%.
Both figures can be true at once: they measure different things. Cook's likely reflects the combined effect of a link plus what tends to accompany it, thinner captions, weaker hooks. Van der Blom's isolates the link itself, across a sample large enough to average out those confounds. A specific number from 1.3 million posts outweighs a rounder one from a single practitioner's observation, so I weight the larger sample: treat "roughly a fifth of your reach, not most of it" as the defensible planning number, and the 60% figure as a worst case, not a baseline.
What both sources agree on, operationally, is that placing the link in the first comment no longer avoids the penalty. LinkedIn's 2026 model reads what analysts call "bridge behavior": a post visibly constructed to push a reader toward a linked comment, penalized close to the same degree as a link in the body. The old workaround is dead; the honest fix is native formatting, upload the document, embed the video, or accept the reach cost when a link genuinely can't be avoided.
Engagement pods: the detection is real, so is the 96 percent drop
An engagement pod is a group of accounts that coordinate, often manually, to like and comment on each other's posts to fake early momentum. LinkedIn's VP of Product Management, Gyanda Sachdeva, laid out the company's position in a statement covered by Social Media Today in February 2026: LinkedIn would act against automated comments posted through third-party scripts or plugins operating without human oversight, stripping them from the "Most Relevant" section and, in repeat cases, restricting the accounts' reach.
Independent reporting on the crackdown, updated in May 2026, put LinkedIn's detection accuracy for coordinated engagement at 97%, built on three visible patterns: comment velocity (fifteen-plus comments within ninety seconds of publication), network analysis (the same small group engaging with each other's posts on a repeating schedule), and semantic analysis of generic, repetitive comment language. One documented case, cited across several outlets, involved a marketing director whose average reach dropped from 8,500 impressions to 340 overnight, a 96% reduction, with no warning issued beforehand. Recovery is reported to require sixty to ninety days of clean posting behavior. I've never run a pod: the math has been bad since before 2026, trading real distribution for a coordination scheme a semantic model now reads as clearly as a human would.
Hashtags stopped mattering, and LinkedIn said so itself
LinkedIn removed hashtag following in 2024 and quietly renamed company-page hashtags to "Specialisms." What survived into 2026 is a search function, not a distribution lever: hashtags help a post surface in topic searches and give 360Brew one more signal about subject matter, but multiple 2026 analyses describe their effect on reach as roughly neutral once content quality and early engagement are accounted for. The guidance hasn't changed, it's just lost urgency: two to three specific, relevant hashtags at the end of a post is enough; optimizing a hashtag set in 2026 spends time on a lever that stopped moving.
Generic AI content gets suppressed, not AI use itself
LinkedIn made its clearest public statement on this on May 20, 2026, when VP Laura Lorenzetti drew a line worth keeping:
"It's ok to use AI to help you write, but your posts and comments need to represent your voice and your perspectives. The ultimate value comes from the human behind the tool."
She was more direct about what the company is trying to stop: "When AI is overused, especially at scale and in an automated way, it dilutes the valuable insights that real human conversations can spark." LinkedIn's detection claims 94% accuracy identifying generic, AI-written content, a figure reported independently in August 2026 coverage, though the company hasn't disclosed a matching false-positive rate. Flagged posts aren't removed, they're suppressed from LinkedIn's recommendation feed while staying visible to the author's direct followers: a softer penalty than it sounds.
Content creation on LinkedIn is up roughly 14% year over year, driven largely by AI-assisted writing: the suppression isn't aimed at the tool, it's aimed at output with no first-party data, no named case, no recognizable point of view. LinkedIn drew the same distinction a week later, adding a member-facing "Seems like AI slop" report button, a separate and more contested feature I broke down in a six-count case against the button itself: worth reading even though I agree with Lorenzetti's underlying diagnosis.
Document posts are the one format holding its ground
Format performance data published by Dataslayer, February 2026, shows a clear hierarchy:
| Format | Engagement rate (2026) | vs. text-only |
|---|---|---|
| Document post (PDF carousel) | 6.60% | +230% |
| Native video, 30 to 90 seconds | 5.60% | +180% |
| Image with text | 3.20% | +60% |
| Text only | 2.00% | baseline |
| Post with an external link | roughly 60% less reach than an identical link-free post | n/a |
The mechanism ties directly back to dwell time: a document post forces a swipe-through interaction that native video and static images can't replicate, and every swipe reads as sustained attention. For a B2B leader with limited production bandwidth, this is the format worth prioritizing for the second half of 2026, a five-to-eight-slide carousel converting one real number your prospects actually argue about beats a well-written paragraph post on distribution alone.
Why the reach numbers contradict each other
Here the numbers stop lining up, and pretending otherwise would be dishonest.
| Source | Metric | Reported figure |
|---|---|---|
| Van der Blom Algorithm Insights, via Dataslayer (Feb 2026) | Views | down 50% |
| Van der Blom Algorithm Insights, via Dataslayer (Feb 2026) | Engagement | down 25% |
| Independent 360Brew rollout analysis (Mar 2026) | Average visibility | down 47% year over year |
| Independent 360Brew rollout analysis (Mar 2026) | Engagement | down 39% |
| Independent 360Brew rollout analysis (Mar 2026) | Follower growth | down 42% |
Neither source published a full methodology I could check line by line. Different sample windows, different definitions of "engagement," and different creator cohorts (van der Blom's sample is broad; the second source doesn't specify its own) would each explain part of the gap.
What every source agrees on, without exception, is direction: reach and engagement dropped substantially and stayed down, rather than dipping and recovering the way algorithm updates typically do. That's the finding worth building a plan around, not the exact percentage; treat anyone quoting a single precise number without a source as guessing.
The four-week reset: rebuilding distribution from zero
This isn't a seasonal ritual. It's the sequence that works whenever a profile has gone quiet and needs to re-earn a stable signal from 360Brew, whose calibration window (roughly ninety days of consistent topic focus) makes this slow to skip.
Week one: narrow the signal. State one area of expertise in your headline and About section, not three. Publish one post drawn from something specific you observed, not a general opinion. Spend extra time on substantive comments, fifty words minimum, on posts from your actual buyer or peer group: that activity feeds your topical signal as much as posting does.
Week two: earn your first saves. Publish twice, including one document post. Every post needs one element worth bookmarking, a number or a comparison table, and every comment gets a reply within thirty minutes.
Week three: raise the cadence. Move to three posts, including one built around a specific, dated experience: a deal that fell through, a number you're not proud of, a decision you'd reverse. That kind of post reliably produces the substantive comments and DM shares that outweigh public reactions in this model.
Week four: check what worked. Hold at three posts. Pull your saves and DM-share estimates, not impressions, and identify the two or three topics that produced the most: those become next quarter's focus. Impressions tell you who scrolled past. Saves and private shares tell you who is planning to act.
The breakdown of the 360Brew model and the format data in the weekly All In newsletter go further than a single article can.
What I don't know
I don't know LinkedIn's actual false-positive rate for AI content detection. The company has published a 94% accuracy claim and no corresponding error rate, and I have no way to independently verify either number.
I don't know why the two reach-decline studies I found diverge by ten to twenty percentage points on similar metrics. I've named both figures and my reasoning for treating direction as more reliable than magnitude, but I haven't seen a methodology from either source detailed enough to reconcile them, and I'm not going to invent one.
I don't know how long the current reach baseline holds. Every past LinkedIn overhaul eventually settled, and operators who adapted early captured a disproportionate share of the readjusted distribution. Whether that pattern repeats this time isn't something the data available to me in mid-2026 can answer.
Ten changes, one mechanism
Ten separate announcements and detected patterns, and they all reduce to the same test: does a specific human reader stop, keep and forward this, or does it get scrolled past in under three seconds. Likes, hashtags, link tricks and pod coordination were shortcuts around that test. 360Brew was built to close them, and the sources that disagree on magnitude don't disagree on direction.
The operators losing the most ground are still optimizing for a version of LinkedIn that stopped existing in early 2026. The ones gaining ground publish less often and give each post something a reader would actually bookmark.
FAQ
Does LinkedIn still penalize links placed in the first comment instead of the post body?
Yes. LinkedIn's 2026 ranking model detects what analysts call "bridge behavior," a post structured to redirect readers to a linked comment, and applies a reach penalty close to what a link in the post body itself would cost. The old first-comment workaround no longer avoids the penalty.
How much does dwell time actually affect LinkedIn reach in 2026?
Every independent 2026 analysis of LinkedIn's update identifies dwell time, the real time a reader spends viewing and reading a post, as the dominant quality signal, ahead of likes and comment counts. LinkedIn measures it through on-screen viewing time and additional time spent after a "see more" click.
Are LinkedIn engagement pods still worth using in 2026?
No. Independent tracking of LinkedIn's crackdown puts detection accuracy at 97%, and documented cases show reach drops of roughly 96% once an account is flagged, with recovery reported to take sixty to ninety days of clean posting behavior afterward. The risk substantially outweighs any short-term boost.
Does LinkedIn's algorithm penalize AI-assisted writing?
Not directly. LinkedIn VP Laura Lorenzetti stated in May 2026 that AI-assisted writing is acceptable as long as posts represent the author's own voice and perspective. What gets suppressed, at a claimed 94% detection accuracy, is generic AI-written content lacking first-party data or a recognizable point of view.
What content format performs best on LinkedIn right now?
Document posts, meaning PDF carousels, led 2026 engagement benchmarks at roughly 6.60%, ahead of native video at 5.60%, image-with-text at 3.20% and text-only posts at 2.00%, according to Dataslayer's February 2026 analysis. The swipe-through format generates the sustained dwell time the current algorithm rewards.
How big is LinkedIn's actual reach decline in 2026?
Sources disagree on the exact figure. Richard van der Blom's Algorithm Insights research points to views down roughly 50% and engagement down 25%; a separate industry analysis of the 360Brew rollout reports visibility down 47% and engagement down 39%. Every source agrees the decline is real and sustained, not a temporary dip.
Do hashtags still help LinkedIn posts get seen?
Barely. LinkedIn removed hashtag following in 2024, and 2026 analyses describe hashtags as having a roughly neutral effect on reach once content quality is accounted for. They still help with on-platform search. Two or three relevant hashtags is enough; more than that has no measurable benefit.
Sources
- UpGrowth, "LinkedIn Algorithm 2026 Explained: What 360Brew Means for Reach & Growth", March 2, 2026, upgrowth.in: 360Brew parameter count, engagement signal hierarchy, reach and engagement decline figures.
- Jodie Cook, Forbes, "The LinkedIn Link Penalty Cutting Your Reach By 60%", July 30, 2026, forbes.com: link penalty figure and mitigation tactics.
- Jodie Cook, Forbes, "5 LinkedIn Content Moves LinkedIn Started Punishing In 2026", July 23, 2026, forbes.com: saves-versus-likes weighting, pod and comment-bait suppression.
- Social Media Today, "LinkedIn wants to limit the reach of AI-generated content", May 21, 2026, socialmediatoday.com: Laura Lorenzetti statements and AI-content policy.
- Social Media Today, "LinkedIn Outlines Measures To Combat Engagement Pods", February 16, 2026, socialmediatoday.com: Gyanda Sachdeva statements on automated comment detection.
- ConnectSafely, "LinkedIn Engagement Pods Crackdown 2026: Complete Guide", updated May 16, 2026, connectsafely.ai: 97% detection accuracy, documented reach-drop case, detection patterns.
- Neil Patel, "LinkedIn Is Suppressing AI Slop: What That Means for You", August 5, 2026, neilpatel.com: 94% detection accuracy figure, 14% year-over-year content creation increase.
- Dataslayer, "LinkedIn Algorithm 2026: What's Working Now", February 12, 2026, dataslayer.ai: format-by-format engagement rates, citation of van der Blom views and engagement decline figures.
- Melanie Goodman, "LinkedIn Algorithm 2026: Why Your Reach Dropped (and 9 Data-Backed Fixes)", 2026, melaniegoodmanlinkedinconsultant.substack.com: van der Blom 1.3-million-post dataset, 18.8% single-link reach reduction.
- Richard van der Blom, Algorithm Insights Report, 2026, richardvanderblom.com: primary research referenced across items 8 and 9.
- ContentIn, "Do LinkedIn Hashtags Still Work in 2026? Data-Backed Insights", 2026, contentin.io: 2024 hashtag-following removal, neutral reach effect.
All In: decode the feed changes before you rebuild a content plan around last year's rules
Ten changes shipped inside one engine in 2026. The operators losing ground are still optimizing for the version of LinkedIn that existed before 360Brew.
All In is the B2B media that decodes LinkedIn: expert blog, weekly podcast and newsletter for SME leaders and sales directors who want to turn LinkedIn into measurable growth. An original creation by Patrick de Carvalho, on LinkedIn since 2004. Motto: "I Never Lose."
Discover All In and get what matters on LinkedIn, every week.