Curated posts vs. original content: what LinkedIn's feed rewards

LinkedIn's feed now runs on 360Brew, a 150-billion-parameter model deployed in March 2026 that reads what a post says, not just how many clicks it attracted. That shift has widened the gap between original and curated content. Personal employee posts generate nine times more engagement than company-curated content, and the data on where the reach ceiling sits is not close.

LinkedIn Original Content vs. Curated Posts: The Reach Gap, Quantified

The numbers are blunt. Personal employee posts generate 9x more total engagements than company-curated posts, averaging 62.74 engagements versus 6.94 for curated content. The breakdown: 9x more clicks, 8.8x more reactions, 17x more comments, per DSMN8's analysis of more than 500,000 LinkedIn posts. A company post that earns 7 engagements would need to become a personal employee post to earn 63.

The structural reason sits in how LinkedIn allocates feed space. First-degree connection posts occupy roughly 42% of visible feed slots. Company organic posts occupy 5.37%. A brand page sharing curated content starts at a distribution disadvantage before a single user has seen the post. That ratio, roughly one-eighth the feed allocation of personal connections, is a ceiling on reach that content quality alone cannot fix.

The gap has been getting worse. AuthoredUp's analysis of more than 3 million posts found median impressions per post dropped from 1,211 in June 2024 to 636 by May 2025, a decline of roughly 47%. That floor sits lower now than at any prior point in the data set, which means the absolute reach available to curated content is smaller even if its relative disadvantage stayed fixed.

Employee reposts of company content carry an estimated -30% reach impact compared to original employee-authored posts, per van der Blom's 2026 analysis of 1.8 million posts across 58,000 profiles and 31,000 company pages. This penalty applies to the repost mechanic specifically, not to content quality.

Put the figures together: company brand pages get fractional feed allocation, employee reposts carry a reach penalty on top of that, and the overall reach environment compressed by roughly half in the last year. These are not solvable by posting more.

Does LinkedIn's Algorithm Penalize Curated Posts or Reward Original Content?

LinkedIn does not publish an explicit penalty for sharing. That framing misses how 360Brew works. The model ranks posts with semantic novelty and profile coherence above repackaged or templated content. The outcome is a structural advantage for original writing, but the mechanism is scoring, not penalizing.

Profile coherence is the part most guides skip. 360Brew evaluates whether a post sounds consistent with the author's historical writing. An account whose posts consistently read like the same person accumulates a coherence score that broadens future distribution. Accounts that alternate between AI-template captions and authentic personal posts confuse the model and earn lower baseline reach over time. The cost of an inconsistent content strategy is not just one post that underperforms; it is a lower baseline on every post that follows.

That consistency matters especially for accounts that share curated content regularly. An account with a stable voice profile earns broader baseline distribution even on posts that are not exceptional, because the model has learned to trust its content type. An account with an erratic pattern earns lower reach even on its best posts. The direction of compounding depends on what the account consistently does.

A repost-with-thoughts reaches only 8 to 10% of a poster's typical audience without fast initial engagement. Crossing 10 or more engagements quickly pushes that to 15 to 20%, still well below the reach ceiling of original content. The ceiling itself is where the two formats diverge most sharply.

The distinction worth keeping: the algorithm does not penalize the act of sharing, but it consistently scores the semantic content of shares lower than posts that carry genuine professional insight or first-person experience. The act is neutral. The content of most shares is not.

360Brew: The Model That Reads Your Posts for Semantic Novelty

LinkedIn's FAIT team deployed 360Brew in the feed on March 12, 2026. It is a decoder-only foundation model with 150 billion parameters, built on LLaMA 3. The key architectural difference from its predecessor is that it uses LLM-generated embeddings to evaluate what a post is about, not just how it performs in early click testing. The model reads the text.

That capability changes the competitive landscape for curated content. Prior engagement-signal models rewarded posts that attracted early clicks and reactions, which gave a well-promoted reshare a plausible path to broad distribution. Under 360Brew, a low-engagement original post carrying specific or novel professional content can outrank a high-engagement reshare because the model evaluates semantic novelty directly from the words on the screen, not from the reaction count.

The model also evaluates what posts say in the context of the account's history. Voice-matching is the mechanism that determines whether a curated post reads as authentic to its sharer. Generic paraphrasing can produce vocabulary overlap of 30 to 50% with the sharer's prior posts while achieving near-zero voice similarity. 360Brew's profile-coherence scoring catches that gap. The rewrite looks different from the original source post but still does not sound like the person sharing it. That specific failure mode is what most AI rewriting tools produce without first analyzing the sharer's voice.

The enforcement posture hardened at the same time as the model rollout. In February 2026, LinkedIn began removing comments posted through third-party scripts or browser plugins from the 'Most Relevant' section, and Lempod was banned from the Chrome Web Store the same month. Coordinated promotion tools that previously gave reshares artificial early engagement no longer produce the signal they once did. The enforcement and the model architecture are solving the same problem from two directions.

0-30% Caption Similarity Is the Rewrite Threshold That Changes Curated Post Reach

DSMN8's analysis of more than 500,000 employee posts found a specific threshold. Captions rewritten so only 0 to 30% of words match the original earn over 4x more engagement, 3x more clicks, and 6x more comments compared to verbatim shares. Captions that are 99% similar to the original, where the sharer changed just one or two words, still perform nearly 3x better than unedited shares. The algorithm responds to any signal of personalization, even minimal ones.

The full multiplier requires crossing the 0 to 30% similarity threshold. What that requires operationally: rewriting the hook, restructuring the body, and changing the CTA in the sharer's own voice. Synonym substitution alone does not reach it. Rearranging sentences without changing vocabulary does not reach it either. The threshold is a full rewrite, not an edit.

Voice-matching is the mechanism that moves a curated post from the approximately 3x tier (minimal edit) to the 4x+ tier (0 to 30% similarity). What voice-matching requires is matching the sharer's habitual sentence length, punctuation patterns, and vocabulary register: whether they write casually or formally, how often they use first person, whether they prefer lists or prose. Generic paraphrasing that substitutes synonyms and rearranges sentences can produce 30 to 50% vocabulary overlap with the sharer's prior posts while achieving near-zero similarity in voice. 360Brew's profile-coherence scoring catches that gap.

SocialNexis's AI analyzes a user's prior posts to extract these fingerprints before rewriting. Accounts that use AI-generated captions without this step often land at 30 to 50% vocabulary similarity but near-zero voice similarity. The result scores well on vocabulary novelty versus the original source but fails profile coherence, capping its distribution before fast initial engagement has any chance to push it higher.

The 96.4% unedited share rate means this opportunity is not theoretical. In a program where nearly everyone shares verbatim, a single advocate who rewrites in their own voice gets 4x the impressions of every colleague sharing the same post. The threshold is not an aspirational goal; it is a floor the algorithm already applies to every shared post in the feed.

Most Employee Advocates Share Verbatim and Give Up Measurable Reach

Only 3.6% of employee advocates edit provided content before posting, per DSMN8's World's Biggest Employee Advocacy Study. The other 96.4% share verbatim and forfeit the 3 to 6x performance upside that rewriting delivers.

The gap this creates inside a single advocacy program is concrete. In a typical 50-person program where everyone shares verbatim, one employee who rewrites in their own voice gets 4x the impressions of every colleague sharing the same post. That one person is not doing anything exceptional by content standards. They are just rewriting.

The 96.4% unedited share rate reflects a workflow problem, not a motivation problem. Advocates are not given a rewriting tool that understands their individual voice, so the default is the path of least resistance: copy, paste, post. Most advocacy platforms surface a suggested caption and a share button. The friction between those two things is where the engagement upside disappears.

Top-performing rewriters in a program can serve as training data for the others. Their rewritten outputs expose the hook structures, personal anecdote placement, and sentence rhythms that the algorithm rewards for their specific audience and professional context. SocialNexis can identify these accounts in a program and use their outputs to train the voice model for the rest of the team, converting the 3.6% who rewrite into something closer to the full group.

Rewrite Curated Posts in Your Own Voice Before You Share

Start with the hook. The opening line is the most critical rewrite point: replacing the original post's hook with one that matches your own opening patterns signals personalization immediately and changes how the algorithm scores the post from the first line of text. This is not about sounding different from the source. It is about sounding like yourself.

Add 200 or more words of original commentary. Commentary under 100 words underperforms by roughly 3x, per van der Blom's analysis of 1.8 million posts. Specific professional experience or first-party data in the commentary scores higher under 360Brew than generic agreement, because the model evaluates what the commentary is actually about. "Great post, agree completely" is commentary. It is not the kind 360Brew rewards.

The 200-word threshold is not arbitrary. Van der Blom's dataset shows the drop-off is sharpest below 100 words, where commentary reads as a caption placeholder rather than a genuine position. At 200 words, there is enough text for 360Brew to evaluate whether the commentary carries a real take. Past that floor, more original text continues to work in the post's favor, though quality drives the gain more than count does.

Place any external link in the first comment, not the post body. A link in the post body suppresses reach by roughly 60%. A first-comment link still reduces visibility, by up to 80% in some observations, so reserve it for content that generates enough engagement on its own to absorb the penalty. A curated post that needs the link to do any work is already starting from behind.

Match your historical voice before rewriting, not the original post's. Review your last 10 posts for sentence length, punctuation style, list use, and first-person frequency before starting. Generic AI paraphrasing without voice analysis tends to produce vocabulary similarity with near-zero voice similarity, and that is the specific gap 360Brew's coherence scoring is built to find.

Stagger timing when coordinating within a team. A minimum 2 to 4 hour spread between shares, distributed across business hours, avoids triggering 360Brew's Coordinated Activity Ring detection. Caption variation is necessary but not sufficient. The timing variance is a separate requirement that caption rewriting alone does not satisfy.

Instant Repost vs. Repost-with-Thoughts: The Strategic Fork Most Guides Skip

An instant repost amplifies the original post's metrics and adds it to your activity feed. It costs the sharer almost nothing algorithmically. It is the right tool when the goal is to support a colleague or creator. The reach goes to the original post, not to the sharer. Using it to build your own authority on a topic is the wrong application.

A repost-with-thoughts creates a separate post in your name with independent reach and its own engagement counter. It is the right tool when you want the topic to appear in your content record and earn distribution through your own network. The two mechanics serve different goals and should be chosen based on those goals, not on which one feels more effort-appropriate.

Mixing the use cases is where reach goes to die. Adding short throwaway commentary to what should be an instant repost delivers the worst of both outcomes: the sharer's reach caps at 8 to 10% of their normal audience while adding no real credibility signal. The commentary is not long enough to move 360Brew's novelty scoring. The post does not amplify the original's metrics the way a clean instant repost would. It is the worst-performing version of both formats.

Employee reposts of company content carry roughly a -30% reach penalty compared to original employee-authored posts. A repost-with-thoughts rewritten in the employee's own voice closes much of that gap. That is what makes this a strategic decision at the scheduling layer, not just a stylistic preference. An instant repost supports the original creator. A fully rewritten repost-with-thoughts builds the sharer's own authority on the topic and earns its own distribution.

When an Advocacy Program Shares in Sync, 360Brew Can Flag the Whole Group

360Brew maps what LinkedIn calls "Coordinated Activity Rings." If the same cluster of accounts engages a post within minutes of publication, the entire group can receive shadow bans with 60 to 90 day recovery periods. LinkedIn VP of Product Gyanda Sachdeva stated in November 2025: "Our goal is to make engagement pods entirely ineffective." 360Brew's ring detection is the technical implementation of that stated intent.

The failure mode is invisible to most practitioners running advocacy programs. Five or ten employees sharing from the same content kit within a 10 to 20 minute window can trigger ring detection, even if each share uses different wording. The captions differ. The timing does not. 360Brew reads both signals.

The suppression is not announced. There is no warning, no notification, no error state. Accounts just see lower reach, and the standard response is to try posting more or at different times, neither of which addresses the ring classification. Recovery reportedly takes 60 to 90 days, meaning a single poorly-timed batch push can cost a program two months of reduced distribution.

The reason advocacy programs are structurally prone to this: program managers schedule shares from a dashboard, and dashboards default to batching. A manager who pushes a content kit to ten employees at the start of the workday creates a cluster that will share within the next hour or two as advocates log in and click approve. The tool's convenience feature becomes the detection trigger.

The fix operates at the scheduling layer, not the content layer. A minimum 2 to 4 hour spread between shares, with human-realistic timing variance across business hours, is what prevents cluster detection. Caption variation is necessary but not sufficient. SocialNexis handles timing staggering at the scheduling layer rather than leaving it to advocates to coordinate manually, which is the structural reason most advocacy programs trigger ring detection in the first place.

Frequently asked questions