AI LinkedIn posts get 45% less engagement than human ones

A 2025 study of 3,368 LinkedIn posts across 99 profiles found that likely-AI-generated content received 45% less engagement than likely-human-written content on average. That headline has circulated widely since Originality.AI published it. What circulates less often: the number is a floor, not a ceiling. Accounts that post AI-heavy content for three or four consecutive weeks show compounding suppression that hits subsequent human-voice posts too. The gap is not per-post arithmetic. It is a progressive tax on account-level reach that most teams never see until the damage is done.

The AI Generated Posts Engagement Gap on LinkedIn: What the Data Actually Shows

The Originality.AI 2025 LinkedIn AI engagement study analyzed 3,368 posts from 99 influential LinkedIn profiles across 11 industries, covering January through November 2025. The finding that spread everywhere: likely-AI-generated posts received 45% less engagement on average than likely-human-written posts. That single number is the least interesting thing in the study.

53.7% of long-form posts (100 or more words) were classified as likely AI-generated. Architecture and Design came in at 100% AI classification. Wellness and Personal Development came in at 92%. These figures suggest LinkedIn is not managing a marginal AI problem. A majority of substantive posts on the platform may already be machine-generated, which changes the competitive context for everyone, including human writers who have never touched an AI tool.

The reach backdrop makes the per-post penalty look mild by comparison. Average organic post reach has fallen to 8-12% of followers, down from 15-20% the prior year. Content creation volume is up 14% year-over-year. AI-flooded feeds are compressing distribution for everyone, not just accounts posting AI content. Even a fully human-written account is reaching a smaller fraction of its followers than it did twelve months ago.

The 45% average obscures substantial industry variance. Some categories show AI content outperforming human posts by 75%. Others show losses exceeding 80%. Treating the headline as a flat universal coefficient applied to every account and every post type misrepresents what the data actually contains. The question is not whether the gap exists. The question is which categories, which behaviors, and which time horizons determine how deep it goes.

How Does LinkedIn's 360Brew Model Identify AI-Generated Posts?

360Brew is a 150-billion-parameter foundation model. LinkedIn published the research paper behind it on arXiv in January 2025. The model evaluates each account through profile coherence, network relevance, engagement patterns, and content consistency across posting history. It is not a single-pass spam filter. It builds a sustained account-level model and measures new content against that model continuously.

For AI detection specifically, LinkedIn uses human editors to annotate thousands of posts and trains machine learning models on those annotations to examine language patterns. The company claims 94% accuracy in identifying generic AI-generated content in initial testing. That accuracy figure implies a systematic labeling and training pipeline, not a set of brittle heuristics that can be bypassed by slightly rephrasing output.

Posts with three or more hashtags receive approximately 70% lower reach under 360Brew. Posts containing external links in the body receive a roughly 60% reach penalty. Both are defaults that AI-generated content workflows commonly produce without intervention. If a writing tool or template adds hashtags and links automatically, it is layering two additional distribution cuts on top of the AI detection penalty before any human reads the post.

The enforcement mechanism is distribution, not deletion. LinkedIn VP and Executive Editor Laura Lorenzetti confirmed in May 2026 that flagged AI-generated posts are not removed but instead receive reduced algorithmic distribution, confined mainly to the author's immediate network. The platform frames this as content quality management. The practical result is the same: a post that reaches only existing connections does not grow an audience.

Dwell Time Is the First Signal to Decay on AI-Generated LinkedIn Posts

Under 360Brew, posts where readers spend 61 or more seconds achieve 15.6% engagement rates. Posts read in under three seconds average just 1.2%. That 13x gap is not a secondary metric. It is the primary mechanism behind the AI engagement penalty. LinkedIn can infer content quality from reading behavior before a single like or comment is recorded.

On accounts we monitor running unedited AI output, dwell times consistently fall under eight seconds, well below the 45-second threshold that starts to register as meaningful signal in 360Brew's scoring. Readers do not make a conscious decision to skip. They recognize within a few words that the opening contains nothing specific enough to warrant stopping, and they scroll. The pattern holds across categories and audience sizes.

The fix is not making AI copy sound more human grammatically. Polished grammar does not stop a reader from scrolling. What stops scrolling is a specific, unexpected tension in the first two sentences. Generic hooks like 'Here is what I learned about leadership' fail this test regardless of how polished the surrounding prose is. The opening needs to name a concrete problem, observation, or result the reader would not have anticipated.

Meaningful comments of 15 or more words, especially those that generate multi-turn threads, carry roughly 15x the algorithmic weight of likes under 360Brew. Saves carry 5-10x more weight than likes. AI-generated posts structurally fail to trigger either of those heavier signals. A post that collects five short affirming comments scores far worse in 360Brew's model than one that generates two 30-word responses and a save. AI output tends to attract the former.

Leadership Posts Outperform by 75%. Strategy Posts Lose 80%. The Gap Is Not Uniform.

The Originality.AI industry breakdown is where the 45% average breaks down as a universal guideline. AI posts in Leadership and Inspiration outperformed human posts by 75%. AI posts in Innovation and Strategy underperformed by 80%. Marketing and Branding underperformed by 73%. The industry an account operates in changes the math entirely, and for some practitioners, the 45% headline is simply the wrong number to be tracking.

The outperformance in Leadership and Inspiration makes sense once you consider the comparison baseline 360Brew is working from. In motivational and inspirational content, both creators and audiences are already producing AI-heavy material. The algorithm cannot establish a clear human-written baseline to compare against, so the relative penalty disappears. 360Brew is not lenient in that category; it simply lacks a credibility floor to measure against.

The penalty is sharpest where audiences hold high expertise expectations and where the algorithm has abundant human-written comparators. Healthcare, strategy, marketing. Readers in those categories evaluate claims against actual domain knowledge. A vague observation about building resilient teams moves through an inspirational feed. That same observation fails immediately when a strategy director reads it and finds no concrete mechanism, no named evidence, no specific context. 360Brew can identify the gap precisely because it has enough human reference material to set the floor.

Industry context determines whether the AI engagement gap is a material risk or effectively irrelevant for any given account. A motivational coach posting AI-heavy content is playing a different game than a B2B marketing strategist doing the same thing. Practitioners in expertise-heavy categories need to treat AI output as draft scaffolding rather than publishable content. Practitioners in inspirational categories have more room, though the platform's enforcement posture continues tightening regardless of category.

What Most Teams Get Wrong About the AI Engagement Gap

The 45% headline frames the problem as per-post performance. That framing is why most teams manage it badly. When an account posts AI-heavy content for three to four consecutive weeks, 360Brew's topic authority score stalls because the system cannot build a coherent credibility profile from structurally identical posts that cover related topics without adding any specific observation or evidence.

The consequence is that human-voice posts also underperform after that period. The account's baseline distribution has already eroded. The gap is not limited to AI posts. It becomes a tax on the account's future reach. Teams that return to human-written content after an AI-heavy run often assume organic growth has stalled for some unrelated reason, because the compounding effect does not appear in any single post's metrics.

The engagement signals that matter most under 360Brew are precisely the ones AI content structurally fails to generate. Meaningful comments of 15 or more words and saves carry dramatically more algorithmic weight than likes. AI posts rarely trigger genuine conversation or bookmarking behavior. The account accumulating short affirming reactions is accumulating low-value signal, and 360Brew registers that at the account level, not just the post level.

Teams tracking performance on a per-post basis cannot see account-level authority erosion, which unfolds over multiple weeks as 360Brew's model of the account shifts. By the time the suppression is obvious in post-level data, several weeks of compounding have already occurred and recovery requires sustained high-quality output over an additional window. The lag between cause and visible symptom is long enough that most teams never connect the two.

AI Comments Carry a Separate Detection Risk Most Practitioners Miss

LinkedIn's 360Brew evaluates accounts across post and comment behavior together, not just individual posts. The detection risk from AI-generated comments is distinct from and, in some contexts, greater than the risk from AI-generated posts. Comments create a cross-post timing and vocabulary signature that LinkedIn's Coordinated Activity Ring detection can flag even on a single-user account.

The signature forms through two patterns. First, similar phrase clusters appearing across multiple posts and comments from the same account over time. Second, response latency: the time between a post going live and the commenting account responding remains suspiciously consistent. Human commenting behavior is irregular. AI commenting workflows are not.

Most practitioners treat AI comment tools as harmless engagement boosters. The tool replies to relevant posts, adds brief affirming text, and appears to improve visibility metrics in the short run. On the basis of how 360Brew processes cross-post behavioral data, AI-generated comments are a faster path to a shadow ban than AI-generated posts alone. The combined post-and-comment signature is a stronger detection signal than either behavior produces individually.

Shadow ban recovery for accounts LinkedIn flags as coordinated activity rings runs 60 to 90 days. During that period, even high-quality human-written posts receive severely compressed distribution. The account's reach does not recover through a single strong post; it requires consistent output across the full recovery window. The short-term metric improvement an AI comment tool produces is not worth several months of suppressed reach.

Closing the AI Generated Posts Engagement Gap: A Hybrid Workflow That Works

The hybrid workflow that closes the engagement gap is asymmetric by design. Use AI for structure, research synthesis, and bullet expansion. Then rewrite the first sentence, the personal stakes line, and the closing call to action in the account holder's actual voice. Those are the three points 360Brew weights most heavily: the hook as a dwell trigger, the personal stakes line as a credibility signal, and the close as a comment trigger.

Rewrites of the middle body produce diminishing returns relative to the time they cost. The algorithm does not score for literary quality in body paragraphs. It scores initial dwell behavior and terminal engagement behavior. Targeted voice injection at the hook and the close recovers most of the engagement gap without requiring a full manual rewrite of every post.

Creators who use AI as a drafting assistant but edit aggressively for voice and specificity outperform those posting unedited AI output by approximately 34% on engagement. That gap is not achieved by replacing AI with fully human writing. It is achieved by adding human specificity at the two points where it changes reader behavior: the opening tension and the closing prompt.

LinkedIn's own guidance states that AI is best used to augment your expression rather than replace it, and recommends disclosure when AI contribution is substantial. That framing aligns with the performance data. The platform is not anti-AI; it is anti-homogeneous output. Treating AI as a drafting layer that still requires a human editorial pass positions an account correctly against both the platform's stated policy and the behavioral patterns 360Brew actually rewards.

One operational note for teams building this workflow across multiple accounts. The first-person specificity that closes the engagement gap is not generic personalization. It is named context: a specific client, a specific result, a specific failure. That level of detail is what creates dwell time, triggers multi-turn comment threads, and gives 360Brew's topic authority scoring something coherent to build on across consecutive posts.

Frequently asked questions