LinkedIn and X carry the worst AI slop problem, Pangram data shows

LinkedIn leads all social platforms for AI-generated longform content, with 41% of posts over 250 words flagged as fully machine-written, according to new data from AI detection firm Pangram.

The figures come from Pangram's Chrome extension, which analysed 1,002,627 posts across LinkedIn, Medium, Substack, X, and Reddit between April and June 2026. Posts were counted once, excluded anything under 50 words, and ran through Pangram 3.3, the company's detection model with a 0.01% false positive rate.

The distribution is what matters for marketers. LinkedIn made up roughly a third of all scanned posts, yet accounted for nearly two-thirds (62%) of all AI content flagged. That structural skew means organic content from real marketers is competing in the most polluted environment across major platforms.

X runs LinkedIn close when AI-assisted posts are included

On X, almost half of longform articles contained AI writing: 23.9% were fully AI-generated and another 22.9% were AI-assisted or mixed, leaving only 53.2% fully human-authored. LinkedIn posts show a sharper all-or-nothing pattern: just 4.3% of longform content was AI-assisted, with most flagged posts either fully human or fully machine-generated.

Shortform LinkedIn content (50 to 250 words) wasn't much better, with 30% flagged as fully AI-written.

Substack had the lowest longform AI rate, yet more than a fifth of posts (21.9%) were still flagged as AI-generated or AI-assisted. Reddit replies were 98% human-written, but standalone posts contained AI text far more often.

The algorithmic penalty is already live

LinkedIn has been moving against AI slop since May. The platform's March 2026 algorithm update penalises generic AI content with up to 47% less organic reach, according to analysis by Future Factors.

LinkedIn's algorithm cannot detect AI-written content directly; what it detects is whether anyone cared enough to finish reading, a June analysis by ZoomSphere noted. AI-generated content without specific professional insight creates near-zero dwell time, earns no saves, and triggers no discussion, and the algorithm reads that behavioural signal and stops distributing the content.

LinkedIn rolled out its biggest algorithm overhaul in history this year, a 150-billion-parameter AI model called 360Brew, first introduced in a research paper in January 2025 and gradually deployed throughout 2026.

The algorithm now downranks content that exhibits low perplexity (predictable, generic sentence structures), according to Linkboost's analysis of the 2026 changes.

On 30 July, LinkedIn unveiled a "seems like AI slop" button so users can report posts that appear to be low-quality AI-generated content. Chief Product Officer Hari Srinivasan said the feedback will train detection models and reduce algorithmic reach for flagged posts.

LinkedIn is also scaling back its own AI writing tools, replacing the "Enhance your post" feature with a conservative proofreading tool that preserves user voice.

Why this matters for social marketers

The data gives marketers hard evidence of where their organic content is competing against bots, and the answer is structural, not anecdotal.

If you're posting longform thought leadership on LinkedIn, more than 40% of comparable posts are fully machine-generated. AI slop writing is not sequestered on unpopular automated SEO farms; humans are regularly wading through AI dreck on hugely popular sites, 404 Media's analysis of the Pangram data noted.

According to Sprout Social's Q1 2026 Pulse Survey, 56% of social users say they see AI slop often on social media, and unlabeled AI content is the top thing consumers want brands to stop doing in 2026. Gen Z and Millennial users are the most likely to unfollow, mute or block accounts because their content feels like AI slop.

Earlier studies reached similar conclusions. Originality.ai reported in January 2026 that 53.7% of long-form LinkedIn posts it analysed were "likely AI", based on a sample of 3,368 posts from 99 influential profiles across eleven industries. The figure and Pangram's 41% longform rate are not directly comparable (different detection models, sampling methods, and definitions of "AI-generated" versus "likely AI"), but both point toward LinkedIn carrying an unusually high concentration of machine-written longform content.

What to do about it

The strategic implication is straightforward: authenticity is no longer a brand value; it's distribution leverage.

The signals LinkedIn's algorithm uses to downrank AI content include lexical diversity (AI repeats similar phrases and transitions), structural predictability (AI uses the same hook-point-CTA format), tonal consistency (humans have emotional range; AI maintains unnaturally steady professional temperature), and personal specificity (AI rarely references actual client names, real dates, specific situations, or genuine uncertainty).

The practical answer most marketers are landing on: use AI as a back-end efficiency tool for research, ideation, and drafting, while keeping human judgment and voice front and centre, according to a February analysis by AdWhite.

Inject personal anecdotes, contrarian opinions, and specific data points that an LLM wouldn't know, Linkboost's algorithm guide advises.

Organic post reach from LinkedIn company pages has dropped 60 to 66% since 2024, and the same content posted from personal profiles generates 561% more reach than from company pages, according to April analysis by Professional Punch.

The Pangram data is the first large-scale view of what users actually see, not what exists somewhere on the platform. Because Pangram works passively in the background while a user browses, it only scans posts that its users actually see, helping answer the question of whether AI slop is actually poisoning the internet that humans use versus polluting the internet more broadly.

The answer is unequivocal: if you're publishing organic content on LinkedIn or X, you're publishing into the most AI-saturated feeds on the social web.