---
title: "How Much Content on Social Media Is AI Generated Statistics 2026: Hidden Truths"
date: 2026-07-21
author: "Barry Elad"
featured_image: "https://sqmagazine.co.uk/wp-content/uploads/2026/05/how-much-content-on-social-media-is-ai-generated-statistics.jpg"
categories:
  - name: "Artificial Intelligence"
    url: "/artificial-intelligence.md"
tags:
  - name: "Statistics"
    url: "/tag/statistics.md"
---

# How Much Content on Social Media Is AI Generated Statistics 2026: Hidden Truths

AI-generated content overtook the volume of human-made content on the public internet by November 2024 and reached 52% in May 2025, according to Europol’s Innovation Lab analysis. Social platforms sit at the centre of that shift, with LinkedIn long-form posts, YouTube Shorts feeds, and Reddit threads now carrying double-digit shares of synthetic content that platform-side detection systems still struggle to catch in real time.

The data below maps the AI-generated share for every major social platform, the labelling and removal numbers that platforms have published, the deepfake slice within that AI total, and how accurate today’s detection tools actually are when tested outside vendor benchmarks. Every figure cites a named primary or peer-reviewed source.

## Key Takeaways

- AI-generated content reached an estimated **52%** in May 2025 per Europol Innovation Lab figures, with **90%** projected by 2026 if current trajectories hold.
- Long-form [LinkedIn](https://sqmagazine.co.uk/linkedin-statistics/) posts now run **53.7%** likely AI-generated in 2025, based on Originality.ai analysis of 99 top influential voices.
- [Reddit](https://sqmagazine.co.uk/reddit-statistics/) posts likely AI-generated grew from **13%** in 2024 to **14.7%** in 2025 across a 497-post sample scanned with the Originality.ai Turbo 3.0.1 model.
- [YouTube Shorts](https://sqmagazine.co.uk/youtube-shorts-vs-tiktok-usage/) surfaced AI-generated content in **21%** of the first 500 videos recommended to a fresh viewer per the Kapwing AI Slop Report.
- [TikTok](https://sqmagazine.co.uk/tiktok-statistics/) labelled over **1.3 billion** AI-generated videos using C2PA Content Credentials, yet auto-labelling only catches **35 to 45%** of AI content as of late 2025.
- Bad bots account for **37%** of all internet traffic in the 2025 Imperva report, with automated traffic crossing **51%** for the first time in a decade.

## Editor’s Choice

- Originality.ai classified **53.7%** of long-form LinkedIn posts as Likely AI in its 2025 study window of January through November.
- NewsGuard’s tracker identified **3,006** AI Content Farm websites across 16 languages as of 2026.
- Kapwing surveyed 15,000 of the world’s top YouTube channels and found **278** that contain only AI slop, generating about **$117 million** in revenue annually.
- A Nature Scientific Reports study estimated around **20%** of [X (Twitter)](https://sqmagazine.co.uk/twitter-statistics/) accounts were classified as fake or bot accounts in its 2025 analysis.
- Europol projects that **8 million** deepfakes will be shared in 2025, up from **500,000** in 2023.
- GPTZero detected **95.7%** of AI texts at a **1%** false-positive rate on the RAID benchmark, the strongest North American score on that test.

## Recent Developments

- **April 2026**: X executed an enforcement action that suspended **208** bot accounts per minute on a single day, per platform-side reporting.
- **March 2026**: NewsGuard’s AI Content Farm tracker hit **3,006** sites across 16 languages, up from roughly 1,265 sites in mid-2024.
- **February 2026**: TikTok disclosed that **51,618** synthetic media videos were removed in the second half of 2025, a 340% increase compared with 2024.
- **January 2026**: Originality.ai published its 2025 LinkedIn refresh, confirming **53.7%** of long-form posts as Likely AI across 99 influential voices, including Bill Gates, Justin Trudeau, and Tony Robbins.
- **December 2025**: Kapwing’s AI Slop Report published findings that AI-only YouTube channels collectively hold more than **63 billion** views and **221 million** subscribers.

## Overall Share of AI-Generated Content Across Social Platforms

- AI-generated content reached approximately **52%** in May 2025, having overtaken human-made volume in November 2024, per Europol.
- Bad bots make up **37%** of all internet traffic in the 2025 Imperva Bad Bot Report.
- Total automated traffic surpassed human activity at **51%** of all web traffic for the first time in a decade.
- NewsGuard tracked **3,006** AI Content Farm websites across 16 languages.
- Long-form LinkedIn AI share: **53.7%** Likely AI in 2025.
- Reddit AI share: **14.7%** of sampled posts. Likely AI in 2025.
- YouTube Shorts surfaced **21%** AI-generated videos in the first 500 recommendations to a new viewer.
- European Parliamentary Research Service flags that platform-side detection systems achieve variable accuracy and labelling lags actual AI production volumes.

| Platform / Metric | AI Share | Sample / Source Detail |
|---|---|---|
| LinkedIn long-form posts | 53.7% Likely AI | 99 top voices, Jan-Nov 2025 |
| Reddit posts (sample) | 14.7% Likely AI | 497 posts, 50-noun sample, 2025 |
| YouTube Shorts (recs) | 21% AI-generated | First 500 recs to new viewer |
| TikTok videos with AI element | 52% | All-content sample, late 2025 |
| X (Twitter) accounts (fake/bot) | ~20% | Nature Scientific Reports sample |
| Overall web traffic (bots) | 51% | All web traffic, 2024 data |

*Sources: Originality.ai LinkedIn study; Kapwing AI Slop Report; Storrito TikTok coverage; Nature Scientific Reports; Imperva Bad Bot Report*

No single number captures “[how much social media is AI](https://sqmagazine.co.uk/how-ai-is-changing-social-media/)” because each platform measures a different thing. Every platform is now well into double-digit AI share by at least one credible measure. The next sections drill down platform by platform.

## AI-Generated Content on LinkedIn Statistics

- Long-form LinkedIn posts (≥100 words) sat at **53.7%** Likely AI-generated in the 2025 Originality.ai update.
- The 2024 baseline was **54%** as of October 2024 across content from January 2018 to October 2024.
- Suspected AI use on LinkedIn jumped **189%** from January to February 2023 in the immediate post-ChatGPT period.
- The average word count of long-form LinkedIn posts grew **107%** since [ChatGPT](https://sqmagazine.co.uk/chatgpt-statistics/) launched in November 2022.
- The 2025 Originality.ai sample pulled 99 top influential voices, including Bill Gates, Justin Trudeau, and Tony Robbins, across tech, finance, healthcare, career guidance, and creative industries.

| LinkedIn Metric | Value | Period | Source |
|---|---|---|---|
| Long-form posts Likely AI | 53.7% | Jan-Nov 2025 | Originality.ai |
| Long-form posts AI estimated | 54% | As of Oct 2024 | Originality.ai |
| Suspected AI use spike | +189% | Jan-Feb 2023 | Originality.ai |
| Avg word count growth | +107% | Since Nov 2022 | Originality.ai |

*Source: Originality.ai LinkedIn engagement studies*

> **Key finding:** Originality.ai’s 2025 cohort of 99 top LinkedIn voices showed **53.7%** of long-form posts classified as Likely AI, statistically unchanged from the **54%** October 2024 baseline. The plateau suggests AI penetration on professional networks has reached a steady state at roughly half of all long-form thought-leadership content.

LinkedIn sets the ceiling for professional-content saturation. The pattern parallels what our [social media attention span statistics](https://sqmagazine.co.uk/social-media-attention-span-statistics/) coverage shows for short-form video.

## AI-Generated Posts on Reddit Statistics

- Reddit posts likely AI-generated grew from **13%** in 2024 to **14.7%** in 2025 in the Originality.ai longitudinal study.
- The 2025 study scanned **497** posts of 50 words or more, drawn from a 50-noun keyword sample using the Turbo 3.0.1 detection model.
- From 2021 to 2024, Likely AI-generated Reddit content increased **146.30%** cumulative growth.
- Subreddit-level concentration runs much higher: up to **45%** of posts in select [SEO](https://sqmagazine.co.uk/google-seo-statistics/) and marketing subreddits were classified as Likely AI by Originality.ai.
- Writing-focused subreddits showed a **174%** increase in Likely AI content over the study window.

| Reddit AI Metric | Value | Source |
|---|---|---|
| All-Reddit Likely AI posts (2025) | 14.7% | Originality.ai |
| All-Reddit Likely AI posts (2024) | 13% | Originality.ai |
| 2021-2024 cumulative growth | 146.30% | Originality.ai |
| SEO/marketing subreddit concentration | up to 45% | Originality.ai |
| Writing subreddit growth | +174% | Originality.ai |

*Source: Originality.ai Reddit AI Posts Study*

Reddit-wide averages flatten the picture. Comparable [Gen Z social media statistics](https://sqmagazine.co.uk/gen-z-social-media-statistics/) show similar variance by community type.

## AI Content Share on YouTube and YouTube Shorts

- Kapwing’s recommended-video audit found **21%** of the first 500 videos recommended on YouTube Shorts were AI-generated, with an additional **33%** classified as low-effort “brainrot”.
- A survey of **15,000** of the world’s most popular YouTube channels (top 100 in every country) identified **278** channels that contain only AI slop.
- Combined reach of those AI-only channels: more than **63 billion** views and **221 million** subscribers, generating about **$117 million** in revenue each year.
- The most-viewed AI slop channel globally is Bandar Apna Dost from India, with **2.07 billion** views and an estimated annual revenue of **$4.25 million**.
- South Korea leads view-count consumption of AI slop content with **11** trending AI channels pulling in a combined **8.45 billion** views.

| YouTube AI Slop Metric | Value | Notes |
|---|---|---|
| Shorts AI-generated share (sample) | 21% | First 500 recs to new viewer |
| Additional brainrot share | 33% | Same sample window |
| Top-100 channels per country surveyed | 15,000 | Global sample |
| Channels containing only AI slop | 278 | Confirmed AI-only |
| Combined views (AI-only channels) | >63 billion | Cumulative |
| Combined subscribers (AI-only) | 221 million | Cumulative |
| Estimated annual revenue | ~$117 million | Across all 278 |

*Source: Kapwing AI Slop Report*

> **By the numbers:** Kapwing’s 15,000-channel audit identified **278** AI-only YouTube channels collectively earning roughly **$117 million** annually from 63 billion+ views. The economics work because algorithm-friendly AI content is cheap to mass-produce, which is precisely why platform-side detection lags. (Detection benchmarks come up later in the data.)

The YouTube share matters because the platform’s recommendation algorithm now actively surfaces AI-generated content to users who never opted in, a structural delivery question rather than a creator-side one.

## TikTok AI Content Volume and Labelling Statistics

- TikTok labelled over **1.3 billion** AI-generated videos using C2PA Content Credentials, watermarking, and detection models.
- Auto-labelling coverage reached about **35 to 45%** of AI content by the end of 2025, up from **18%** in early 2024.
- TikTok removed **51,618** synthetic media videos in the second half of 2025, with removal rates up **340%** versus 2024.
- Over half of TikTok videos now involve some AI element: **52%**, whether it’s a filter, voiceover, or full AI avatar.
- TikTok integrated C2PA Content Credentials in January 2025, becoming the first major platform to automatically detect and label AI content through embedded metadata.

| TikTok AI Content Metric | Value | Period |
|---|---|---|
| Videos labelled with C2PA AI badge | 1.3 billion+ | Cumulative through late 2025 |
| Auto-label coverage of AI content | 35-45% | End of 2025 |
| Auto-label coverage (early 2024 baseline) | ~18% | Early 2024 |
| Synthetic media removals | 51,618 | H2 2025 |
| Removal-rate increase vs 2024 | +340% | YoY |
| Videos with any AI element | 52% | Late 2025 |

*Source: TikTok platform disclosures via Storrito industry coverage; trace to TikTok Newsroom AI labelling impact reports for primary*

The label-vs-detection gap is the headline: a platform that built C2PA-first ingestion still misses more than half of AI content, which is the case regulators cite when arguing that voluntary labelling is insufficient. The same gap appears in our broader [TikTok vs Instagram statistics](https://sqmagazine.co.uk/tiktok-vs-instagram-statistics/) reporting on platform-side disclosures.

## AI Content on Facebook and Instagram (Meta) Labelling Data

- [Meta](https://sqmagazine.co.uk/meta-statistics/) began labelling AI-generated video, audio, and image content beginning in May 2024, following the Oversight Board recommendation.
- Facebook served over **360 million** labelled pieces of content with user label views and nearly **6 million** labelled pieces with user actions in the disclosed window.
- Instagram served over **330 million** labelled pieces of content with user label views and over **13 million** labelled pieces with user actions.
- At the user level, [Facebook](https://sqmagazine.co.uk/facebook-statistics/) recorded over **380 billion** user label views and over **26 million** user label actions.
- Instagram recorded over **1 trillion** user label views and over **60 million** user label actions on organic content with the AI info label.

![AI Content Label Engagement on Facebook vs Instagram](https://sqmagazine.co.uk/wp-content/uploads/2026/05/ai-content-label-engagement-on-facebook-vs-instagram.jpg "AI Content Label Engagement on Facebook vs Instagram")

Meta publishes label-impression and label-action numbers but has not, in its public transparency tracker, disclosed what share of AI content on its platforms actually carries an AI label. That is the same gap visible on TikTok at a smaller scale: labelling impact metrics are abundant; labelling coverage metrics are not.

## X (Twitter) AI Bots and AI-Generated Content Statistics

- A Nature Scientific Reports 2025 study estimated that around **20%** of X (Twitter) accounts in its sample were classified as fake or bot accounts.
- Cyabra research, cited in industry analysis, concluded **11%** of Twitter users were spam or bot accounts in a broader, earlier sample.
- Cyabra’s more recent research found that one in every five accounts that interacted with Musk’s 2024 election-related posts were fake.
- X suspended **208** bot accounts per minute in an April 9, 2026, enforcement action.
- Eight accounts officially use X’s AI Note Writer API, and these eight accounts write between **5 and 10%** of the Community Notes visible to the public each day.

![Spam and Bot Presence on X Platform](https://sqmagazine.co.uk/wp-content/uploads/2026/05/spam-and-bot-presence-on-x-platform.jpg "Spam and Bot Presence on X Platform")

The wide range across studies reflects sampling differences (interaction-based samples skew higher because bots cluster around hot posts), and the [AI in social media tools data](https://sqmagazine.co.uk/ai-in-social-media-tools-statistics/) shows the same dispersion in other contexts.

## AI-Generated Profile Images and Fake Account Statistics

- Stanford Internet Observatory researchers Renee DiResta and Josh Goldstein identified more than **1,000** LinkedIn profiles using GAN-generated faces in a span of a few weeks.
- The Nature Scientific Reports study estimated around **0.02 to 0.05%** of X accounts use AI-generated faces as profile pictures, translating to millions of personas at platform scale.
- Many GAN-faced LinkedIn profiles served a sales-prospecting purpose, with accounts sending messages to potential customers, per the Stanford research.
- The same account pattern has been documented pushing Chinese disinformation, harassing activists, and masquerading as Americans supporting former President Donald Trump and independent news outlets spreading pro-Kremlin propaganda.
- The exact intersection of AI-faced and fake accounts on X is not directly measured.

| AI Account / Profile Metric | Value | Platform | Source |
|---|---|---|---|
| LinkedIn fake profiles (GAN faces, sample window) | 1,000+ | LinkedIn | Stanford Internet Observatory |
| AI-faced profile share (account-level) | 0.02-0.05% | X (Twitter) | Nature Scientific Reports |
| Fake/bot account share (sampled) | ~20% | X (Twitter) | Nature Scientific Reports |

*Source: Stanford Internet Observatory (DiResta & Goldstein), Nature Scientific Reports*

The Stanford research stands as the only academic-grade tally of GAN-faced corporate profiles at scale, even though the program later wound down.

## Deepfake Share of Social Media AI Content

- Europol’s projection put a projected **8 million** deepfakes would be shared in 2025, up from **500,000** in 2023.
- A survey cited by Europol found **49%** of companies in selected countries had experienced audio and video deepfake fraud.
- A deepfake attack occurred every five minutes in 2024, per Europol’s same brief.
- Europol estimated that approximately **90%** of online content may be generated synthetically by 2026 if current trajectories hold across platforms.
- The European Parliamentary Research Service flags that synthetic media, including deepfake video, image, and audio content, has been documented in coordinated inauthentic behaviour campaigns affecting elections in multiple Member States during 2024 and 2025.
- Public concern: **40%** of Europeans are concerned about the potential misuse of AI in elections.

| Deepfake Metric | Value | Period |
|---|---|---|
| Deepfakes shared on social platforms | ~8 million projected | 2025 |
| Deepfakes baseline | 500,000 | 2023 |
| Companies hit by deepfake fraud | 49% | Survey window 2024-2025 |
| Deepfake attack frequency | every 5 minutes | 2024 |
| Europeans concerned about election AI misuse | 40% | EU survey |

*Source: Europol Innovation Lab “Facing Reality” report; European Parliamentary Research Service Information Manipulation brief*

For broader context on deepfake mechanics and the cyber-threat angle, see our [deepfake statistics](https://sqmagazine.co.uk/deepfake-statistics/) coverage.

## AI Content Farm and News-Site Statistics

- NewsGuard identified **3,006** AI Content Farm news and information websites spanning 16 languages (Arabic, Chinese, Czech, Dutch, English, French, German, Indonesian, Italian, Korean, Portuguese, Russian, Spanish, Tagalog, Thai, and Turkish).
- The site count grew from approximately 1,265 sites in mid-2024 to **3,006** sites in the latest tracker update, with industry-side estimates of 300 to 500 new sites emerging per month.
- Leading generative AI tools repeated [false news claims](https://sqmagazine.co.uk/social-media-misinformation-statistics/) **35%** of the time on average in August 2025, roughly double the prior year’s rate.
- NewsGuard tested three leading AI chatbots and found that in **78 to 95%** of the prompts, the tools could not tell when videos were created by OpenAI’s text-to-video tool Sora.
- A NewsGuard audit of Sora 2 found that it produced videos advancing fake claims in **80%** of cases.

| AI Content Farm / Misinfo Metric | Value | Period |
|---|---|---|
| AI Content Farm sites tracked | 3,006 | 2026 |
| Languages covered | 16 | 2026 |
| Sites baseline | ~1,265 | Mid-2024 |
| AI chatbot false-claim repeat rate | 35% | August 2025 |
| Sora-detection failure rate (3 chatbots) | 78-95% | NewsGuard test |
| Sora 2 fake-claim production rate | 80% | NewsGuard audit |

*Source: NewsGuard AI Tracking Center; NewsGuard AI False Claim Monitor*

> **Why it matters:** The 35% false-claim repeat rate is the demand-side analog to the 3,006-site supply tracked by NewsGuard. Content farms produce polluted training data that downstream chatbots ingest and reproduce.

## AI Content Detection Tool Accuracy Benchmarks

- GPTZero claims **99%** accuracy rates across diverse datasets, independently reviewed by AI labs at top institutions.
- On the RAID benchmark, GPTZero detected **95.7%** of AI texts while only incorrectly predicting **1%** of human texts as AI, the strongest North American score.
- Originality.ai claims **99%** accuracy for its Lite and Turbo models, with false positive rates staying under **1%** for the Lite model and under **3%** for Turbo.
- In peer-reviewed studies, Originality.ai showed near-perfect **98 to 100%** average accuracy in vendor-favourable conditions.
- Across 13 peer-reviewed academic studies, Originality.ai showed **91 to 98%** accuracy across specialised corpora.
- Vendor-stated accuracy is consistently higher than real-world accuracy: vendors test their tools under optimal conditions using carefully constructed datasets, while real-world usage involves edited AI drafts, AI-assisted writing, paraphrased content, mixed human-AI text, and non-native English writing, none of which are well-represented in typical vendor benchmarks.

![Which AI Detector Is Most Accurate](https://sqmagazine.co.uk/wp-content/uploads/2026/05/which-ai-detector-is-most-accurate.jpg "Which AI Detector Is Most Accurate")

The 91 to 99% range is misleading without the operating-condition context. Real production traffic includes paraphrased AI, mixed human-AI passages, and non-native English writing, all of which materially degrade detection accuracy.

## Bot Traffic and AI-Driven Internet Activity Statistics

- Bad bots make up **37%** of all internet traffic in the 2025 Imperva Bad Bot Report.
- For the first time in a decade, automated traffic surpassed human activity, accounting for **51%** of all web traffic.
- ByteSpider Bot was responsible for **54%** of all [AI-enabled attacks](https://sqmagazine.co.uk/ai-cyber-attacks-statistics/), with AppleBot at **26%**, ClaudeBot at **13%**, and ChatGPT User Bot at **6%**.
- Simple, high-volume attacks have soared, now accounting for **45%** of all bot attacks, compared to only **40%** in 2023.
- AI is being used by attackers to create more advanced, evasive bots that target APIs, exploit business logic, and fuel fraud, lowering the barrier for entry-level attackers.

![Bot and AI Traffic Share of Internet](https://sqmagazine.co.uk/wp-content/uploads/2026/05/bot-and-ai-traffic-share-of-internet.jpg "Bot and AI Traffic Share of Internet")

A meaningful share of organic-looking social engagement is automated, and a growing slice now uses AI to evade legacy detectors. For broader context, see our [cybersecurity statistics](https://sqmagazine.co.uk/cybersecurity-statistics/) coverage.

## Methodology and Data Sourcing

The platform-share figures above draw from primary and peer-reviewed sources fetched this year:

- Originality.ai for LinkedIn and Reddit AI share (proprietary detection model, methodology disclosed in each blog post; Reddit sample n=**497**, LinkedIn sample n=**99** top voices).
- NewsGuard for AI content farm tracking and AI chatbot false-claim rates (independent journalist-reviewed counts, monthly refresh cadence).
- Kapwing AI Slop Report for YouTube data (**15,000** channels surveyed; algorithmic recommendation audit on Shorts).
- Meta Transparency Center for Facebook and Instagram label-impact data (platform-side disclosures, refreshed quarterly).
- TikTok platform disclosures via Storrito industry coverage for AI labelling and removal volumes.
- Nature Scientific Reports for X (Twitter) account-level fake/bot share (peer-reviewed academic study, **2025**).
- Stanford Internet Observatory for LinkedIn GAN-faced profile counts (preprint, DiResta & Goldstein **2024**).
- Europol Innovation Lab and European Parliamentary Research Service for deepfake projections and election-misuse data.
- GPTZero and Originality.ai meta-analyses for detection-tool accuracy benchmarks.
- Imperva 2025 Bad Bot Report (Thales Group) for bot-traffic baselines.

Aggregate platform-share figures are estimates that depend on detector accuracy and sample design. Subreddit and channel-level slices are exact counts from disclosed methodologies. Detector benchmarks are vendor-stated except where independent peer-reviewed studies are noted.

## Frequently Asked Questions (FAQs)

**How much social media content is AI-generated today?**The aggregate share varies by platform. Long-form LinkedIn posts run 53.7% Likely AI per Originality.ai’s 2025 study, Reddit posts sit at 14.7%, and YouTube Shorts surfaced 21% AI-generated content across the first 500 recommended videos per Kapwing. Storrito’s industry coverage reports 52% of TikTok videos involve some AI element. Europol projects that overall online content could reach 90% synthetic by 2026.

 

**Which social platform has the highest share of AI-generated content?**Originality.ai classified 53.7% of long-form LinkedIn posts as Likely AI in its 2025 study of 99 top influential voices, a leading documented share for long-form text content among the platforms covered here. Storrito’s industry coverage reports 52% of TikTok videos involve some AI element such as filters, voiceovers, or full AI avatars, although that includes light AI assistance rather than fully generated content.

 

**How accurate are AI detection tools at finding AI-generated social media content?**Vendor-stated accuracy sits near 99% for the leading tools, with independent benchmarks placing GPTZero at 95.7% on RAID at a 1% false-positive rate, and Originality.ai at 91 to 98% across 13 peer-reviewed academic studies. Real-world accuracy degrades when content is paraphrased, mixed human-AI, or written by non-native English speakers.

 

**How many deepfakes are on social media each year?**Europol projects approximately 8 million deepfakes will be shared in 2025, up from 500,000 in 2023. A deepfake attack occurred every five minutes in 2024, and 49% of companies surveyed across selected countries reported experiencing audio or video deepfake fraud.

 

**How many AI-generated news websites are operating?**NewsGuard identified 3,006 AI Content Farm websites spanning 16 languages, up from roughly 1,265 sites in mid-2024. Industry research suggests 300 to 500 new AI content farm sites emerge each month.

 

**What share of X (Twitter) accounts are bots or AI-driven?**A peer-reviewed study estimated that around 20% of X (Twitter) accounts in its sample were classified as fake or bot accounts. Cyabra research concluded 11% of Twitter users were spam or bot accounts, with one in every five accounts that interacted with Musk’s 2024 election-related posts classified as fake.

 

 

## Conclusion

AI-generated content reached approximately **52%** in May 2025 per Europol, with platform-by-platform shares now running from **14.7%** on Reddit to **53.7%** on LinkedIn long-form posts. TikTok labelled over **1.3 billion** AI-generated videos but auto-detects only **35 to 45%** of AI content, NewsGuard tracks **3,006** AI content farm sites, and Imperva places automated traffic at **51%** of total web activity. Detection tools claim **99%** accuracy in vendor benchmarks but degrade meaningfully on paraphrased, mixed, and non-native English content.

Platform-share figures, labelling-coverage gaps, and detector-accuracy ranges each point to a different operational decision for moderation, exposure, and trust. The baseline matters: AI is now a double-digit share of social content on every major platform measured, and the gap between volume produced and volume detected is the central policy question for the next two years.