The honest answer, across every controlled study we could verify, is that most people can no longer reliably tell. That includes a blind test Runway ran on its own model, where detection accuracy landed only slightly above a coin flip, and a separate consumer study that found the exact same thing using a completely different methodology.
It also means a meaningful share of what's already in your social feed is AI-generated, whether or not it's labeled as such, and the platforms doing the labeling are candid that they can't keep up with the volume.
Part of our AI Video Generation Statistics 2026 series.
57.1%
Detection accuracy in Runway's own blind study of its Gen-4.5 model, barely above a coin flip.
Only 9.5% of 1,043 participants could reliably tell AI video from real footage.
Our take: Runway published this about its own flagship model. That's either unusual transparency or a sign the industry knows detection is a losing argument either way.
Runway, The Turing Reel01 / 3
How good are people at spotting AI video
The two studies below used different participants, different countries and completely different methodologies, and landed on nearly the same result. That convergence matters more than either number on its own: when a company benchmarking its own model and an independent research firm testing unrelated deepfake awareness both find people performing barely above chance, it's not a fluke of one study's design.
Runway's own "Turing Reel" study (n=1,043): only 9.5% of participants could reliably tell Gen-4.5 video from real footage; overall accuracy was 57.1%.
Our take: Only 9.5% could reliably tell the difference, meaning over 90% of people were essentially guessing.
Runway, The Turing ReelVeriff (research by Kantar, n=1,000 US adults within a 3,000-person global study): US adults scored 0.07 on a -1 to 1 deepfake-detection scale, described as "effectively a coin flip."
Our take: General deepfake awareness, not video-specific, and still barely above zero on a scale where zero is pure chance.
Veriff / Kantar
02 / 3
How much AI video is already in your feed
Kapwing didn't survey opinions here, it created fresh accounts and manually counted what the algorithm actually served, which is a much harder number to argue with than a self-reported estimate. The gap between platforms (59% on TikTok vs. 21% on YouTube Shorts) likely reflects how each platform's recommendation system weights engagement over source, not a real difference in how much AI content exists on each one.
On a brand-new TikTok account, 59% of the videos served to the For You feed were AI-generated (294 of the first 500 reviewed).
Our take: Nearly 6 in 10 videos on a fresh account, this is the default experience for a new user, not an edge case.
Kapwing, TikTok AI Slop ReportOn a brand-new YouTube Shorts feed, 21% of videos were AI-generated (104 of the first 500 reviewed).
Our take: Under half of TikTok's rate, worth asking whether that's less AI content overall or just a different algorithm surfacing less of it.
Kapwing, TikTok AI Slop ReportIn TikTok's Kids category specifically, 57.4% of videos were AI slop (1,147 of 2,000 reviewed).
Our take: Roughly the same rate as the general feed, meaning Kids content isn't getting any extra filtering for AI-generated material.
Kapwing, TikTok AI Slop ReportTikTok has applied AI-content labels to over 1.3 billion videos to date, a cumulative count, not a share of all content.
Our take: Cumulative since the policy started, not a current snapshot, don't read this as "1.3 billion AI videos exist right now."
TikTok Newsroom
03 / 3
Consumer trust in AI-made content
Three separate consumer surveys, run by three unrelated organizations, all converge on the same pattern: awareness of AI content is now nearly universal, but trust in it hasn't followed. That's a real headwind specifically for advertising, where the Harris Poll's numbers suggest disclosure itself, not just quality, is what erodes trust once a viewer suspects AI involvement.
63% of consumers say they'd be less likely to buy from a brand using AI-generated ads; 73% less likely to trust an ad they suspected was AI-made; 78% say AI makes ads feel "less authentic."
Our take: 78% saying AI makes ads feel less authentic is the number advertisers should worry about most, it's about the medium itself, not any one bad execution.
The Harris Poll (with 4As and Infillion)53% of consumers say they're less likely to buy from a brand they know uses AI in social content; 86% say they've encountered AI-generated content on social media.
Our take: 86% having already encountered AI content means the awareness battle is basically over, the trust battle is what's left to fight.
Clutch.coDeloitte's 2026 Digital Media Trends (n=3,575): ~24% of fans want to co-create content with GenAI, like alternative endings; roughly a third say it doesn't matter to them whether ads use GenAI.
Our take: Only a third being indifferent means two-thirds have an opinion, and based on the other two stats here, it's mostly not a positive one.
Deloitte, 2026 Digital Media Trends
Frequently asked questions
Can people tell AI-generated video from real footage?
Not reliably. Runway's own blind study found viewers correctly identified AI video only 57.1% of the time, barely better than chance, and a separate Kantar/Veriff study using a different methodology found US adults scored close to a coin flip on a standardized deepfake-detection test.
What percentage of social media video is AI-generated?
On a brand-new TikTok account, Kapwing's AI Slop Report found 59% of videos served to the For You feed were AI-generated, versus 21% on YouTube Shorts. TikTok says it has applied AI-content labels to over 1.3 billion videos to date.
Do consumers trust AI-generated ads?
Largely no. The Harris Poll found 73% of consumers would be less likely to trust an ad they suspected was AI-made, and 78% said AI makes ads feel less authentic. Trust is the biggest headwind facing AI video adoption in advertising specifically.
Are platforms doing anything to label AI video?
Yes, TikTok has applied AI-content labels to over 1.3 billion videos, but Kapwing's independent audit still found a majority of a new account's feed was AI-generated, suggesting labeling hasn't caught up with volume.