Can People Actually Tell an AI-Edited Video From a Human-Edited One?
We showed 315 people the same footage edited two ways, once by a professional human editor, once entirely by AI, and asked them to spot the difference. Most couldn't.
📌 A Note on This Study
Editvideo.io is a video editing company. Our human editors are one of the two subjects being compared here. We built this study to be as fair as we could make it: identical raw footage, one locked AI prompt, no manual cleanup on either side. We're publishing the full methodology, the full dataset, and the results that don't flatter us right alongside the ones that do. Judge the design, not just who paid for it.
Watch Both Videos First
Before you see a single number, watch what we're actually comparing. Same raw footage, same script, two different editors.
One of these was edited by a professional human on our team. The other was edited entirely by Descript's Underlord AI co-editor, from a single prompt. Watch both, make your guess, and keep scrolling to see if you beat the 315 people we surveyed.
Featuring Brittany Flammer
The raw footage used in this study comes from Brittany Flammer, a personal finance creator and Editvideo.io client who helps parents ditch debt and build wealth while raising money-smart kids. She and her husband grew their net worth to nearly $1,000,000 on an average household income of $40k/year, proof, as she puts it, that "it's not about how much money you make, it's how you manage what you make."
Watch her channel →Why We Ran This
AI video editing tools are good now. Good enough that we, a company that sells human video editing, wanted to know exactly how good. Not "can AI edit a video," because obviously it can. The sharper question is whether ordinary people can tell the difference when they watch a human edit and an AI edit side by side, and whether it even changes how they feel about the video.
We're publishing this because the honest answer surprised us in two directions at once. Humans still win, clearly, on the things that matter for a business: engagement and trust. But AI editing has gotten good enough that most people can't spot it when you ask them to.
Video A was edited by a professional human editor on our team. Video B was edited entirely by Descript's Underlord AI co-editor, from one prompt, with zero manual touch-ups. Now let's see how they actually performed.
Human Editing Still Wins on Engagement, Clearly
248 of 315 respondents, 78.7%, found the human-edited video more engaging. PickFu flags this as statistically significant even in the full run alone (234 of 300, 78.0%).
The reasoning came down to two things. On-screen graphics and text made the video easier to follow and hold attention (30.5% of respondents cited this directly), and the human edit's audio sounded clearer and more natural to people (24.1% cited this).
"The graphics that show up on the screen are eye catching and therefore keep attention. Bulleted lists help solidify what is being said."
A minority, 21.3%, actually preferred the AI edit because it was plainer:
"By using less pop-ups and flash bang, you actually listen to her message, which made you think about what she was saying and the importance of it."
Human Editing Also Wins on Trust and Subscribe Intent
199 of 315 respondents, 63.2%, said the human-edited video felt more trustworthy and made them more likely to click subscribe or follow.
10.2% of respondents used words like "genuine," "authentic," or "real" specifically to describe the human edit. 7.9% pointed to "professional" or "polished" as their trust signal.
"Comes across as more mindfully and intentionally made, which feels trustworthy and grounded."
A smaller group, 36.8%, actually read the AI edit's plainness as more honest, not less:
"It looks like they didn't try as hard, which feels more honest, actually."
Most People Couldn't Tell, and When They Guessed Wrong, They Guessed Backwards
This is the headline finding, and it's the one that actually surprised us.
When we told respondents one video was AI-edited and one was human-edited and asked them to guess which was which, 168 of 315 people, 53.3%, guessed wrong. They thought the AI edit was the human one.
The reasoning behind the wrong guesses was remarkably consistent. Most people assumed simpler editing means a human did it, and heavier production means AI did it. In our case that assumption ran exactly backwards. Our human editor is the one who added the branded motion graphics and yellow highlighted captions. The AI kept its captions simple and minimal.
"I would think the simplest video, video B was done by a human editor as there is very little that needed to be done to it other than add subtitles. Video A looks like the AI edited version with all the edits and visual elements added."
"It seems like AI 'jazzed' the other one up with the extra flair and a human did this one."
Plenty of people also just told us outright they were guessing:
"I honestly can't tell which one it is. You asked to pick one so I took a guess."
Some respondents' own reasoning even contradicted their vote. One person wrote that a video "was edited by AI because it is so simple," then voted that same video as the human edit. We're leaving that in rather than cleaning it up, because it's honest proof that a lot of these guesses were closer to a coin flip than a confident read.
Higher Income and More Education Didn't Help People Spot the AI, It Made Them Worse at It
We broke question three's accuracy down by demographics across all 315 respondents, and the pattern ran the opposite way from what we expected.
Accuracy went down as income went up. The highest earners in our sample were the least likely to correctly identify the human-edited video.
"B was 100% human edited and it shows. It flowed like a sane conversation, it felt natural. A was all over the place, felt weird and obnoxiously bad. It was obvious because no human would have made A as bad as it was."
— $75-150k household income, guessed incorrectly, stated with full confidence
Education showed a similar pattern, though less of a straight line. Respondents with a graduate degree were noticeably less accurate (34.8% correct) than those with a bachelor's degree (51.0% correct) or a high school education (45.5% correct).
"I would think that choice B was human edited because it was just the video and subtitles. I would think AI would be punching it up with the graphics like in choice A."
— Graduate degree, guessed incorrectlyWho We Surveyed
315 US-based respondents, a near-even gender split (50.5% male, 49.2% female), concentrated in the 35-54 age range, spread across income and education brackets with no demographic filtering applied. We wanted a general read, not a niche one.
What This Means for Your Channel
Professionally edited video still wins where it counts. Engagement and trust, the two things that actually drive subscribers and business outcomes, both favor human editing by a wide margin.
AI editing has crossed a line most people can't see past. The visual and audio giveaways that used to expose an AI edit aren't a reliable tell anymore for an average viewer.
Polish doesn't mean AI, and plainness doesn't mean human. Most viewers have this backwards. That assumption cost over half our respondents the right answer.
If you're using AI editing tools, know what you're trading. It might fool a casual viewer. It's still measurably losing on the things that make someone stick around and hit subscribe.
Where This Study Has Limits
We're presenting these results as directional, not final.
- One creator, one video, one niche (personal finance). We'd expect some variation with a different style of content.
- One AI tool (Descript Underlord) and one prompt. A different tool, or a more detailed prompt, could change the outcome.
- A general PickFu audience, US only, with no demographic filtering. That was a deliberate choice for a broad claim, but it means we can't speak to how any one specific audience would score.
- The 315 total respondents came from two survey waves (15 and then 300) run with identical setup. We reported them combined because the results held steady between the two waves.
Full Report & Press Resources
Everything above is the summary. The full write-up, methodology, and dataset are available below for anyone who wants to dig deeper or cite this study.
Frequently Asked Questions
Can viewers tell if a video was edited by AI or a human?
In our 315-person study, most viewers could not reliably tell. 53.3% incorrectly identified the AI-edited video as the human-edited one.
Does AI-edited video perform as well as human-edited video?
Not in this study. The human-edited video was preferred for engagement by 78.7% of respondents and for trustworthiness by 63.2%, even though most people couldn't correctly identify which video was AI-edited.
What AI video editing tool was used in this study?
Descript's Underlord AI co-editor, using one natural-language prompt with default settings and no manual edits afterward.
Why did people mistake the AI edit for the human edit?
Most respondents assumed that simpler, less graphics-heavy editing meant a human editor, and that more polished, graphics-heavy editing meant AI. In this study that assumption was backwards.
Want Video That Actually Holds Attention?
The data says engagement and trust. If you want video that earns a subscribe, that's what our editors are for.