Original research
What AI actually changed about making short videos — and what it didn't
We run three faceless channels and have published 283 AI-made shorts since June 2026. Here is what we saw, in plain language.
The short version
| Before AI tools | Now (late 2026) | |
|---|---|---|
| Time to make one narrated short | an evening of work | about 5 minutes of generation, plus review |
| Skills you need | writing, voice, image search, editing, captions | picking a story and checking the result |
| Raw tool cost per short | your time (or paying an editor) | about $0.15–0.55 at public list prices |
| Platform rules on AI | none specific | label AI that looks or sounds real; templated look-alike videos can’t earn on YouTube |
| What decides views | the story, the first seconds, showing up regularly | the same |
What changed: making a video
A narrated short has five parts: a script, a voice, pictures, captions and the edit that puts them together. Until recently each part was a separate job. Doing them all once is a fun evening. Doing them every night is a second job.
Now each of those parts is a tool. A language model writes the script, a voice model reads it, an image model draws the scenes, and software times the captions to the voice. On our setup, the median time from "go" to a finished video is about five minutes of computer time (half of our videos took between 3.8 and 5.5 minutes). The human part that is left is choosing the story and checking the result.
The money side moved just as far. At public list prices, the raw tool cost of one 60-second short made of eight still pictures, about 900 characters of narration and one script is roughly $0.15–0.55, depending on which models you pick. (Prices from fal.ai, ElevenLabs and Anthropic, November 2026.) Real video clips instead of still pictures are a different story: animating a whole minute costs 15–25 times more, so most faceless channels still use still images with camera movement.
Two other things changed as well. The platforms wrote rules: label AI that looks or sounds real, and don’t mass-produce look-alike videos (YouTube won’t pay for templated, interchangeable content, but it explicitly allows using AI to tell a story you invented). And audiences formed opinions: in large surveys people are far more relaxed about AI in entertainment than in news, but most want to be told when it was used. We’ll cover both in their own pieces.
What did not change: views are still earned
Cheap videos did not make views cheap. If anything, the opposite: more than 20 million videos are uploaded to YouTube every day (Alphabet, October 2026), so the competition for each viewer’s first two seconds is fiercer than ever. Three things we measured on our own channels make the point.
A few videos carry the channel. On YouTube, our best 10% of shorts brought in 51% of all views. On Instagram, the best 10% brought in 39%. You cannot tell in advance which ones those will be. That is the real benefit of cheap production: you can afford to try more stories.
The first seconds decide. YouTube shows how many people who saw a short kept watching instead of swiping away. Across our shorts, the median was about 35%. In other words, roughly two in three people decided in the first seconds. Our shorter openings did better: with hooks of 11–14 words, viewers watched 37% of the video on average, against 29% with hooks of 15 words or more.
The story matters more than the tool. On our true-crime channel, the same tools, voice and style produced very different results depending on the kind of story. Unexplained mysteries were watched almost to the end or even rewatched (78–106% of the video on average), while murder cases were watched only 8–17% of the way through.
| What we measured (our channels, Jun–Oct 2026) | Result |
|---|---|
| Share of YouTube views from the best 10% of shorts | 51% |
| Share of Instagram views from the best 10% of shorts | 39% |
| Viewers who stayed past the first seconds (YouTube, median) | about 35% |
| Average watched, hooks of 11–14 words vs 15+ words | 37% vs 29% |
| Average watched, unexplained mysteries vs murder cases | 78–106% vs 8–17% |

And showing up still matters. When we checked 115 faceless channels in September, 47% had not posted for over a month. The channels that kept going were not the ones with the best niche; they were the ones that kept publishing.
What this means if you are starting a channel
- Use AI for the assembly line, not the idea. The part AI made cheap is the production. The part that still decides views is which story you tell and how it opens.
- Spend your saved time on the first two seconds. Read your opening line out loud. If it takes more than about 14 words to get to the strange part, cut it.
- Make more, then keep what works. Because a few videos carry the channel, the cheapest way to find your winners is to publish more stories and look at what people finish, not only what they click.
- Pick a rhythm you can keep. The channels that stopped did not run out of ideas. They ran out of evenings.
How we collected this
Our own numbers come from 283 shorts we published on three faceless channels (mostly horror and true crime) between June 25 and October 26, 2026, using YouTube and Instagram analytics. Only videos at least seven days old are included. We report shares and percentages rather than raw view counts. Generation times are from our own logs.
What this is not. Three channels are a small sample, and they are ours, run by the same person with the same tools. The hook and story comparisons rest on small groups (as few as 7–14 videos per group), so treat them as strong hints, not laws. "Watched more than 100%" happens because YouTube counts rewatches.
Tool prices and platform figures come from the sources linked in the text, checked on November 3, 2026.
Questions people ask
How long does it take to make a short video with AI?
On our setup, generating a finished narrated short takes about five minutes of computer time (median; half of our videos took 3.8–5.5 minutes), plus a few minutes for a person to choose the story and check the result. Doing the same by hand usually takes an evening.
How much does it cost to make an AI short?
At public list prices in November 2026, the raw tool cost of a 60-second short made of eight still images, about 900 characters of narration and one script is roughly $0.15–0.55. Animating the whole minute with an image-to-video model costs 15–25 times more.
Does AI make it easier to get views?
It makes production easier, not attention. On our three channels, the best 10% of shorts brought in 51% of YouTube views, and about two in three viewers swiped away in the first seconds. The story and the opening still decide.
What makes an AI-made short hold viewers?
On our channels, the kind of story and the opening. On our true-crime channel, unexplained mysteries were watched 78–106% of the way through on average (rewatches count), while murder cases were watched 8–17%. Shorter hooks of 11–14 words kept viewers for 37% of the video on average, against 29% for hooks of 15 words or more.