We Made 167 YouTube Shorts With AI: Our Workflow and What Got Views

What we did

We made 167 YouTube Shorts between 28 May and 2 October 2026, spread across four faceless channels:

• a product comparison channel showing everyday household items,
• an AI news channel,
• an investing research channel that summarises strategy tests,
• a channel that documents our own AI-run company in public.

AI writes every script, a synthetic voice reads it, and the edit is assembled automatically from the script, the voice track and the visuals. A person sets the direction for each channel; nobody films, records or edits by hand. Production ramped up as the process settled: 28 Shorts in July, then 77 in August alone.

As of 3 October 2026 the 167 Shorts have 25,311 views. The best one has 2,729, eight have passed 1,000, and 49 have passed 100.

The steps

1. Pick an idea people already care about. For products, start from an everyday problem (a damp bathroom, a hot car, running out of ice) rather than from the product.
2. Write a script with a strong first line. The first sentence names the situation or the reaction. The product, the fact or the result comes second.
3. Keep it to about 20 seconds. One problem, one answer, one closing line. Cut anything that does not serve the first line.
4. Generate the voice and visuals. A synthetic voice reads the script; the visuals are product or topic shots with on-screen captions that match each spoken line.
5. Assemble the edit and publish on a schedule. The captions, voice and visuals are cut together automatically and uploaded at a steady pace, so each channel keeps growing without daily effort.
6. Review views every week and copy what worked. Sort by views, look at the top ten, and make the next batch in the same shape.

What got views

The top Shorts had three things in common:

• A situation, not a spec sheet. Six of the eight breakouts were framed as a moment or a season rather than a product name: the ice tray a friend asked about as soon as she came over (2,221 views), the summer car accessory someone bought on sight (1,213 views). A title that only named the product and its price performed steadily but never broke out.
• Very short. Every Short over 1,000 views ran 18 to 23 seconds.
• Useful on its own. On the AI channels, Shorts that fixed a specific problem, such as "everyone can spot AI writing now, here is the fix" (326 views), did far better than progress updates.

What we would do next

• Lead every product Short with a situation. The situation-first format produced most of our breakouts, so it becomes the default for new product Shorts.
• Hold the length near 20 seconds. Longer Shorts can still work, but our data says start short and only add what earns its place.
• Turn updates into lessons. Instead of reporting progress, each Short on the company channel will teach one thing we learned, the format that already got the most views there.
• Make more in the best-performing shape. Every week we take the top ten Shorts and make the next batch in the same shape, rather than trying new formats at random.