Long faceless documentaries, the 20-minute to one-hour videos about history, disasters, science or how something was built, have become a common format on YouTube. Many of them are now made with AI narration and AI-made or stock-style pictures. This guide explains, step by step, how one of these videos is put together, where AI helps, and where a person still needs to check the work.
It is written for creators deciding whether to make documentaries themselves, hire an editor or use a service. The steps are the same either way.
1. Pick a topic with a story in it
A documentary needs more than a list of facts. It needs a beginning, a turning point and a consequence: why something happened, what it cost, who was affected and what changed afterward. "The Transcontinental Railroad" is a subject; "How two companies raced to build the Transcontinental Railroad, and who paid for it" is a story.
Before any script is written, it helps to check that the topic has enough reliable sources and enough to show on screen for the length you want. A one-hour video needs far more material than a ten-minute one.
2. Research and an outline in chapters
Long videos are easier to make, and to watch, when they are split into chapters. Each chapter covers one part of the story and ends with a reason to keep watching. A one-hour documentary might have five or six chapters of about ten minutes each.
AI research tools can gather background quickly, but they can also mix up dates, names and numbers or state guesses as facts. Good practice is to keep a short source list for each chapter and check every name, date and number against at least two reliable sources, such as museum and university sites, government archives, encyclopedias and published books.
3. The script
Narration is usually written at a speaking pace of around 150 words a minute, so a ten-minute chapter is roughly 1,400 to 1,600 words. Things that make a documentary script work:
- An opening that sets up the question the video answers, without giving everything away.
- Short, spoken sentences. Writing that reads well on a page often sounds stiff out loud.
- Specific details (a street name, a price, a quote from a letter) instead of general statements.
- No repeated sign-offs in the middle. In a long video, "don't forget to subscribe" every few minutes breaks the story.
- Care with real people. Victims of crimes and disasters were real; describe what happened with respect and without graphic detail.
4. AI narration
Text-to-speech voices have improved to the point where a well-written script, read by a good AI narrator, is pleasant to listen to for an hour. A few practical points:
- Use one voice per channel. Viewers come back to a channel partly for its voice, so it should sound like the same show every time.
- Listen to the whole thing. AI voices sometimes mispronounce place names, old units or foreign words. Spelling a word out the way it sounds in the script usually fixes it.
- Do not copy a real person's voice without their clear permission. A cloned voice of a real person also needs YouTube's altered content disclosure; a generic AI narrator reading your script usually does not.
5. Pictures: AI-made, archival and stock-style
A faceless documentary changes picture every few seconds, so an hour-long video needs hundreds of images or clips. They usually come from a mix of sources:
- Public domain archives. In the US, works published before 1931 are generally in the public domain as of 2026, and many US federal government photos and documents are too. The Library of Congress and the National Archives are good places to start; check the rights note on each item.
- Stock libraries. Licensed stock photos and footage, used within the license terms.
- AI-made pictures. Useful for scenes nobody photographed. For historical topics, the prompts need to name the period, so that a story set in 1869 shows 1869 clothes, buildings and vehicles, not modern ones.
- Maps and simple graphics for routes, borders and numbers.
A realistic AI picture of a real event can mislead viewers into thinking it is a photograph. Use such pictures as illustration, avoid inventing realistic "photos" of real people, and disclose them on YouTube when they look real (see our checklist before posting AI videos).
6. Motion, edit and captions
Still pictures come alive with slow pans and zooms (often called the Ken Burns effect), cross-fades and, where the budget allows, short AI-generated motion clips. The edit follows the narration: each picture should match what is being said at that moment.
Captions help viewers who watch without sound or in a second language. A separate captions file (SRT) can also be uploaded to YouTube so the platform does not have to guess the words.
7. Music and sound
A quiet music bed under the narration sets the mood. It should stay well below the voice, change between chapters if the tone changes, and come from a source you have the right to use. Music rights are the most common cause of copyright claims on these videos; the posting checklist covers where to get music safely.
8. Chapters, title, thumbnail and description
YouTube shows chapters on the progress bar when the description lists timestamps: the first must be 00:00, there must be at least three, and each chapter must be at least 10 seconds long. Chapter titles help viewers jump to the part they care about.
The title and thumbnail should promise what the video actually delivers. A dramatic but accurate title works better over time than one that viewers feel tricked by.
9. A final watch-through before posting
Before posting, someone should watch the whole video once at normal speed, checking for wrong facts, mispronounced names, pictures that do not match the narration, modern objects in historical scenes, and audio that is too loud or too quiet. It is the least automated step and the one that catches the most problems.
Making documentaries yourself or having them made
Every step above can be done by one person with the right tools; it just takes time, often several days for an hour-long video. If you would rather have them made, we make documentaries and ten-minute videos on a monthly plan: research, script, AI narration, pictures, motion, captions and music, with chapter markers on long documentaries. You watch each one, ask for changes in plain words, and post it yourself.
Still choosing a subject? Start with how to choose a faceless YouTube niche.