Documentary Script Example: From Research to Scenes

You have a topic, a pile of open tabs, and a rough sense of the story. What you don't have yet is a plan for what to say, what to show, and how you know each line is true. That plan is your documentary script.

A documentary script ties every spoken line to a picture and every factual line to a source. The second half matters more on YouTube than ever. Viewers check your facts in the comments, and YouTube asks creators to disclose realistic AI-generated scenes.

The fastest way to learn it is to watch one get built. So this guide takes a single real topic from the first question to finished scene plan.

Documentary Script Example Used Throughout This Guide

Each step below explains the method first, then shows it applied to the molasses film, then gives you a blank template where one applies.

What You Will Build

1. A documentary angle, written as one question the film answers
2. A source log that tracks every factual claim
3. A paper edit that turns research into story beats
4. A two-column audio and visual script, with a source column and a visual-status column
5. A revision log for when the evidence shifts
6. A YouTube description built from the same research

Step 1: Pick a Documentary Angle Before You Research Deeply

A topic is what the film is about. An angle is the question the film answers. "The Great Molasses Flood" is a topic. It could become a disaster recap, a courtroom drama, or a physics lesson. Pick one before research swallows your week.

For this example, the angle is:

Why did a thick, slow liquid kill 21 people, and why was it so hard to pull survivors out?

That question does three jobs. It creates tension, because molasses and "fast and deadly" don't belong in the same sentence. It points to evidence that exists: a 2016 fluid dynamics study found that the cold January air thickened the spreading molasses and made rescue harder. And it tells you what to leave out. The Prohibition backstory is interesting, but it only earns a line or two.

Angle template

Topic: [the event, person, or idea]

Angle question: Why / how did [surprising outcome] happen?

What makes it surprising: [the contradiction the viewer feels]

Evidence that can answer it: [studies, records, archives, experts]

What this angle cuts: [interesting material you will NOT cover]

Run one test before moving on. If you can't name at least two sources that could answer your angle question, change the question. A great angle with no evidence becomes a script full of guesses.

Step 2: Build a Source Log for Factual Claims

A source log is a table with one row per factual claim. Each row records where the claim comes from and how confident you are. It exists so every fact in the final script can be traced back in seconds.

Journalists call a version of this an annotated draft, where each statement is tied to a source a fact-checker can verify. Documentary fact-checkers want the same thing. One told the International Documentary Association that the best script a filmmaker can bring is one fully annotated with its sourcing.

Here is the log for the molasses film. Notice how often trusted sources disagree.

Source log template

Three rules keep the log honest:

• One claim per row: "21 dead and 150 injured" can stay together only because the same sources report both.
• Write to the confidence level: An estimate gets estimate wording. A conflict gets vague wording or gets cut.
• Log the claims you reject: C10 stops a dramatic but shaky word from sneaking back in during the edit.

Step 3: Organize Research Into a Paper Edit

A paper edit is your film assembled on paper before anyone touches a timeline. Filmmakers usually build it after shooting, from transcripts and a "selects" list that records the file, timecode, and speaker for each chosen moment. Some teams lay it out like a spreadsheet, with columns for source, timecode, visuals, audio, and notes.

If you make voiceover-led videos, your "footage" is your research. Your selects are archival photos, documents, published expert findings, and the claims in your source log. The job stays the same: put the strongest material in an order that answers your angle question.

The molasses film targets about 8 minutes, split roughly 25 / 50 / 25 across three acts. That split follows the three-act guide Documentary Film Academy recommends.

Paper edit template

The "Job in the story" column is the one to guard. If you can't say what a beat does for the angle, cut it. Beat 2 almost lost its place until C11 gave it a real job: the warning signs everyone ignored.

Step 4: Use a Two-Column Audio and Visual Template (Then Add Two Columns)

The classic documentary script template has two columns. Visuals sit on one side, and the audio that plays over them sits on the other, lined up row by row. Start a new row whenever the picture changes.

For factual videos, two columns leave out two questions a viewer, an editor, or a fact-checker will ask. Where did that line come from? And is that picture real? So this template adds a Source column and a Visual status column.

Documentary script template (four columns)

To time a draft, count narration words. Most creators speak between 130 and 160 words per minute. An 8-minute film lands near 1,000 to 1,300 words of narration, minus the space you leave for sound and archive audio.

Step 5: Separate Interviews, Archives, and Narration

Tag every audio line by where it comes from. Viewers trust these voices differently, so your script should never blur them.

• NARR: your narration. You wrote it, so every factual sentence needs a Source ID.
• INT: an interview you recorded. Log the file name and timecode.
• ARCH: audio or text from the historical record, like a newspaper headline read aloud or a court transcript line. Log where it lives.
• SFX / MUSIC: sound design. No source needed, but note anything AI-generated (Step 6 explains why).

The same logic applies to pictures. A 1919 photo and a modern street shot are both "real," but only one shows the event. The Visual status column keeps that difference visible all the way to export.

Step 6: Label Reconstructions and Illustrative Visuals

Most history videos need pictures that don't exist. Nobody filmed the molasses wave. You will fill that gap with one of two things, and they need different labels.

YouTube requires creators to disclose AI content that generates a realistic scene that didn't actually occur, or that alters footage of a real event or place. A photorealistic AI wave in 1919 Boston fits that description. YouTube also lists AI-generated music as something to disclose. Clearly unrealistic content and production help, like using AI to draft a script or thumbnail, does not need disclosure. Creators who repeatedly skip required disclosure risk labels being added for them, content removal, or suspension from the YouTube Partner Program.

The documentary industry goes further. The Archival Producers Alliance's best practices suggest a temporary watermark on generative AI material during editing, plus audience-facing disclosure through lower thirds, narration, or top-of-show language. They also recommend a cue sheet listing every generative AI element.

For a solo creator, that translates to three habits:

1. Mark RECON and ILLUS in the script, so the label survives into the edit.
2. Put a small on-screen label on every RECON shot, like "Reconstruction" or "AI-generated illustration."
3. List reconstructed and AI-generated elements in the video description.

Step 7: The Worked Example, From Research to Scene Plan

Here is the documentary script example itself, written in the four-column template. It covers the key rows of beats 1 through 6. Every narration line matches the "approved wording" column of the source log.

A few choices are worth copying:

• Row 4 is the only RECON shot. It carries an on-screen label, and the narration beneath it uses estimate wording. The picture shows what the accounts describe. It doesn't pretend to prove them.
• Row 5 turns a sourcing problem into a story beat. The early headline undercounted the dead. Instead of hiding that, the script uses the gap between first reports and the final toll.
• Rows 2, 3, and 7 are ILLUS, not RECON. They are stylized on purpose, so no viewer mistakes them for 1919 footage.
• This film has no interviews. If you add one, it gets its own INT row with a file name and timecode in the Source column.

Step 8: Revise the Script as Evidence Changes

A documentary script is never locked while you are still learning. Information that looked solid during research can turn out irrelevant or false later, so every claim needs checking right up to the end.

Published sources revise too. The Live Science report on the 2016 molasses study carries an editor's note saying it was updated to correct the temperature of the molasses. If you had copied the first version into your script, your film would now be wrong.

Claim IDs make revision fast. When a fact changes, you fix the log row first, search the script for that ID, and rewrite every row that cites it. Then you record the change.

Here is how the molasses script changed across its drafts:

Revision log template

If the video is already live, update the sources and corrections in the description, and pin a short correction comment when the change matters. The source log tells you exactly what to fix.

Step 9: Turn Your Source Log Into the YouTube Description

The description is the last page of your documentary script. It is where viewers check your sources, where your AI disclosure lives, and where search engines learn what the video covers. This is the end-to-end playbook: why it matters, how to structure it, where keywords go, what to link, and how to write it in the same session you generate the video.

Why the Description Matters

• It tells YouTube what the video is about. YouTube says descriptions inform both the algorithm and viewers, and that search terms are matched against the video and its description text.
• The first lines do most of the work. Viewers see only the opening lines until they tap "Show more" on desktop or "more" on mobile.
• It proves the research is real. A documentary without sources asks for blind trust. A linked source list lets curious viewers check you, and it separates your work from low-effort uploads. YouTube's spam policy specifically targets channels that mass-produce near-identical AI videos.
• It carries your disclosure. Step 6 flagged every RECON shot and AI music cue. The description is where viewers can read that list.

Description Structure: Best-Practice Order

YouTube allows up to 5,000 characters and recommends a unique description for every video. Use the space in this order:

1. Opening lines (above the fold). One or two sentences with your main keyword and what the viewer will learn. Third-party guides treat the first 100 to 160 characters as the visible window, since it varies by device.
2. Short summary. Two or three sentences that restate your angle question. Work in one or two secondary keywords where they read naturally.
3. Chapters. Timestamps with titles. YouTube requires the first to start at 00:00, at least three in ascending order, each chapter at least 10 seconds long.
4. Sources. The confirmed rows of your source log, with links.
5. Visuals and AI disclosure. Which shots are reconstructions or AI-generated, plus any AI music. Also set "AI use" to Yes in YouTube Studio when your content meets the disclosure rules.
6. Links and one CTA. A related video or playlist, credits for collaborators, and a single ask.
7. Hashtags. Two or three relevant ones at the end. Up to three appear near the title, and a video with more than 60 hashtags has all of them ignored.

Links and CTA

Keep one primary call to action. Everything else is navigation: a playlist of related videos, credits for people who helped, and your channel links. YouTube recommends linking playlists and mentioning collaborators for exactly this reason. Put the CTA after the sources, so trust comes before the ask.

YouTube Description Template

[Main keyword] + [what the viewer learns, in one sentence].

[Second sentence: the surprising angle, in plain words.]

[2 to 3 sentence summary. Restate the angle question. Use 1 to 2 related terms naturally.]

CHAPTERS

00:00 [Beat 1 title]

[mm:ss] [Beat 2 title]

[mm:ss] [Beat 3 title]

SOURCES

- [Source title], [publisher]: [link]

- [Source title], [publisher]: [link]

ABOUT THE VISUALS

Archive: [where historical images come from]

Reconstructions: [which scenes are staged or AI-generated]

AI audio: [AI music or voice, if used]

WATCH NEXT: [related video or playlist link]

[One CTA]

#[hashtag1] #[hashtag2] #[hashtag3]

Filled Example: The Molasses Film

The Great Molasses Flood of 1919 killed 21 people in Boston. This documentary explains how a slow, sticky liquid turned deadly.

On January 15, 1919, a 50-foot tank in Boston's North End split open and released about 2.3 million gallons of molasses. We follow the physics of the wave, the cold that trapped survivors, and the court case that decided who was to blame.

CHAPTERS

00:00 A warm lunch hour

00:45 The tank that leaked

02:00 The rupture

03:15 How a slow liquid moved fast

04:45 The cold trap

06:00 Who was to blame

07:15 What the flood left behind

SOURCES

- Boston.gov: 100 years ago today, molasses crashes through Boston's North End

- APS Division of Fluid Dynamics (2016): Sharp, Kennedy & Rubinstein, the physics of the Boston Molasses Flood

- Mass Moments: Hugh Ogden issues report on cause of the Molasses Flood

- Boston Post, January 16, 1919, via Boston Public Library

ABOUT THE VISUALS

Archive: 1919 photographs and newspaper scans, credited on screen.

Reconstruction: the wave sequence at 02:00 is AI-generated and labeled on screen.

Illustrations: the tank and cooling sequences are stylized, not historical footage.

WATCH NEXT: [your history playlist]

#history #boston #documentary

Check the chapter times against your final export before publishing. The paper edit gives you targets, and the edit decides the truth.

Write It While the Video Generates

You don't need a separate writing session. Every part of the description already exists in your script files:

So draft it in the same sitting you produce the video. Paste your finished script into your video tool, start generation, and fill the template while it renders. When the edit is final, lock the chapter timestamps and publish both together.

From Script to Video With ByThen

A finished four-column script is a production plan. ByThen is an AI video producer built for long-form storytelling, and it can turn that plan into a video up to 30 minutes long. Script, voice, music, visuals, and editing all happen in one place.

Here is how the script maps onto it:

• Your approved wording stays approved. Paste your own final script and ByThen visualizes it, instead of rewriting it from a topic.
• The storyboard is your label check. Review the film scene by scene before full generation. ILLUS scenes should look stylized. RECON scenes need a label.
• Archive stays real. Upload your licensed photos and newspaper scans in the editor and place them on the ARCHIVE and DOCUMENT rows.
• Labels go on in the editor. Add a text overlay like "AI reconstruction" to every RECON shot. Captions are generated automatically.
• Disclosure follows the audio too. If you use generated music, list it under "About the visuals" and answer YouTube's AI use question honestly.

The research is yours. The production doesn't have to take a week.