Fact checking a documentary Short

A documentary Short built by ViewMade ships with a media credits file that names the source of every clip in the render, alongside a Starter plan priced at $29 per month for 50 renders, or $0.58 per Short, measured on 22 August 2026. This article explains how fact checking works on a Short like that, which parts of the pipeline can be checked against an external source, and which parts cannot. It also gives you a procedure you can run yourself before you publish.

The short answer

Fact checking a documentary Short means verifying three layers separately: the claims in the script, the provenance of every visual clip, and the captions that carry both. On a ViewMade Short, the script is researched with sources, the visuals come from real archive material rather than a video model, and the media credits file lists where each clip came from. You check the script against primary sources, check each clip against its credited origin, and confirm the subtitles match the narration word for word. If any layer fails, you fix it before upload, not after.

What a source verification pass actually covers

Source verification on a documentary Short is not one step. It is three checks that fail independently, and treating them as one is how errors survive to publication.

The first check covers the script. Every factual claim in the narration needs a named source: a date, a document, a recording, an official report. The research stage attaches these during writing, so the check is a matter of confirming the claim matches what the source says, not finding a source after the fact.

The second check covers the visuals. A Short about a real event uses archive footage, maps, and charts. Each clip carries an entry in the media credits file naming its origin. The check confirms the file exists, matches the credited source, and shows what the credit says it shows. Footage of the wrong year, the wrong aircraft, or the wrong coastline passes a casual watch and fails this check.

The third check covers the captions. Captions are burned into the frame word for word, so a transcription error becomes permanent once rendered. The check compares the caption track against the final narration audio.

The table below summarizes the three layers:

Layer	What it contains	How it is checked
Script	Narrated claims with attached sources	Claim read against its named source
Visuals	Real archive clips, maps, charts	Clip matched to its entry in the media credits file
Captions	Word-for-word subtitle track, burned into the video	Caption text compared against final narration

A Short that passes all three is publishable. A Short that passes two is not, because viewers cannot tell which layer failed.

Why generated footage changes the question

Most tools in this category generate their visuals. Revid produces high-volume short-form output with generated visuals. InVideo combines generation and stock. Pictory draws from a stock library. StoryShort uses web-sourced imagery, and our own comparison table records its source citation as partial. For a fictional topic, generation raises no accuracy problem at all, because nothing in the video claims to be a record of something that happened.

For a documentary about a real event, generation creates a specific failure mode: the video looks like evidence but is not. A model asked to depict a 1970s airliner will produce an aircraft that resembles one, and a viewer has no way to know the difference. There is no original to compare against, because the clip never existed outside the model. Fact checking therefore cannot operate on the visual layer at all; it collapses into script-only checking.

This is why ViewMade pulls footage from real archives, maps, and charts instead of generating it, and why every render arrives with the media credits file. The credit converts each clip from an uncheckable image into a checkable claim: this footage comes from here, go look. Verification stays possible because the visual layer keeps a verifiable relationship to the world.

The honest limitation sits on the other side too. Archive sourcing does not make a Short accurate by default. A correct clip cut into a misleading sequence, or paired with a wrong date in the narration, is still wrong. Sourcing makes the visual layer checkable; it does not do the checking for you.

How to fact check a Short before you publish

If you have a finished Short in hand, run this sequence. It takes less time than re-rendering after a public correction.

1. Read the script alone, without the video. List every claim that states a fact: a date, a number, a name, a cause. Skip atmosphere and framing sentences.
2. For each listed claim, open the source attached to it in the research stage. Confirm the claim matches. Mark any claim whose source you cannot open or that the source does not clearly support.
3. Open the media credits file. For each entry, confirm the clip in the render is the clip the credit describes. Watch for lookalikes: the right aircraft type in the wrong livery, the right city in the wrong decade.
4. Play the final render with captions visible. Compare the burned-in captions against the narration audio line by line. Any mismatch is a re-render, because the captions are baked into the frame.
5. Fix failures at the layer where they occurred: rewrite the script claim, swap the clip, or regenerate the caption track. Then re-run steps 2 through 4 on the changed parts only.

Two habits catch most errors early. Check the first thirty seconds hardest, because that is where retention is decided and where a wrong claim does the most damage before anyone pauses to doubt it. And check anything involving numbers twice, since a transposed digit survives every casual review.

What this does not tell you

We have not measured how often a fact-checked Short still contains an error that reaches publication, because we have no data on post-publication corrections and will not invent a rate. We have not measured how long the verification sequence above takes per Short, so treat any time estimate you form from this article as your own experiment, not ours.

The procedure also assumes the credited source itself is reliable. If an archive mislabels its own footage, matching the clip to the credit confirms consistency, not truth. We have not audited upstream archive labeling and this article does not cover how to do it.

Finally, this article answers how to verify a Short, not whether a given niche tolerates lower verification effort. That tradeoff depends on your audience and your risk tolerance, and we have not measured it.

<!-- faq -->

Frequently asked questions

Can I fact check a Short made entirely with AI-generated visuals?

You can check the script and the captions, but the visual layer offers nothing to verify against. A generated clip has no external origin, so there is no source to compare it with. Your check reduces to confirming the narration is accurate and accepting that the imagery is illustrative rather than evidential. State that distinction in your description if the topic is real events, because some viewers will assume otherwise.

Does the media credits file prove a clip is authentic?

No. It proves the clip came from the stated source, which is a different claim. Authenticity depends on the source having labeled its material correctly in the first place. Treat the credit as a pointer that makes checking possible, then do the checking by comparing the clip against what the credited source says it depicts.

What should I do when the script and the credited footage disagree?

Trust neither automatically. Re-check the script claim against its named source first, because scripts are cheaper to fix than renders. If the script is right and the footage is wrong, replace the clip and note the correction in the credits. If the footage is right and the script is wrong, rewrite the narration line and regenerate the affected segments. Never publish knowing the two disagree.

How often should captions be checked if they are generated automatically?

Every render, without exception. Captions are burned into the frame word for word, so an error ships permanently and cannot be patched after upload. The check is mechanical: play the render, compare the caption text against the narration audio, flag mismatches. Because it is mechanical, it belongs at the end of every production run regardless of how confident the pipeline is.