How to Make Explainer Videos With AI (Without Letting Errors Slip In)

The main risk: AI sounds sure even when it's wrong

An AI assistant writes with the same confidence whether it's right or has made up a number, a date or a quote. In an explainer video that mistake costs more, because a polished voiceover and clean visuals create an impression of authority viewers rarely question.

That's why the order matters: sources first, then script, then visuals and voice. Reverse it and you end up fact-checking a video that's already built, which is when errors are easiest to wave through.

Pass 1: gather sources before you prompt anything

Pick two or three primary or well-established sources on your topic: official reports, the organization's own documentation, specialist publications, the original article. Read them yourself. The AI can help summarize them, but it can't replace your reading.

I'm going to paste the text of a source. Work only from that text and don't add anything from your general knowledge.
Extract the key claims as a numbered list. For each one give me: the sentence from the source that supports it (copied exactly), whether it's a fact, an opinion or an interpretation, and what would be left unsupported if I removed it.
If something isn't in the text, write 'not in the source'.

Source: [paste the text and the link here]

Pass 2: build a claim-and-check table

Before writing a script, collect everything the video will assert and how you'll confirm it. This table becomes the backbone of the video and saves you from deciding mid-edit what stays and what goes.

I'm making a [length] explainer video about [topic] for [audience].
Using the claims extracted from my sources (pasted below), build a table with these columns: claim, source where it appears, how I can verify it (which document or official data to check), confidence (confirmed / uncertain / single source) and whether the video needs it.
Don't add new claims. If sources contradict each other, point it out.

Claims and sources: [paste here]

Cut anything you can't back up with a second reliable source or the original document. A shorter, solid video beats a long one with one shaky number.

Pass 3: write a script that stays inside the table

Once the table is checked, ask for the script and fence it in. Tell the AI explicitly not to fill gaps with guesses, and not to include direct quotes or names that aren't in your sources.

Write the script for a [length] explainer video for [audience] using only the claims in the verified table pasted below.
Structure: hook, context, three key points with one example each, and a closing takeaway.
Format: two columns, 'On screen' and 'Voiceover', in short scenes.
Rules: don't add facts, numbers, names or quotes that aren't in the table; don't fill gaps with assumptions; if a scene lacks information, write '[data missing]' instead of making something up; keep the language clear and neutral.

Verified table: [paste here]

Then reread every line of narration and match it to a row in the table. A line with no matching row gets cut or verified.

Pass 4: choose visuals that don't mislead

An explainer doesn't need photorealistic footage. Illustrations, animated charts, maps and diagrams lower the risk that anyone mistakes a generated scene for a real one. If you do use AI-generated images, make it clear they're illustrative.

• Don't generate 'stock footage' of specific events, places or people as if it were real recording.
• Don't use real people's faces or voices, and don't recreate statements they never made.
• Put numbers in charts only if they come from your verified table, and show the source on screen.
• Say in the video or the description that the visuals were generated with AI, and check the rules of the platform where you'll publish.
• Avoid copyrighted material and tools whose terms don't allow your use.

Pre-publish checklist

• Every fact, date, number and quote in the narration is in the table and traces to a source I've read.
• I've checked key facts in the original document, not only in the AI's summary.
• On-screen text (labels, charts, captions) matches the narration.
• No image could be mistaken for real footage of something that wasn't filmed.
• The video credits its sources on screen or in the description.
• I've disclosed the use of AI and checked the platform's rules.
• If the topic is sensitive, someone who knows the subject has looked at it first.

News and current topics change fast. Note the date you consulted each source, and consider updating the video or its description when something changes.

Templates that save you the blank page

PromptNow has templates built for this workflow: one to plan an informational video, one for an explainer based on verifiable data, and one to summarize an article or source that you provide. They all work the same way: fill in a short form and get a prompt ready to paste into your AI assistant.

No template verifies anything for you. The prompt helps you organize the work and tells the AI to stay within your sources, but the final check is yours.