How to fact-check an AI-written script before you publish

Treat every number, name, date and cause-and-effect claim in an AI-written script as unverified until you have seen it in a source you trust. Pull those claims out, check each against a primary or reputable source, rewrite or cut what you cannot confirm, then read the final script aloud once more. Kadapt Creator researches with sources and runs a fact-check pass, then asks you to tick the few figures that need a human eye; it does not claim a machine can verify everything, so the last check is yours.

Creator's process is described as on kadapt.com/creator as of 20 October 2026 and can change. The checklist is general practice, not a guarantee of accuracy.

An AI-written script can be well structured, pleasant to hear and wrong in one sentence. In a video that sentence is narrated confidently, drawn, and published under your name. The fix is a short, repeatable check before anything is built.

Why it needs a human step

A language model produces plausible text. It does not know which of its sentences are true, and a made-up statistic looks identical to a real one. So the question is never “did the AI write this well?” but “what would I point to if someone challenged this line?”

A checklist for any AI script

1. Highlight every checkable claim. Numbers, percentages, dates, names, places, quotations, rankings (“the largest”, “the first”), and cause-and-effect statements (“this happens because…”).
2. Find the source for each. Prefer the primary source: the original paper, the official statistics page, the company’s own documentation. A blog quoting another blog is a lead, not a source.
3. Check the date. Figures go stale. Confirm the number is current, and say “as of” in the script if it will age.
4. Check the exact wording. “Up to 40%” and “40%” are different claims. So are “linked to” and “causes”.
5. Look for the missing qualifier. Models flatten nuance. If your source says “in one study” or “in some cases”, the script should too.
6. Cut what you cannot support. A removed line costs you nothing. A wrong one can.
7. Check names and spellings aloud. The voice will pronounce whatever is written, so an error in a name is also an error you can hear.
8. Re-read the final version. Edits introduce new errors. Read the finished script once more, start to end, before approving it.

Where the effort should go

Not every sentence deserves the same scrutiny. Spend your time on claims that are specific (figures, dates, names), consequential (health, money, legal, safety) or surprising (if it sounds too neat, it may be invented). General framing and plain explanations of well-established ideas need a lighter pass.

How Kadapt Creator splits the work

Kadapt Creator is built around this order of operations. If you ask it to research a topic, it researches with sources, outlines, writes and then fact-checks the script. You then read the script and tick the few figures that need a human eye. Nothing is built until you approve it, so a wrong figure caught here costs you a quick edit, not a re-render.

To be clear about the limit: Creator’s page says some figures cannot be checked by machine, which is why that confirmation step exists. Sourced research and a fact-check pass reduce what you have to check. They do not remove your responsibility for what you publish.

If you paste your own script instead, the claims are yours to source before you paste. That is often the better route for a topic you already know well: see turning a blog post into an explainer video.

Cost of checking early

Review happens before the paid drawing step. Creator shows the research price when you tap “Write my script” (6 credits per started minute, only if you ask it to research) and a separate quote for the video work before you approve it. Full figures are in the cost guide and on the pricing page.

Next

Once the facts hold, make sure the script is worth watching: how to write an explainer video script that holds attention, and decide on how long the video should be.

Common questions