The AI Rewrote Your Script. Legal Approved the Old One. For most content, an AI rewrite is a favor. For content that has already been reviewed, signed, and dated, it is the most expensive thing an AI tool can do.
Nnibu Thomas
Here is a claim that sounds backward in 2026.
For a whole class of content, the most valuable thing an AI video tool can do with your script is nothing. Not polish it or tighten. But leave every word exactly where a human put it.
Not for everything. If you are a founder recording a rough demo at midnight, let the AI clean up your rambling. It will do a better job than you will. If you are drafting a marketing explainer, let it play. But if the words in front of you have already been through clinical, legal, and regulatory review, and someone has signed and dated them, then the AI's improvement is not an improvement. It is a change. And in a regulated company, a change is never free.
Dana's hard limit
Dana writes training content for a company that makes smart infusion pumps.
Her current project is a set of short videos for the software hospital pharmacists use to build the pump's drug library: the medications, concentrations, and dose limits the pump will enforce at the bedside. Every script she writes goes through clinical, regulatory, and legal review before anyone records a word. The last batch took three weeks to clear. Nobody complained. That is what review is for.
This quarter, her team adopted an AI video tool to speed up production. Record the screen, paste the script, let the AI handle the voice and the polish. The first video came back in an afternoon. It looked great. Honestly, it sounded better than the approved script. Smoother. More conversational.
Then Dana listened to it properly.
Her script said: Set the hard limit for heparin. The narrator said: Set the maximum dose for heparin.
To most people, those are the same sentence. To a pharmacist, they are not.
In a smart pump, a hard limit is a ceiling the nurse cannot override at the bedside. A soft limit can be overridden with a confirmation. The approved script used the exact term the software uses, the term the pharmacist will see on screen, the term three reviewers checked. The AI swapped it for something friendlier. Something that quietly erases the one distinction the whole step exists to teach.
Nobody wrote anything false. The AI was trying to help, and that is exactly the problem.
The rewrite is not a feature. It is a new review cycle.
Here is what Dana's afternoon of saved production actually bought her.
She now has to compare every line the narrator speaks against the approved script, for every video in the batch, because she has no way of knowing which sentences were improved without listening to all of them. Or she sends the reworded scripts back through review, which is the honest option and the expensive one.
How expensive? The average medical, legal, and regulatory approval cycle runs about 21 days by industry benchmarks, and reviewers describe pieces that bounce back ten times or more (PharmaText, 2026). Two or three rounds is normal planning, and more is common for sensitive material (Revisto, 2026).
Review cycles run for weeks, and one of the named culprits is "round trips triggered by one change." (CruxDigits, What Is MLR Review, 2026)
Read that last phrase again. That’s one change.
An AI that rewrites twelve sentences in a five-minute script has not made one change. It has made twelve. Each one is a question a reviewer now has to answer. The hours the AI saved in production come back, with interest, in review.
This is the arithmetic the demo never shows you. Production time goes down. Total time to approved, published content goes up.
Can you make an AI video from a script without the AI rewriting it?
Yes, but you have to check, because many AI video tools rewrite by default. A large share of them were built for people who do not have a script: they transcribe rough narration and turn it into polished copy, which is the opposite of what a team with approved wording needs.
The test takes ten minutes. Paste a reviewed script into the tool, export the video, and compare the narration to the source, word for word. If even one word moved, ask whether you can stop it from moving. If you cannot, the tool was built for someone else.
Small words, large consequences
The changes that matter in regulated content are rarely dramatic. They are small, reasonable, and wrong. Here is what polish tends to look like up close. The lines are illustrative, but anyone who has sat on a review committee will recognize the pattern.
Approved line | The polished version | What quietly changed |
Verify the infusion rate against the physician's order before selecting Start. | Double-check your rate, then hit Start. | The check against the order is gone. That check was the safety step. |
Set the hard limit for heparin. | Set the maximum dose for heparin. | A hard limit cannot be overridden at the bedside. 'Maximum dose' erases the distinction the step exists to teach. |
This device must not be used with a pressure line. | This device shouldn't be used with a pressure line. | 'Must not' reads as a prohibition. 'Shouldn't' reads as advice. |
Select Confirm. | Click the confirmation button. | The narration no longer matches the label the user sees on screen. |
See section 4 of the Instructions for Use. | Check the user guide for more details. | The specific, reviewed reference is gone. |
Notice that none of the rewrites are bad English. Most of them are better English. That is exactly why they slip through. A reviewer would catch a hallucinated dosage in seconds. A friendlier synonym for a defined term is much harder to see, and much more likely to ship.
Where the rules actually bite
Not every training video lives inside a quality system. But plenty do, especially content that teaches customers how to configure or use a device. Since February 2, 2026, the FDA's Quality Management System Regulation has incorporated ISO 13485:2016 by reference (The FDA Group, 2026). Under that framework, documents are reviewed and approved before they are issued, changes go back through the same control, and the FDA reads approved as meaning a signature and a date (NAMSA, 2026).
So think about what that signature means. A named person read these exact words and accepted responsibility for them. A rewrite after the signature is not editing. It is publishing words nobody signed.
That is the line legal cares about. Not whether the new version reads better, but whether anyone approved it.
The objection, and why it misses the point
The obvious pushback is that the AI version is better. Clearer, friendlier, easier to follow. Sometimes that is true.
But in approved content, better is not the standard. Approved is. If a script genuinely needs to be clearer, the place to fix it is upstream, in drafting and review, where a human makes the change and a human signs it. Not downstream, inside a video tool, where the change happens silently while the signature still sits on the old words.
And notice what the rewrite assumes: that the script is a rough draft. For a founder's demo, it often is. For Dana, the script is the most finished part of the entire project. It is the one artifact three departments already agreed on. Treating it as raw material gets the whole thing backward.
What AI should do instead
None of this is an argument against AI in regulated content. It is an argument about where AI belongs.
The words are the part that is already done. Everything around the words is where the work still is: a clear voice, narration timed to the right screen, captions that match, pacing that does not rush a pharmacist through a dose limit. That is production, and production is exactly what AI is good at.
Let AI do the production. Leave the words to the people who signed them.
That is how Zenious treats an approved script. Your script is used exactly as written. Every word you submit is every word the narrator says, and the captions match it, while Zenious handles the voice and syncs the narration to the right moments on screen. If you want to run the test the way Dana would, paste your approved script into Zenious, export the video, and check every word.
So what this means
For two years, the question everyone has asked about AI and content is how much of it the AI can write.
For approved content, that is the wrong question.
The right question is how much the AI is willing to leave alone.
In most of the content world, a rewrite is a gift. In the world Dana works in, it is a liability with a nice voice. The tools that win there will not be the ones that write the best sentences. They will be the ones that never touch a sentence a human already signed.
Dana's script said hard limit. The video should say exactly that - hard limit.
That is not a feature request, it is the whole job.
Sources
PharmaText. (2026). What Is MLR Review? Medical, Legal and Regulatory Review Explained. pharmatext.ai/blog/what-is-mlr-review
Revisto. (2026). MLR Review Process: Stages, Roles, and Key Bottlenecks. revisto.com/news/mlr-review-process-stages-roles-and-key-bottlenecks
CruxDigits. (2026). What Is MLR Review? Process and Where AI Helps. cruxdigits.nl/blog/what-is-mlr-review
Improvado. (2026). MLR Review Process Guide: Steps, Best Practices and Tools. improvado.io/blog/mlr-review-process
The FDA Group. (2026). The FDA's Quality Management System Regulation (QMSR): An Introduction and Compliance Guide. thefdagroup.com/blog/qmsr-quality-management-system-regulation
NAMSA. (2026). Industry Update: Quality Management System Regulation (QMSR). namsa.com
Zenious. (2026). zenious.ai
