Pharma teams can use AI-generated video and digital avatars in promotion, and the content is judged by exactly the same standards as any other video. Claims must match the approved labeling, risk information must be as prominent as the benefits, and the finished piece goes through MLR review and, where it applies, FDA submission. What AI adds is a second set of questions about the people on screen: could a viewer mistake a synthetic presenter for a real doctor or patient, was a real person's face or voice used with consent, and where does the law require a label?
This article covers where AI video helps pharma teams, how FDA's promotional rules apply to it, the FTC, state and EU rules on synthetic people and digital replicas, and how to keep claims and ISI accurate through MLR review. As of October 2026 FDA has not issued guidance specific to AI in drug promotion, so most of this is existing rules applied to a new way of making content. It is not legal advice.
Where does AI video help pharma marketing most?
In the formats that were always worth making and rarely got made. Mechanism of action explainers, disease education for patients, short HCP modules on dosing or patient selection, and rep-ready clips that answer one question in under a minute all use the same approved claims in a different shape. Traditional production made each one a shoot, an edit and a review cycle. AI video turns an approved script into a draft cut in hours.
Localization and variation are the other big gains. The same explainer can be rendered for different audiences, lengths and channels, and voiced in another language without a reshoot.
As with generative AI in pharma marketing generally, the gains come from adapting approved content, not from inventing new claims. A video that only says what the claims library already says is fast to make and fast to review. A video whose script a model improvised is neither.
Do FDA's rules change when a video is AI-generated?
No. FDA's advertising regulation, 21 CFR 202.1, and the misbranding provisions behind it apply to promotional content whatever made it. An AI-rendered explainer has to be consistent with the label, support every efficacy claim with substantial evidence as worded, state the indication accurately and present risk information with prominence comparable to the benefits. Promotional pieces are submitted to FDA on Form FDA 2253 at first use.
Video has its own fair balance traps, and AI makes them easier to fall into. FDA's final rule on the major statement in TV and radio ads, with compliance required from November 2024, says the risk statement must be clear, conspicuous and neutral: consumer-friendly language, audio and text together on TV, and no sounds or visuals that compete with it. Strictly, the rule covers broadcast DTC ads, but it is the right bar for any video. A generated presenter smiling through the risks over upbeat music fails it as surely as a filmed one.
Visuals count as claims. A synthetic patient hiking, sleeping well or playing with grandchildren implies an outcome, and FDA reads imagery as part of what an asset says. Generative models fill scenes with plausible detail nobody asked for, so every generated shot needs the same scrutiny as the script. Does it suggest a benefit the trial did not measure, or a patient outside the indication?
Can an AI avatar play a doctor or a patient?
It can, as long as nobody is misled about who they are. The FTC's Endorsement Guides, revised in 2023, describe an endorser as someone who is or appears to be an individual, group or institution, which brings virtual and fabricated presenters into scope. Ads that present people as actual consumers must use actual consumers or clearly disclose that they are not. An endorser presented as an expert must have the expertise the ad implies: the Guides' own example is an endorser called "Doctor" who is not a medical doctor with experience in the field.
The FTC's rule on consumer reviews and testimonials, in force since October 2024, goes further. It prohibits testimonials that misrepresent that they come from someone who does not exist, and the FTC named AI-generated fake reviews when it announced the rule. A synthetic patient describing their own experience with a drug is the clearest case to avoid.
Pharma has its own convention too. PhRMA's Guiding Principles on direct-to-consumer advertising, which member companies commit to, say that ads using actors to portray healthcare professionals should say so, and that a real professional's paid appearance should be acknowledged. The same logic covers avatars. If a synthetic presenter wears a white coat, speaks with clinical authority or could be read as a physician, show a clear on-screen statement that it is an AI-generated presenter, not a real doctor. Better still, use a presenter who does not imitate a clinician at all.
FDA has primary authority over prescription drug advertising, and these FTC and industry standards are not FDA rules. But a viewer misled about who is speaking is a misleading ad under any of them, which makes them the practical reference.
What consent do you need to use a real person's face or voice?
Written, specific consent, and more than a standard talent release. If you build a digital replica of a real physician, patient ambassador or actor, the agreement should say what the replica may say, in which assets, channels and markets, for how long, and how the person can withdraw. A replica can be made to say anything, so consent should cover the scripts, not only the likeness. A real physician shown endorsing a product also brings the FTC's expectation that a paid expert's connection is disclosed.
State law is moving. California's AB 2602, in effect since 1 January 2025, makes a contract term allowing a digital replica of a performer's voice or likeness unenforceable if it lacks a reasonably specific description of the intended uses and the person was not represented by a lawyer or union when it was negotiated. Tennessee's ELVIS Act, in effect since July 2024, protects a person's voice alongside their name and likeness against unauthorized use, including AI imitation.
New York goes furthest on advertising. Since 9 June 2026, an ad that includes a synthetic performer, a digitally created person who appears to perform but is not recognizable as any real individual, must conspicuously disclose it when the advertiser knows. Penalties are $1,000 for a first violation and $5,000 for each one after. AI used only to translate a real performer's speech is excluded, as are audio-only ads. For a national campaign, the simplest approach is to treat that disclosure as the default everywhere.
Does the EU AI Act require labels on AI-generated video?
Often, yes. The transparency obligations in Article 50 of the AI Act have applied since 2 August 2026. Deployers who use AI to generate or manipulate a deep fake, image, audio or video that resembles real people, places or events and could falsely appear authentic, must disclose that it is artificially generated or manipulated. Providers of generative tools must also mark their output in a machine-readable way, with a grace period to 2 December 2026 for systems already on the market.
An entirely fictional avatar may fall outside the deep fake definition, but the machine-readable marking still applies, and a visible label is the safer choice for any presenter who looks human. The 2026 Digital Omnibus delayed the Act's high-risk obligations but left most of Article 50 on its original timetable.
Medicines rules sit on top. The EU does not allow prescription medicines to be advertised to the general public, and national codes govern how HCP promotion is presented. For an AI-generated HCP video in Europe, plan for a visible disclosure, check the code for each market, and keep the provider's marking intact when you edit.
How do you keep claims and ISI accurate through MLR review?
Lock the script before you render anything. Every line of voiceover and every on-screen super should come from approved claims with their references attached, and the safety information should be the approved text, not a summary a model wrote. Generate from that script. Never ask a model to write and render in one step.
Then check what only video can get wrong. Does the synthetic voice pronounce the drug name, dose and indication correctly? Do the risks get comparable time, readable text and a calm delivery? Do the supers and voiceover say the same thing? Does any generated shot imply a benefit or a population the label does not support? Is the AI or actor disclosure on screen?
Review the final render, not the script and storyboard separately. Reviewers should see the video at the size and on the device it will play on, with comments tied to timestamps, the reference for each claim, the consent record for any real person's likeness or voice, and the disclosure decisions. The MLR review process is the same as for any video. The evidence the reviewer needs is wider.
Treat every change as a new version. Regenerating a scene, even from the same script, can produce different visuals, so it needs approval again. Each language version and each cut-down is its own asset: a translated voiceover must be checked against that market's label, and a 15-second cut of an approved 60-second video can lose the risk information entirely if nobody looks.
Where SwishX fits
Magic Studio's Video studio makes explainers, avatars and reels in a brand's own characters and voices, generated from its Brand Dossier and claims library rather than from an open prompt. Every claim is cited to its source and checked against the label for its market, fair balance and safety information are placed in the video as it is made, and anyone can ask the asset where a line came from. Finished work goes into the MLR queue and on into Veeva PromoMats with every claim and its reference attached.
Decisions about who appears on screen, what consent covers and where a disclosure goes stay with your legal and regulatory teams. What changes is that the claims and ISI in the video are already the approved ones when it reaches them. See how it works in Magic Studio.
Frequently asked questions
Can pharma companies use AI avatars in drug ads?
Yes. There is no US rule against it, but the video must meet the same FDA standards as any promotion, and a synthetic presenter should not lead viewers to think they are a real doctor or patient. Disclosure laws in New York and the EU may also apply.
Do you have to disclose AI-generated people in pharma ads?
It depends on the market. New York requires a conspicuous disclosure in ads that include a synthetic performer, the EU AI Act requires deep fakes to be disclosed, and the FTC treats an undisclosed fake expert or fake patient testimonial as deceptive. Disclosing by default is the simplest approach.
Can an AI avatar give a patient testimonial for a drug?
It should not present itself as a real patient describing a real experience. The FTC's rule on consumer reviews and testimonials prohibits testimonials that misrepresent that they come from a person who does not exist, and its Endorsement Guides require ads showing actual consumers to use actual consumers or say otherwise.
Does FDA have guidance on AI-generated promotional content?
As of October 2026 FDA has not issued guidance specific to AI in drug promotion. Its existing rules apply to AI-generated content in full, including fair balance, consistency with the label and Form FDA 2253 submission at first use.
How should MLR review an AI-generated video?
On the final render, like any video, plus the extra evidence AI needs: a reference for every claim, the consent record for any real person's likeness or voice, the disclosure decisions, and a fresh approval whenever a scene is regenerated.