Healthcare and wellness marketing with AI: claims, consent and what to leave alone

Health and wellness is the one area where I would slow down before using AI for marketing. The tools are fine. The problem is that health marketing already carries claim rules, privacy rules and ad platform rules, and an AI tool makes it easier to break all three quickly. A fake-looking patient, a "cleaned up" testimonial or a retouched result photo can turn a normal ad into a complaint.

This page is a plain guide for owners and marketers in clinics, med spas (clinics that offer cosmetic treatments), dental practices, therapy practices and wellness brands. It uses official pages that are listed at the end. It is information, not legal or medical advice, and where a rule is uncertain it says to check. Talk to your own lawyer or compliance lead before you publish anything about treatments or results.

Why health marketing is different

Three layers of rules sit on top of each other.

• Claims. What you say a product or treatment does must be backed by evidence. The FTC enforces this for advertising in general.
• Patient privacy and consent. If a patient appears in your marketing, or their story is used, you need the right kind of permission. HIPAA is the main US rule for patient health information. For the risks of invented patients and results, read where AI video is risky for clinics and regulated businesses.
• Platform and product rules. Google Ads limits many health topics. For drug advertising, the US Food and Drug Administration (FDA) is a separate regulator with its own rules.

AI adds a fourth problem: it makes things that look real. A generated patient, a generated doctor or a generated result can look like a real person with a real outcome. That is exactly the kind of thing these rules were written to stop.

The US Department of Health and Human Services (HHS) has published no AI-specific guidance on this. HHS runs the office that enforces HIPAA. What exists is general guidance that predates AI video, so I treat it as the floor and apply it to AI content the same way I would apply it to a photo or a filmed video.

Other AI uses in healthcare marketing, and the data rule

Most pages that rank for this topic are about something other than video. Vendor and trade pages list chatbots for patient questions, personalized content, predictive targeting and campaign automation as the main uses. They are mostly marketing pieces and give few specifics on rules. One of them mentions partnering with vendors that sign a business associate agreement, and another says AI-drafted scripts need human review for regulatory problems. I agree with both points. The rule of thumb is that every one of these uses runs on patient data, so each one brings back the HIPAA question: do not feed patient information into a tool until counsel has confirmed the data terms. Creative uses, which this page covers, carry the extra risk that a generated image or voice looks like a real person.

Claims: what you must be able to prove

The FTC's health products guide says advertisers need competent and reliable scientific evidence for health claims. It also says two things that surprise owners. A sincere testimonial is not enough to support a claim. And a line such as "results not typical" does not fix a testimonial whose results really are atypical. If the result you show is not what most people get, a small disclaimer does not rescue the ad.

Before you write a script, sort each claim into a ladder. I use three levels.

Level	Type of claim	Examples	Before you make the video
Green	General education	What a procedure involves, how to prepare, what a visit looks like, how to book	A clinician checks the facts
Amber	Service descriptions	What you offer, who it is for, how long it takes, what the equipment is	Clinician and compliance check the wording; no outcome promised
Red	Outcome, cure, weight-loss or anti-ageing claims	"Lose 10 pounds", "reverses ageing", "cures", "treats" a condition	The evidence file exists first, before any script is written

The rule for red claims is simple: no evidence file, no script. If you cannot show the study or the data behind the sentence, the sentence does not go into a video, an image caption or an AI voice-over. AI makes it cheap to produce ten versions of a claim. It does not make the claim any easier to prove.

Wellness, supplements and weight loss are the riskiest group for ad rejection whether or not AI was used. Budget extra review time for them.

Consent: patient stories, photos and voices

The HHS page on HIPAA marketing says the Privacy Rule requires written authorization before protected health information (PHI, meaning health details that identify a person) is used for marketing, with limited exceptions. The rule is cited as 45 CFR 164.508(a)(3). That page was written around 2002 and 2003. It does not talk about photos or AI, so it does not tell you what to do with a face or a voice. What it does establish is the principle: marketing use needs its own written permission.

The key point many clinics miss is that the consent a patient signs for treatment, or for clinical photos in their chart, is a different thing from marketing permission. Do not assume one covers the other.

A per-patient marketing consent file

Keep one file for every real person who appears in your marketing. It should travel with the asset, so anyone who finds the video or image can find the permission.

• Marketing authorization, separate from treatment consent. A signed form that says the person agrees to appear in marketing.
• Scope. Which uses are allowed: video, still image, audio, the face, the voice, the story, and which channels.
• AI use. A line that says whether AI tools may be used on their image or voice, for example to caption, translate or edit. If it is not in the form, treat it as not allowed.
• Duration. How long you may use the material, and what happens at the end.
• Revocation route. A clear way for the person to withdraw, and a promise of what you will do when they do.
• Where the asset lives. The file name and location, so you can pull it down fast.

A tool for collecting a signature online can help you keep these forms tidy, but the content of the form is what matters. The consent checklist for real faces and voices goes through the wording in more detail.

Do not "clean up" a testimonial with AI

If a testimonial cannot be used as it was recorded, do not repair it with an AI voice or lip sync (software that changes mouth movement to match new audio). Changing the words changes what the patient said, and the person did not consent to the new version. If the audio is poor, record it again with their agreement.

Before and after images: leave them alone

My rule is firm: never generate, retouch, upscale (add detail to make an image larger or sharper) or "enhance" a before-and-after image with AI. A result photo is a claim about what happened to a real person. Any change, even a small clean-up of lighting or skin, can turn a true result into a misleading one. Gyms and studios face the same trap, covered in AI class promos for fitness studios and gyms. The FTC's point that atypical results are not cured by a disclaimer applies directly: the image is the claim.

Some practical habits follow from that.

• Take before and after photos under the same conditions: same light, angle, distance and camera settings.
• Do not use filters, smoothing or AI tools on either image. Crop only if the crop does not hide anything.
• Show typical results, or say clearly what typical results are, with evidence. A "results not typical" label does not cure an unusual result.
• Get marketing authorization for each person, as above.

One vendor blog on AI video for med spas takes the same position: never AI-touch before and after images. I treat that as a vendor opinion, not as law. Some state medical boards also have rules about result photos. Check your own state board for its rules.

Synthetic patients and AI doctors

This is the area with the least official guidance. Check HHS for any guidance on generated patients, and until then do not assume a synthetic patient is outside or inside HIPAA. Do not rely on anyone, including a vendor, who tells you it is clearly allowed.

What I would do, from a risk point of view:

• Do not present a generated person as a real patient. A testimonial implies a real customer with a real result. The FTC has a rule about fake reviews and testimonials, which is covered on the fake testimonials page. Check the rule text with counsel.
• If you need an illustration, make it obviously an illustration. A drawn or clearly stylized figure, labelled as such, is far safer than a photoreal "patient" who looks like a person with a result.
• Do not use an AI doctor as your spokesperson. Patients assume a person in a white coat is a licensed clinician. A real clinician on camera, speaking their own words, is better for trust and for your state board. State boards may have rules about how clinicians are presented, so ask yours.
• Prefer a consenting real patient or a licensed stock photo. A licensed model photo bought from a stock library, used as an illustration, is simpler than a generated one.

Safe uses of AI in healthcare marketing

Lower-risk uses are the ones where AI helps with production, and the facts and the people stay real.

• Education explainers. Plain-language videos on how a procedure works or how to prepare, using green-level claims, checked by a clinician.
• Captions and subtitles. They make a real clinician's video easier to follow without changing what is said.
• Translation of real, approved content. Have a native speaker check the result. Do not alter the recorded face to match the new language without a separate consent line.
• Abstract B-roll. Background footage that shows no person, no result and no claim: a clean room, a pattern, an object.
• Drafting and editing help. First drafts of scripts, which a clinician then rewrites and signs off.

Keep real-patient assets separate from AI B-roll in your library, so a reviewer can see at a glance which is which. A shared asset library is a simple way to do that.

A word about the tools. Ask every AI vendor where your uploads are stored and whether they are used to train models. Higgsfield, an AI platform for making images and video, says in its Terms of Use that you must hold consent for any person's likeness or voice you upload, and that you must not put sensitive personal information, health information included, in text prompts. Its Terms also bar using its developer access to process protected health information without a written addendum, and say it may use content to train and improve models except under a separate enterprise agreement. I would not upload patient media to any general AI tool unless counsel has confirmed the data terms. Whether a business associate agreement (a contract HIPAA can require when a vendor handles patient information for you) is needed is a question for counsel, not a point I can settle here. For more questions to ask, see does your AI tool train on client work.

Ad platform and regulator rules to check

These are the official pages behind this guide. None of them talks about AI-generated content specifically, so read them as the general rules that AI content must also meet.

• Google Ads healthcare and medicines policy. Google restricts prescription drug services, unapproved substances and speculative treatments. Some categories need certification from Google first, for example addiction services, health insurance in the US and pharmaceutical manufacturers. The policy page has no section on before-and-after images that I could see, so read the live page and the policies for your own category.
• FDA drug advertising. On 9 September 2025 the FDA announced a crackdown on deceptive drug advertising. It said it was sending thousands of letters and about 100 cease-and-desist letters, and that it was looking at social media and influencers. It also said it uses AI to monitor ads. The announcement did not mention AI-generated content. Drug ads generally need balanced risk information, and I would ask regulatory counsel how that applies to short video before anything is made.
• FTC. The health products guide and the endorsement guide FAQ cover claims, testimonials and disclosure of paid or free endorsements.
• Meta health and wellness policy. Meta's own page is not cited here. A vendor blog says Meta changed its health ad policy in July 2026, but that is unconfirmed, so read Meta's own policy pages before you run health ads there.
• California. Some law-firm blogs mention a California bill about disclosing AI-altered images of health care providers. Check the bill text before you rely on it. Ask your lawyer whether any state law on this applies to you.
• State boards. Medical, dental and other boards may have their own advertising rules, including for result photos. Check yours.

For platform labelling of AI content in general, see do you have to label AI content, and for policy traps in ads see AI ad platform policy traps.

The review step and consent file

Put a short gate between "made" and "published". It should be a named person, not a vague "the team".

1. Claim log. List every claim in the asset, its ladder level and where the evidence is. Red claims with no evidence stop here.
2. Clinician sign-off. A licensed clinician in your practice confirms the facts are right and nothing implies a guaranteed result.
3. Consent check. Every real person in the asset has a marketing authorization on file that covers this use, including any AI editing.
4. Reality check. Nothing in the asset is a generated person presented as real, and no result image has been touched.
5. Platform check. The ad fits the current policy of each platform and any category certification.
6. Record. Save the final file, the claim log and the consent reference together, with the date and the reviewer's name.

The same gate works for outside agencies. A simple approval workflow keeps the sign-offs in one place.

When to skip AI video

• When the main point of the asset is a patient result. Use the real photo, with consent and no edits.
• When you cannot name the person who signs off clinical claims.
• When the ad is for a drug, a device or a treatment that has regulated advertising rules and you have no regulatory adviser.
• When the plan relies on a generated patient or doctor to look credible. That is the exact situation the claim and testimonial rules exist for.
• When your state board has not been checked.

A real clinician filmed on a phone, with captions added, beats a polished AI avatar for trust, and it is safer. For general use of AI inside a clinic, not marketing, see AI for clinics.

How I'd approach this

I would start with the claims, not the tool. Write down what you want to say, put each line on the ladder, and drop or evidence the red ones. Then build the consent file template, because that is the part clinics most often skip. Only then pick the small set of jobs where AI helps: captions, translation of approved scripts, education explainers, abstract B-roll.

I would keep every result photo untouched, use real clinicians on camera, and run each asset through the review gate above with a named signer. When a rule is unclear, which in this area it often is, do not guess. Check the regulator's published guidance, your state board, or your lawyer.

If you want the gate, the consent file and a shared library set up as a repeatable process for your practice, Blur Studio builds that kind of workflow. See how I work with small businesses. This page is information and not legal or medical advice.