Where AI video is risky for law firms, clinics and other regulated businesses

Regulated businesses are different from a cafe or a gym because their marketing is judged by a professional regulator as well as by the platform and the consumer-protection agency. A video that is only a little misleading in retail can be a conduct problem for a lawyer or a doctor. I am not a lawyer and this page is information, not legal or medical-regulatory advice. Rules change by country, state and profession, so use this page to ask better questions of your own regulator, not to replace it.

This guide sticks to principles on purpose. Bar and court rules, including the American Bar Association (ABA) opinions, New Jersey court guidance and individual state rules, vary by place and change. Check the primary text of any rule named below before you rely on it.

Why regulated businesses are different

Three things make this sector tricky.

• The viewer relies on the result. Someone choosing a lawyer or a clinic is making a high-stakes decision. A misleading video does more harm than a misleading burger ad.
• Several rule books apply at once. Professional conduct rules (for lawyers, state bars; for clinicians, boards and colleges), consumer-protection law, privacy law, and the platform's own ad policy.
• You carry the risk, not the tool. AI video makers say in their terms that the user is responsible for the content and for the consent of anyone shown. An agency or a vendor will rarely take on your regulatory risk. See can you use AI video in paid ads for clients? for the general rights checklist.

The rules to check, and who writes them

Rule book	Who writes it	What it usually asks
Lawyer advertising rules	Your state or national bar. The ABA Model Rules are a template that many US states adapt. Rule 7.1 is, in principle, the rule against false or misleading communications about a lawyer's services.	No false or misleading statements, including by what is left out. Many states add rules on testimonials, results and dramatisations.
Clinician advertising rules	Medical or dental boards, colleges or councils for your profession and place	Accurate claims, no guaranteed outcomes, rules on before and after images and testimonials
Consumer-protection law	In the US, the Federal Trade Commission (FTC)	No deceptive ads; clear disclosure of paid endorsements; no fake testimonials
Health privacy	In the US, HIPAA, enforced through the Department of Health and Human Services (HHS)	Written authorisation before a patient's protected health information (PHI) is used for marketing
Platform ad policies	Meta, Google, TikTok and others	Extra limits on health and legal ads and on synthetic media

Outside the US, replace each row with your own regulator. In the UK, for example, solicitors, doctors and other professions have their own regulators with advertising guidance. In the UAE and elsewhere, the health authority or bar may need to approve content. The step that never changes is this: find your regulator's current advertising rules, read the part on testimonials, results and dramatisation, and if you are unsure, call the regulator's ethics or advice line and write down what they said. Clinics should also read health claims and consent, and finance firms compliance first for insurance and mortgage ads.

The FTC items below come from the agency's own pages as of 4 November 2026. The FTC says its final rule on fake reviews and testimonials bans fake or false testimonials, including ones from people who do not exist, and allows civil penalties against knowing violators. It also says that when an endorsement is both seen and heard, the disclosure should be both visual and audible, and that disclosures hidden behind a "more" link or in comments are generally inadequate. This is a summary, so read the pages directly. The wider consumer-protection point is covered in fake testimonials and AI reviews: the legal risk, so it is not repeated here.

The one test I would use

Almost every rule above comes back to a single question. Would a reasonable viewer think this person exists, or this result really happened? If the answer is yes and it is not true, that is the problem, no matter what tool made the video, and a small caption rarely fixes it.

Use that test on every scene, every line of voice-over and every thumbnail. It covers the person on screen, the setting (is this a real office?), the result implied (a courtroom win, a healed patient) and the credentials implied (a white coat, a framed degree).

Red, amber, green by asset type

Asset	Rating	Why
Invented client or patient testimonial	Red	Fails the test. An older but useful ethics opinion, New York State Bar Opinion 661 (1994, under a former code, so historical only), called fictional client testimonials inherently false and misleading and said a disclaimer could not cure them.
Invented lawyer, doctor or nurse as presenter	Red	The viewer assumes a licensed person is speaking. A generated stand-in has no licence.
Scenes that suggest an outcome (a verdict, a recovered patient, a settlement cheque)	Red	Implied results. Most rule books want results claims to be accurate and not typical-looking unless proven.
AI-altered before and after images	Red	For clinics, altered results are the classic misleading-ad problem. AI makes them easier to produce and harder to spot.
AI avatar or voice clone of the real professional reading an approved script	Amber	Needs the professional's written consent, a clear label, a check against your regulator's rules, and a script the professional has approved word for word. Some regulators may treat it as a communication the professional must stand behind personally.
Real client or patient testimonial, with an actor re-enacting it	Amber	Historic guidance allowed dramatisation of a real, authorising client's words with clear captions. Rules today vary by state and profession, so check yours.
Generic stock-style visuals (a courthouse exterior, a stethoscope on a desk)	Amber	Fine if no one could think it is your office or your staff. Licensed stock footage with a clear licence is lower risk than generated scenes.
Captions, subtitles, translation and dubbing of a real video	Green	Does not change who is speaking or what is claimed. Have a native speaker check translated legal or medical wording.
Abstract animation or diagram explaining general law or health education	Green	No person or result is implied. Keep it accurate and say it is general information.
The real professional on camera, edited with normal tools	Green	The safest option in this sector, and often the most persuasive.

Clinics: patients, privacy and AI faces

For clinics there are two extra points.

Do not generate patients. An AI patient who "recovered" is a made-up result. A made-up patient face may also look like a real person, which raises its own problem. If you want patient voices, use real people who gave written permission for that use.

HHS says the HIPAA Privacy Rule requires a person's written authorisation before their protected health information is used for marketing, with limited exceptions (45 CFR 164.501 and 164.508(a)(3), as described on the HHS marketing page). A patient's face in a video can itself identify them as a patient. The authorisation form should name the use, and if the clip will be edited or dubbed with AI, say so. Whether your vendor needs a business associate agreement is a separate question for your privacy officer or lawyer.

For deeper background on AI in a clinic generally, see AI for clinics.

Law firms: avatars, results and confidentiality

The same ideas apply to law firms. If you are considering an AI version of a lawyer, see talking presenter video without filming yourself for the consent mechanics. The extra points for law:

• The ABA's 2024 guidance on AI (Formal Opinion 512) is reported to cover competence, confidentiality and supervision when lawyers use generative AI. Confirm it directly in the primary text. Even if you do not read it, the principle is that the lawyer stays responsible.
• State rules differ. Vendor summaries mention rules in states such as Florida, Texas and New York, but the summaries flagged their own citations as unverified, and none are confirmed here. Check your own bar.
• Testimonials need their own check. A New Jersey ethics opinion on testimonials from 2025 is reported, so check the primary text. Some states restrict testimonials or require disclaimers. Ask your bar.

For general AI use in a firm, see AI for law firms.

Safe uses: education, intros, captions, translation

This is the useful side. The following are normally the lowest-risk ways to use AI in regulated video, subject to your regulator:

• Education explainers. "What happens in a first meeting", "how a will works in general", "what to expect at a check-up". Animated diagrams, no invented people.
• Intro videos with the real professional on camera. Use AI only for captions, a clean background or a tidy edit.
• Captions and subtitles for accessibility.
• Translation of a real video into the languages your clients speak, reviewed by a fluent speaker.
• Abstract B-roll (short background clips) that does not look like your office or any person.
• Internal training videos that never go to the public. Still, never paste client facts into the tool.

Plan the disclosure into the video. A 15 second vertical clip cannot carry a 15 second text disclaimer. If the claim needs a long caveat, change the claim or use a longer format.

Confidentiality: what never goes into a generator

Do not type client facts, case details, patient names, scans, notes or recordings into any AI video or image tool. Treat the prompt box like an email to a stranger. Even if a tool says it does not train on your data, ask for that in writing. Does your AI tool train on client work? lists the questions to ask. If you use a platform that offers many models behind one login, such as Higgsfield, remember that your content may pass to several model providers, so read the terms. Higgsfield's Terms of Use (Section 4.4, as of 4 November 2026) say inputs and outputs may be used to train its models, with different terms for customers under an Enterprise Agreement, and Section 5.4 tells users not to put health information or full names into text prompts; the Higgsfield review covers what to check.

The review step: who signs off and what is logged

The reviewer must be a licensed professional or the firm's compliance officer. It must not be the agency or the intern who made the video. Use a sign-off log, kept with the asset, with these columns:

Column	What to write
Asset and version	File name, version number, where it will appear
Tools used	Each AI or editing tool, and what it did
People shown	Real or synthetic; if real, the consent form reference
Claims made	Every statement of fact, result or credential, with its proof
Viewer test	"Would a viewer think this person or result is real?" yes or no, and why
Rule check	Which regulator rules were checked, by whom, and the date
Disclosure	The wording and where it appears on screen and in audio
Reviewer sign-off	Name, licence or role, date
Retention	Where the approved file and the log are stored, and for how long

This log is the thing you will want if a client, a patient or a regulator ever asks how a video was approved.

Questions to put to an AI video vendor before paying

• Where do my prompts and uploads go, and are they used to train any model?
• Is there a data processing agreement? (For a clinic, a business associate agreement is a separate matter to ask your privacy adviser about.)
• Does the tool add invisible marks or labels to its outputs, and does it allow me to disclose AI use?
• Who owns the output, and what does the vendor's indemnity (promise to cover legal costs) say, if anything?
• Can I delete my content, and how long do you keep it?

If you are an agency serving these clients

Put it in the contract that the client approves the final cut and that the client's licensed reviewer signs off on compliance. The agency usually cannot take the regulatory risk. Say in the contract which AI tools you use. Should agencies disclose AI use? covers that conversation. Do not offer invented testimonials as a package, even if the client asks.

When to skip AI video

Skip it if your regulator has said no or you cannot get a clear answer, if the only appeal of the video is an invented person or result, if you cannot name a reviewer, or if your video would need client or patient information to be useful. A real professional on camera with plain editing beats any AI presenter in this sector.

How I'd approach this

Start with the green list only: captions, translation and plain explainer animation, with the real professional on camera. Write the red/amber/green rules into one page that your team can read in two minutes. Name the reviewer and keep the log. Before you try an amber item such as an AI avatar of yourself, ask your regulator in writing, and keep their answer. If you want help setting up the log, the approval step and a shared library of approved assets, Blur Studio builds that workflow, and you can see how it works.