The hardest part of AI video for a law firm isn’t the video

What a month of producing AI spots for a Chicago injury firm taught me about disclaimers, a federal rule almost nobody reads, and why you should never let the model render the legal text.

Every firm that asks me about AI video asks the same question first. Does it look real?

Wrong question. I spent the last month producing AI spots for a Chicago personal injury firm, and realism was never the constraint. The disclaimer was.

The rule that actually governs this

Here is the line that governs all of it, and I would guess most firms running AI creative have never read it.

Under 16 CFR Part 465, if a character in your ad describes what it was like to be your client, that is a testimonial. Testimonials carry a whole body of rules, including rules about whether the person is real. If that same character narrates a situation instead, you are outside it.

Same face. Same footage. Completely different regulatory posture.

Testify, or narrate

So every script ran through one filter. Does this character testify, or does this character narrate?

Nobody says “they settled my case.” The character says “this is what happens when a dog bites someone in Illinois.”

It sounds like a small distinction. It is the difference between an ad and a problem. You can watch the finished version of that one here — the dog bite spot — and listen for it. The character never once claims to have been a client.

Same test on the food illness spot and the rideshare spot. Every one of them narrates.

Then there is the state layer

Illinois has no ARDC opinion on AI-generated advertising. Neither do most states.

So you are not complying with a rule. You are predicting one.

We wrote a disclaimer with the firm’s founder and put it in exactly two places. Burned into the opening seconds of every video, and once in the page footer. Two, consistently — a defensible standard instead of a scatter of good intentions.

I keep talking myself into a third placement and back out of it. If you have a view on that, I want to hear it.

Never let the image model render the legal text

This is the practical one, and we learned it the hard way.

We tried it once, early. The model produced a beautiful frame with a disclaimer across the bottom reading “Ankin Ofio LC doent accept every case.”

The firm’s name, misspelled, in an advertisement, attached to a legal disclaimer. In Illinois that is arguably a Rule 7.1 problem — a false or misleading statement about a lawyer’s services — and it is unquestionably worse than having no disclaimer at all.

Now the model makes the picture and nothing else. The letters are composited afterward, by software that can spell, from a text file a human wrote. It costs nothing and takes seconds, and it means the legal language on a hundred videos is identical to the language a lawyer approved.

The risk is not where people think it is

The part I keep chewing on. The real enforcement risk here probably is not your state bar. It is the platforms.

Meta, TikTok and YouTube all have AI-labeling rules with teeth and detection, and they act in days, not years.

Your bar might look at this in 2028. Instagram will look at it Tuesday.

Whose ads these are

Ankin Law is a Chicago injury and workers’ compensation firm that has represented injured people here since 1940. Howard Ankin read every word of that disclaimer before a frame was rendered, and sent back edits.

That is not typical, and it is the only reason this project could be done carefully. The full set is at the situation index if you want to see how the standard holds across fifteen of them.

---

Spencer Forman builds AI media and marketing systems with MinuteLaunch. The work described here was done for Ankin Law, a Chicago injury and workers’ compensation firm. The videos are AI-generated and labeled as such on every page.