How to write a script for an AI spokesperson ad

This article is general information about advertising practice and the policies we read at the time of writing. It is not legal advice, and we don’t claim any particular ad is compliant. Get advice for your own market and category.

The hard part of an AI spokesperson ad is not the face. Tools have made the presenter the easy bit. The script is where the money and the risk both sit: a weak one wastes the spend, and a careless one puts a claim in a real person’s mouth that nobody can stand behind. Here is how to write one.

Start with one job, not three

Before writing a word, finish this sentence: after watching, one person should ___. Visit a page, start a trial, book a call, remember a name.

Thirty seconds of speech is far less text than people expect — well under a hundred words at a natural pace. Room for one idea with evidence, or three with none. Pick one.

A 30-second structure that holds up

Five short beats. The timings follow what the platforms themselves say about where attention goes.

0–3 seconds: the hook. TikTok’s creative guidance tells advertisers to “introduce your content proposition in the first 3 seconds for better recall and awareness” and to “prioritize your hook in the first 6 seconds” (TikTok Ads Manager). Name the problem or the audience immediately. No logo intros, no “hi everyone”.

3–8 seconds: the stake. Why it matters, in one sentence, in the viewer’s words rather than your category’s.

8–18 seconds: what the thing is and does. One concrete mechanism, not a mood. “It drafts the quote from your job notes” beats “it transforms your workflow”.

18–25 seconds: proof you can hand to someone. A specific feature, a published guarantee, an independent standard. See the next section for what counts.

25–30 seconds: one action. One verb, one destination, no alternatives.

Then write a second version with a different hook and the same body. Hooks are the cheapest thing to test and the easiest thing to be wrong about.

Write for the sound being off

Assume the first viewing is silent, so every line that carries meaning also appears on screen — TikTok suggests “displaying 5-10 words per second when using text” (TikTok Ads Manager). Write the caption track alongside the script, and keep the hook and the offer short enough to fit a phone screen without wrapping.

Build the evidence column before the copy column

Draft in two columns: the line on the left, the evidence on the right. If the right column is empty, the line does not go in the ad.

That is close to what regulators ask for. The UK’s CAP Code requires marketers to “hold documentary evidence to prove claims that consumers are likely to regard as objective and that are capable of objective substantiation” (CAP Code 3.7). Australia’s ACCC puts it plainly: “A business must be able to prove a claim of a product having a particular quality or benefit” (ACCC).

Platforms apply their own version. Google’s misrepresentation policy says “making inaccurate claims or claims that entice the user with an improbable result (even if this result is possible) as the likely outcome a user can expect is not allowed” (Google Ads). Note the parenthesis: a result being possible is no defence if you present it as typical.

What survives: features that exist, published prices and terms, warranties, certifications, delivery times you meet. What doesn’t: “the best”, “instantly”, “guaranteed results”, untraceable percentages, and any implied typical outcome you have not measured.

The lines an AI presenter must never say

An AI spokesperson is not a customer. The moment your script slips into first-person experience, it becomes a testimonial — and a synthetic one.

Delete, every time:

  • “I use this every day.”
  • “This changed my life / my skin / my business.”
  • “I was sceptical, but…”
  • Anything in the past tense about using the product.

The US Federal Trade Commission’s endorsement guidance is direct about it: “You can’t talk about your experience with a product if you haven’t tried it” (FTC). Its Consumer Reviews and Testimonials Rule, effective 21 October 2024, targets testimonials that misrepresent coming from someone who does not exist or who lacked actual experience with the product (FTC). The CAP Code requires marketers to “hold documentary evidence that a testimonial or endorsement used in a marketing communication is genuine”, with contact details for whoever gave it (CAP Code 3.47).

An AI presenter can describe, demonstrate and recommend. It cannot remember.

If you want customer experience in the ad, get it from a real customer, on the record. Labelling the video as AI does not fix a testimonial nobody gave — the label and the claim are separate problems, and only one of them has a toggle. The same trap catches ads written in the style of user content, where the format itself implies a first-person experience before the script says a word.

Health, money and body claims: a higher bar

If your script touches health, weight, income or financial outcomes, assume ordinary marketing evidence is not enough. The FTC expects “competent and reliable scientific evidence” for health claims, usually “in the form of randomized, controlled human clinical testing”, and warns that advertisers “must have appropriate scientific evidence to back up the underlying implied claim that the product is effective and will work for buyers as it did for the endorser” (FTC).

Read that last line next to a synthetic presenter: there may be no real endorser, but there is still an implied claim to prove. In these categories the safest script is descriptive — what the product is, what is in it, and where to read the terms.

Keep the script inside the licence

If the presenter is an AI version of a real, consenting person, there is a second constraint: what that person agreed to say.

A fair licence gives that person script approval, a blocked list of categories, and per-brand approval for anything resembling a testimonial — see what a fair AI likeness licence includes. So the approval step belongs before generation, not after, and the approved version is the one you keep on file.

Three things to check against the licence first:

  • Category. Gambling, alcohol, weight loss, crypto and political content are commonly excluded.
  • Claim type. Many people will present a product but refuse to endorse one.
  • Placements and countries. A script written for one market can land in another through a broad audience setting.

Write the disclosure into the script, not around it

Realistic AI content generally needs disclosing, and the platform rules differ in wording and placement — the detail is in AI disclosure rules for ads on Meta, TikTok and YouTube. Two script-level consequences:

  1. Leave room for the label. TikTok’s AI-generated content disclaimer “will show at the bottom of the in-feed video for the duration of the ad” (TikTok Ads Manager), so keep your caption and offer out of that strip.
  2. Paid promotion is separate. If a presenter has a commercial relationship with the brand, YouTube’s paid-promotion setting “adds a disclosure label that appears at the beginning of your video” (YouTube Help) — an AI label does not cover it, or the reverse.

A skeleton you can fill in

Swap in your own words and keep the word count.

> (0–3s, on screen and spoken) Still [doing the painful thing] by hand? > (3–8s) Most [audience] lose [time or effort — only if you can show it] to it every week. > (8–18s) [Product] does [one concrete mechanism]. You [specific user action], and it [specific output]. > (18–25s) [Verifiable proof: published terms, a standard, a guarantee, a free tier]. > (25–30s) [One action]. [Destination].

Read it aloud with a timer. If you cannot finish comfortably, cut the stake beat, not the proof beat.

Before you generate: a checklist

  • One desired action, written at the top of the script
  • Hook lands in the first three seconds, and reads with the sound off
  • Every objective claim has evidence next to it, held on file
  • No first-person experience, and no implied typical result you haven’t measured
  • Health, income and body claims removed or properly substantiated
  • Script approved in writing by the person whose likeness is used, before generation
  • Category, claim type, placements and countries checked against the licence
  • Caption track written, lower strip left clear for platform disclaimers
  • AI disclosure and paid-promotion disclosure handled separately
  • Approved script, evidence and licence reference stored together

How Likeness fits in

We’re building Likeness so script approval is part of the workflow rather than an email thread: the person sets the categories they’ll appear in, approves the script before anything is generated, and the business’s finished video leaves the platform labelled as AI with a licence reference attached. It is in development and not yet open. See for businesses and trust and consent, or email [email protected].


General information only, not legal advice. Advertising rules, substantiation standards and platform policies differ by country and change often; take advice for your own markets.

Sources

Previous post
Next post