AI Voices in Ads: When a Synthetic Voice Needs Consent, a Label, or Neither

Quick answer: a synthetic voice needs consentwhen it is attributable to a real, identifiable person, whether or not you sampled that person's recordings. Tennessee's ELVIS Act writes that test into law and makes unauthorised use in advertising a civil action. It needs a labelwhen a platform rule says so, and the platforms do not agree. TikTok's Community Guidelines exempt generic text-to-speech narration, TikTok's Ads Manager policy does not, YouTube exempts a person cloning their own voice, and TikTok Shop bans AI voices from LIVE outright. It needs neither when the voice sounds like nobody in particular, runs on a surface without an audio-label rule, and says nothing a real customer could be mistaken for saying. The rest of this post is the three questions that sort any ad into one of those buckets.

Question one: is the voice attributable to a real person?

This is the consent question, and the most precise wording of it comes from Tennessee. The Ensuring Likeness, Voice, and Image Security Act of 2024, enrolled as Public Chapter 588, adds a definition to Tennessee Code 47-25-1102: "'Voice' means a sound in a medium that is readily identifiable and attributable to a particular individual, regardless of whether the sound contains the actual voice or a simulation of the voice of the individual". It then rewrites 47-25-1103(a) so that "Every individual has a property right in the use of that individual's name, photograph, voice, or likeness in any medium in any manner."

Three things follow. The trigger is not "AI" and not "cloned"; it is readily identifiable and attributable. A preset that sounds like a generic adult is outside the definition. A synthesis a listener would pin on a named person is inside it, however it was built. The statute protects "Every individual", not only celebrities. And it reaches advertising by name: the new 47-25-1105(a)(1) makes a person liable in a civil action who knowingly uses an individual's voice "for purposes of advertising products, merchandise, goods, or services" without prior consent, and (a)(2) covers anyone who "publishes, performs, distributes, transmits, or otherwise makes available to the public an individual's voice or likeness, with knowledge that use of the voice or likeness was not authorized by the individual or, in the case of a minor, the minor's parent or legal guardian, or in the case of a deceased individual, the executor or administrator, heirs, or devisees of such deceased individual". The Act took effect on 1 July 2024.

It is one state's law, but it is the clearest statutory statement of the right of publicity that makes voice an explicit property right. Note what it is not: a disclosure law. New York's GBL 396-b is the disclosure law, and it turns on a synthetic performer who is not recognisable as any real person, so the two statutes bite on opposite ends of the same spectrum. The New York post covers that one.

The federal layer asks a different question again. The FTC's Consumer Reviews and Testimonials Rule defines a consumer testimonial in 16 CFR 465.1(f) as "an advertising or promotional message (including verbal statements, demonstrations, or depictions of the name, signature, likeness, or other identifying personal characteristics of an individual) that consumers are likely to believe reflects the opinions, beliefs, or experiences of a consumer who has purchased, used, or otherwise had experience with a product, service, or business." Under 465.2(a) it is unfair or deceptive to create such a testimonial that materially misrepresents "That the reviewer or testimonialist exists", that they used the product, or their experience with it. The FTC's own Q&A then says, of AI stock avatars: "The rule has no blanket prohibition on the use of AI-generated avatars in marketing." And, asked whether actors may portray testimonialists in a television ad: "Such conduct is not covered by the rule unless the underlying testimonials were fake or false. But using actors to portray testimonialists could be deceptive in violation of the FTC Act."

So federally a generic voice is never the problem; a recognisable one can be. The same Q&A answer adds: "It would also violate the rule for someone to use a celebrity avatar without the celebrity's permission to speak favorably about a product, if reasonable consumers would think that the celebrity actually gave a testimonial for the product." That is 465.1(b)'s celebrity testimonial, and a cloned celebrity voice is the identifying personal characteristic it names. A synthetic voice reading "I used this for a month and my skin cleared" is a testimonial from a testimonialist who does not exist. The same voice reading "here are the three ingredients and what each one does" is a presenter script outside 465.2. The words decide it, not the synthesiser. The full reading of that rule against an AI actor's script is in the FTC fake testimonial post.

Question two: which surface, and what does its label rule say about audio?

Here the answers diverge, inside TikTok as much as between platforms. Every cell below is a sentence read on a page in the Sources list.

The row that catches people is the second one. TikTok's Misleading and false contentad policy states the rule as "Significantly edited media and AIGC content is allowed if the following requirements are met: Apply the AIGC label, or by adding a clear disclaimer, caption, watermark, or sticker of your own." Its minor-tweaks list is lighting, backgrounds and denoising. The generic-TTS carve-out in the Community Guidelinesis not on the ads page. A synthetic voiceover over real product footage is therefore label-free as an organic post and label-required as an Ads Manager upload, where "If we identify AI-generated content that has not been disclosed, your ad will be rejected or restricted." The toggle mechanics are in the TikTok disclosure post. The Spark Ads row behaves differently again, because the label belongs to the organic post, which the Spark Ads post covers; the Shop rows, including LIVE, are in TikTok Shop's AI content rules.

Vendor blogs claim TikTok Ads Manager requires uploaded consent documentation for a cloned voice. We could not find that on the advertising policy page, the disclaimer help page, or either Seller Center AIGC page as read today. What exists is a contractual version: the TikTok One advertiser terms(last updated May 2026) define Advertiser Assets to include "any actual or simulated likeness, voice, photograph, personal characteristic, name, performance, compositions and master recordings of music", and then place the burden on you: "You shall be solely responsible for securing all applicable licenses, clearances, and permissions required for TikTok and any applicable Creative Partner to display, distribute, exhibit, and otherwise use the Advertiser Assets and any third party rights of any kind or nature, including, but not limited to, any rights under copyright, trademark, or any rights of privacy or publicity included therein ('Third-Party Intellectual Property Rights') in and in connection with the TikTok One Creative Content and the applicable Project in accordance with these Terms and as otherwise set forth in an applicable Creative Brief or agreed to by you and TikTok within the TikTok One Platform." "Simulated" voice is named explicitly. Those terms only bind advertisers who commission creative through TikTok One; a self-served Ads Manager upload is governed by the advertising policy above, which asks for a label, not paperwork.

The UK Seller Center page adds: "Don't use AI-generated avatars, voices, personas, presenters or hosts that impersonate, portray or falsely imply medical or professional expertise to promote products or medical claims." That catches a synthetic voice introduced as "as a dermatologist, I recommend" even when it resembles no real dermatologist. Consent is irrelevant there; the claim of expertise is the violation.

Two surfaces are left out of the table on purpose. Google's ads policy offers an optional AI label and no audio-specific rule, covered in the Google Ads post. And Article 3(60) of the EU AI Act defines a deep fake as "AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful", so audio sits inside the Article 50(4) disclosure duty that the EU AI Act post works through.

Question three: where did the voice come from?

The first two questions turn on this one, and it is answered by the pipeline, not the policy. We can describe ours exactly because it is in our model registry. A voice gets into a render on UGC Vids AI by one of three routes.

Native joint audio. Veo 3.1, Kling 3.0, Seedance 2.0 and 1.5 Pro, and Happy Horse 1.1 generate speech and picture in one pass from the prompt. There is no voice input. The voice is whatever the model composes, so it cannot be a clone of anyone and the consent question does not arise. The label question still does, because the audio is completely AI-generated.

Built-in text-to-speech.Pruna Avatar takes a still image and a script and speaks it with its own TTS. Our registry passes one of two named presets by the actor's gender, Puck or Zephyr, in one of nine languages; Replicate's model page describes 30 named voices. Presets by design, no cloning path.

Audio-driven.OmniHuman 1.5 and VEED Fabric 1.0 take a still image plus a finished voice track and animate the face to that waveform. Our pipeline generates the track first with MiniMax speech-02-hd on Replicate, using stock preset IDs (English_Trustworth_Man or English_Graceful_Lady by default), and the same code path accepts a cloned voice ID in place of the preset. That substitution is the single point in our stack where a real person's voice can enter an ad, and the only place question one has teeth. If you supply a clone, the ELVIS definition is describing your file, and if the ad runs through TikTok One, so is that warranty. The lip-sync comparison explains why you would want that path anyway: it is the only one where the spoken words match your script exactly.

Cost follows the route. The two audio-driven models and Pruna are billed per spoken second, because output length equals the length of the speech. From our credit table: Pruna Avatar bills 18.4 credits per spoken second at 1080p; OmniHuman 1.5 and VEED Fabric 1.0 inherit their rate from a 15-second anchor, which works out to roughly 32.7 and 30.7 credits per spoken second. The estimator assumes about 2.3 spoken words per second and rounds up, so a 60-word script bills as 27 seconds. During the free 3-day trial OmniHuman 1.5 and VEED Fabric 1.0 are among the blocked models, and the 250-credit per-render trial cap means a Pruna Avatar trial render tops out at about 13 spoken seconds.

The three questions, applied

• Sounds like nobody, presenter script, Ads Manager upload. No consent needed. Label it, because the ads policy lists completely AI-generated audio and has no TTS exemption. FTC: not a testimonial.
• Sounds like nobody, first-person "my results" script. No consent needed. Label it on Ads Manager. FTC: this is a testimonial from a testimonialist who does not exist, and 465.2(a)(1) is the paragraph that describes it. Rewrite the script or make the dramatisation obvious.
• Cloned voice of your own founder, with their sign-off. Consent exists; keep it in writing, because the TikTok One terms make you warrant it if the ad is commissioned there. Label: on YouTube, "Cloning one's own voice to create voice overs or dubs" is exempt; on TikTok organic, the label attaches when AI audio "mimics the voice of a real person", which this does, so label it there and on Ads Manager.
• Voice resembling any real person, public or private, no consent. Stop. Tennessee 47-25-1105(a)(1) on the consent side, TikTok Community Guidelines NOT ALLOWED on the platform side, and the Seller Center's "false endorsements" sentence on the Shop side. No label cures it.
• Any AI voice in a TikTok Shop LIVE. Prohibited regardless of source or label.

None of this is legal advice; it is a reading of the documents below, which can change without notice. Only the second question needs re-checking when a platform rewrites a page. The first is settled by whose voice it is, the third by which model you picked.

Sources

• State of Tennessee, Public Chapter No. 588 (HB 2091), Ensuring Likeness, Voice, and Image Security Act of 2024, approved 21 March 2024, effective 1 July 2024 (read as the enrolled PDF)
• Legal Information Institute, 16 CFR 465.1 Definitions and 16 CFR 465.2 Fake or false consumer reviews, consumer testimonials, or celebrity testimonials
• Legal Information Institute, 16 CFR 255.0 Purpose and definitions (endorser "could be or appear to be an individual, group, or institution")
• Federal Trade Commission, Consumer Reviews and Testimonials Rule: Questions and Answers
• TikTok Community Guidelines, Integrity and Authenticity: Edited Media and AI-Generated Content (AIGC)
• TikTok Advertising Policies, Misleading and false content
• TikTok Ads Help Center, How to add disclaimers to ads in TikTok Ads Manager
• TikTok Shop Seller Center (US), AI-Generated Content Restrictions and Requirements, updated 2 September 2026
• TikTok Shop Seller Center (UK), Artificial Intelligence Generated Content (AIGC), updated 28 August 2026
• TikTok Shop Seller Center (US), Requirements for High-Quality Videos and LIVEs, updated 31 July 2026
• TikTok Ads Help Center, TikTok One Creative Services Advertiser Terms of Service, last updated May 2026
• YouTube Help, Disclosing use of GenAI content (answer 14328491)
• EU AI Act, Article 3, point (60) (definition of deep fake)
• Replicate model pages for minimax/speech-02-hd and prunaai/p-video-avatar

All of the above were read directly on 4 September 2026. The ELVIS Act was read as the enrolled public chapter because the Tennessee Code host and the bill PDF were unreachable that day.

The voice is one of several places an AI ad can pick up a rule. For the visual side, Meta labels detected AI itself, and YouTube's upload rules and Google's ad rules land on the same Shorts file. UGC Vids AI renders the ads on the models above, from $49/month, free for 3 days.