AI Voice Commercial Use: Licensing and Copyright

Yes, you can use an AI voice commercially. On a paid plan, Kukarella and most of the voice platforms we checked let you put the audio in a course you sell, training for your own staff, a monetized video or an ad. But a plan licenses only one thing: the voice. Three more questions decide whether you are safe: whose words the voice reads, whether the voice belongs to a real person, and what the place you publish requires. And one surprise. In the United States, the voice an AI generates from your typed text is very likely nobody's copyright. Your script stays yours.

The four checks:

1. The voice. Does your plan allow commercial use, and what survives if you cancel?
2. The words. Did you write the script, or do you hold a licence for it?
3. The person. If the voice is a clone or a sound-alike of a real person, do you have their written consent?
4. The place. What do YouTube, Udemy, Audible or your podcast app require?

We make Kukarella, an AI voice platform, so read us with that in mind. Where a human narrator or a lawyer is the better answer, this article says so. It is general information, not legal advice. Every rule below is as we read it on 3 October 2026.

Answer four questions and get a checklist for your own project:

What a commercial licence gives you, and what it does not

A commercial licence is a contract. The platform lets you use the audio it generates for business, and the contract decides where that audio may go. But a licence cannot give you rights the platform never had, and it gives you no rights in the words you feed it.

"Commercial" covers more than selling. It sounds like a paid course or an ad. The vendors draw the line much earlier. NaturalReader writes that "even internal training videos or classroom presentations that are not public still require a commercial license". TED's usage policy counts "employee learning" and "training and educational materials" as commercial use of its talks. Our own home page answers the question "Can I use the audio in a course I sell, or in our own training?" with "Yes, on any paid plan". So treat internal training as commercial unless your vendor says otherwise in writing.

Free plans are usually for trying, not for publishing. ElevenLabs says its free plan "cannot be used for any commercial purpose", and anything you publish from it must credit "elevenlabs.io" or "11.ai" in the title. On Kukarella, free-plan audio is for personal use only. Some free tools do allow commercial use with a credit line. Each character of the Japanese engine VOICEVOX has its own terms, and most require a credit such as "VOICEVOX:" and the character's name.

What survives when you stop paying is the question few guides answer. A course sells for years. A phone greeting plays for years. ElevenLabs says that once your subscription ends, "you will still have a commercial license to use whatever you generated during that subscription forever". Murf calls its rights lifetime, but you cannot download new copies after you cancel. Both promises sit in help articles, not in the terms of service, and not every vendor makes one. Read both before you publish.

On Kukarella, audio you made while on a paid plan keeps its commercial rights after you cancel. Audio you make after that, on the free plan, is for personal use only.

Some products inside a paid account carry their own limits. ElevenLabs' Studio terms make Studio output non-commercial unless you are on an Enterprise plan, while its own Studio guide says your plan decides. ChatGPT's voice output is "for non-commercial use only", although audio from OpenAI's developer API belongs to you. And Murf's help centre reserves TV and radio broadcast for its Enterprise plans. If your audio is heading to broadcast, ask in writing.

A library voice is shared. Every subscriber can pick the same stock voice, and the providers say so: OpenAI, Amazon and Google all warn that output may not be unique. A brand that needs a voice no competitor can use needs an exclusive contract: with a human actor, or for a custom voice.

Platforms disappear. By January 2026 Play.ht, a popular voice platform, had shut down its service. Its terms had made keeping copies the customer's job. Keep your files, and keep a dated copy of the terms you relied on.

On Kukarella, every paid plan shows the same line on the Pricing page: "Unlimited commercial use for any project". One condition sits in our FAQ: you must "hold the rights to the text you turn into speech". A Kukarella file carries no plan stamp or licence code, so your proof is your receipt and your script.

Kukarella's Prime plan card on 3 October 2026: commercial use is one line of the plan.

How to check any tool's terms in two minutes

Search the vendor's help centre and terms for these seven questions. Each one comes from a clause we found.

1. Does the free plan allow commercial use, and does it demand a credit? ElevenLabs wants it in the title. Speechify wants "Speechify.com" in the description.
2. Does the licence cover every product in my account? ElevenLabs Studio and ChatGPT's voice do not, as above.
3. Does "commercial" include my use: a classroom, internal training, broadcast, an ad? NaturalReader includes the classroom. Murf carves out broadcast. WellSaid bans whole topics, such as alcohol and gambling.
4. What happens when I cancel? "Forever" at ElevenLabs. Murf: lifetime, but no new downloads.
5. Who owns the audio, and is the voice mine alone? Most terms say the audio is yours, as far as any rights exist in it, and the voice stays shared.
6. Must I tell listeners it is AI? OpenAI's text-to-speech guide says yes. More on this below.
7. What if the voice or the vendor disappears? Microsoft promises 12 months' notice before it retires a prebuilt voice, unless security, legal or performance reasons force it sooner. Play.ht shut down its whole service.

Two of the four checks: a licence for the voice and the rights to the script.

Whose words? What is fine in class may not be fine in your shop

Your own script is yours to use anywhere. But words you borrowed follow their owner's licence, and the exceptions written for teaching stop where selling begins.

The biggest teacher marketplace, by its own account, says it plainly. Teachers Pay Teachers tells its sellers: "While it may have been okay to use these in your classroom, different rules apply when you post resources for sale or even share free resources on TPT." The UK's teaching exception allows "minor uses, such as displaying a few lines of poetry on an interactive whiteboard", only for a non-commercial purpose, and never "uses which would undermine sales of teaching materials".

Three kinds of borrowed words are easy to get wrong:

• A TED talk. TED licenses its talks under CC BY-NC-ND: no commercial use, no edits. TED owns the transcripts. It does not allow translation or dubbing outside its own volunteer translators. Using a talk "as a regular, structured component of a course" may need a separate licence from TED. So an AI re-voicing of a TED transcript breaks the licence even in a free lesson.
• A textbook passage or a song. Song lyrics are a written work. A chapter belongs to its author or publisher. A voice plan cannot license either; only the rights holder can.
• A lesson you wrote for your employer. In the US, a "work made for hire" is "a work prepared by an employee within the scope of his or her employment", and the employer owns it. Check your contract before you sell what you wrote at work.

A script an AI wrote is a special case. The U.S. Copyright Office says that with today's technology, "prompts do not alone provide sufficient control" to make you the author of what a model writes. What you add, cut, reorder and rewrite by hand can be protected. So rewrite the draft in your own words: the parts you write are yours.

You write, the AI reads: the words are the part you can own.

Who owns the audio?

Three different things decide it, and people mix them up. Your licence decides where you may use the audio. Copyright protects your script inside it. And the voice performance itself, generated by AI from typed text, is very likely nobody's in the United States. But a recording you perform yourself, then re-voice with AI, can be protected.

The U.S. Copyright Office put it in its January 2025 report. "Copyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements." And: "Human authors are entitled to copyright in their works of authorship that are perceptible in AI-generated outputs". Your script is perceptible in the narration. So is the way you built the course.

The courts agree that only a person can be an author. In Thaler v. Perlmutter, the appeals court in Washington held that the law "requires all eligible work to be authored in the first instance by a human being". It added that this "does not prohibit copyrighting work that was made by or with the assistance of artificial intelligence." The Supreme Court declined to hear the case on 2 March 2026, so that is final.

The clearest example is a comic. In 2023 the Copyright Office reviewed Zarya of the Dawn, whose author had made the images with Midjourney. It cancelled the registration and issued a narrower one. The text, and the way the author selected and arranged the words and images, stayed protected. The generated images did not, because "prompts function closer to suggestions than orders".

Voice has its own example, and it points the other way. The Office registered the recording of Where That Came From (registration dated 29 May 2024), a song released in Randy Travis's voice after a stroke had left him with limited speech. A human singer recorded the song, and an AI model turned that performance into Travis's voice. The Office's explanation: "Because the sound recording used AI as a tool, not to generate expression, the Office registered the work." A voice generated from typed text has no human performance in it. What you can protect there is the script, and any creative selection, arrangement or editing you do by hand.

The UK is different. There, a recording is protected even when no human performed it: the government's own report says sound-recording copyright "will also apply to a recording of music, film, etc. generated by an AI system". It belongs to the recording's producer; who that is for an AI voice has not been tested in court. British law also still protects "computer-generated" text and images, such as a script an AI wrote. On 18 March 2026 the government proposed removing that protection, but not the one for recordings, and no bill has followed yet.

On Kukarella, we claim no rights over what you make. That is a promise about us, not a copyright for you. Over what you upload, our Terms of Use take only the licence we need to run and improve the service.

Copyright protects the human part: the script you wrote.

A real person's voice: actors, colleagues, celebrities

A voice cannot be copyrighted. A real person's voice is protected anyway, by publicity law and by new digital-replica laws. So before you clone a real person, or imitate their voice to sell something, get their consent in writing. But even a stock voice was recorded by someone, and its licence is only as good as the platform's own consent.

The case that shows it. On 11 July 2023, two voice actors, Paul Lehrman and Linnea Sage, were driving in their car, listening to a podcast about the dangers of AI. The AI voice on the podcast was Lehrman's. Years earlier, buyers on Fiverr had paid him $1,200 and her $400 for recordings they said were for research and internal tests. The actors say the company behind those buyers, Lovo, then sold their cloned voices under the names "Kyle Snow" and "Sally Coleman". The podcast had been made on a paid Lovo plan, and Lovo told its customers its paid plans carried "full commercial rights for all content generated". In July 2025 a federal judge in New York dismissed the copyright claims over the voices: copyright does not protect a voice as such. But the actors' claims under New York's publicity law, consumer-protection law and contract law went ahead. The judge wrote: "Lovo cannot escape liability merely because it appended fictitious names to those appropriated voices." Nobody has won yet: the case has been on hold since Lovo filed for bankruptcy in May 2026.

Closer to home, Ontario voice actor Bev Standing recorded about 10,000 sentences for what she understood was a translation app. Her voice became TikTok's text-to-speech voice. She sued in 2021. The case settled, and her lawyer said TikTok would license her voice as part of the deal. Her words: "I just wanted a choice."

How to license a voice actor's voice for AI

It is the search that most often brings up this page in Google, and the answer has known parts. Between them, the union deals for digital voice replicas, the voice actors' own checklist and two state laws set out these terms:

Two warnings. First, the actor's existing contracts may already forbid all of this. The National Association of Voice Actors publishes a free rider for contracts that bars clients from using an actor's recordings to simulate their voice, to build a blended voice from them or to train AI. Read the actor's existing contracts before you build anything. Second, in California and New York, a clause that lets a digital replica replace work the person would have done can be unenforceable if it does not describe the intended uses. It stays enforceable if a lawyer negotiated the replica licence for the person and the terms are set out clearly in a writing they signed, if their union's agreement expressly addresses digital replicas, or if the uses fit the contract and the character of the original recording.

The union's own deal with the ad platform Narrativ shows what good terms look like. Actors set their own price above a union minimum, can opt out of categories of ads, and are paid for every use. And if an actor leaves, "Narrativ must delete your digital voice replica, and any recordings used to create it."

Cloning a colleague, or yourself

A trainer can clone a colleague's voice in minutes. Undoing it is harder. Get consent in writing that names the uses and how long they last. Microsoft, for example, makes its customers obtain "explicit written permission from voice talent" and a recorded acknowledgement before it builds a custom voice. In the EU, an employee's yes is weak ground: European data regulators find it "unlikely to be freely given". Tennessee's ELVIS Act, in force since July 2024, protects a person's voice even when the sound is "a simulation of the voice". And in Canada, Quebec's Civil Code says that using a person's "name, image, likeness or voice" for anything but informing the public may invade their privacy, and privacy needs their consent. Manitoba and Newfoundland and Labrador treat using an identifiable person's voice for advertising, or for the user's gain or advantage, as a breach of privacy.

On Kukarella, our Terms require "all necessary rights, licenses, consents, and permissions" for any voice sample you upload. Our clone screen does not ask for that consent or keep a record of it, so keep the signed consent yourself. And there is no button to delete a cloned voice. A deletion request goes through our support to the company that makes cloned voices, which holds the voice under its own retention policy. If the person may want their voice back one day, plan for that before you clone. A voice designed from a written description starts from no one's recording, unless the description is written to sound like a real person, and then the consent question comes back.

Your own voice is the simple case. YouTube does not even ask you to label "Cloning one's own voice to create voice overs or dubs". And you cannot copyright your voice itself, only your recordings.

A famous voice

Imitating a famous voice to sell something is a classic lawsuit, and no recording needs to be copied. In 1988 an appeals court ruled for Bette Midler after Ford's ad agency hired a singer to sound like her: "A voice is as distinctive and personal as a face." In 1992 Tom Waits kept $375,000 in damages and $2 million in punitive damages over a Doritos radio ad that imitated his voice. In April 2024 Drake took down a track with an AI voice of Tupac Shakur within days, after the estate's lawyer called it a "flagrant violation of Tupac's publicity and the estate's legal rights".

Even a legitimately cast actor can sound too close. When ChatGPT's "Sky" voice reminded people of Scarlett Johansson in May 2024, her lawyers wrote to OpenAI and OpenAI paused the voice. She did not sue. So never brief a voice as "like" a famous person.

A correction. An earlier version of this page said that in 2020 a company used AI to make Jay-Z recite Shakespeare, and that this violated his right of publicity. It was not a company, and no court said so. The videos came from an anonymous hobby channel on YouTube. Jay-Z's company filed copyright takedown requests, and YouTube put the videos back up for the time being, calling the requests "incomplete". No court ruled.

The US still has no federal right in a person's voice. A bill that would create one, the NO FAKES Act, passed the Senate Judiciary Committee in June 2026. On 30 September 2026 a senator blocked a quick passage, and it is still waiting on the Senate calendar. Until it passes, state laws decide.

If you need a famous voice, ask the rights holder. Some estates now license voices through brokers, and the rights holder decides on each request.

Where you publish: what YouTube, Udemy, Audible and podcast apps require

No major platform bans AI narration outright. Audible's self-publishing platform, ACX, is the exception, for now. But YouTube refuses to pay for mass-produced videos, and several platforms want a label in some cases.

Platforms set their own rules for AI voices, on top of the law.

Here is what each place asked for when we checked:

The sources, all read on 3 October 2026: YouTube's monetization policies and rules for disclosing AI content; Udemy's course quality checklist on AI and its course description rules; ACX's audio requirements; Google Play Books' guide to adding a book; Apple Podcasts' content guidelines, rule 1.11; and the US Federal Communications Commission's ruling of February 2024.

YouTube never banned AI voices. On 15 July 2025 it renamed its "repetitious content" rule to "inauthentic content" and called the change "a minor update". What fails is content made with "generic or unoriginal templates giving the impression of mass production". An original lesson with an AI narrator is on the right side of that line.

Udemy asks for one exact sentence. If AI tools "generate artificial video or speech meant to emulate human instruction", your course description must begin with: "This course contains the use of artificial intelligence."

Audible is the hard no, for now. ACX says: "Your submitted audiobook must be narrated by a human unless otherwise authorized." Audible says it is working to accept outside text-to-speech, but has not said when. If your course is heading to Audible, budget for a narrator.

Do you have to tell listeners it is AI?

Often, yes, and it costs a sentence or two. OpenAI, which makes some of the voices in our library, tells developers in its text-to-speech guide: "Our usage policies require you to provide a clear disclosure to end users that the TTS voice they are hearing is AI-generated and not a human voice." The EU now requires it for deep fakes. Usually a spoken line at the start and a line in the description are enough, and YouTube says a label "won't limit a video's audience or impact its eligibility to earn money."

The EU rule. Since 2 August 2026, anyone who uses AI to make a deep fake for work or business must disclose that it was artificially generated or manipulated (AI Act, Article 50). The Act defines a deep fake as content that resembles existing people or things and "would falsely appear to a person to be authentic or truthful". The Commission's guidelines of 20 July 2026 give a voice-cloned podcast presenter as an example, and the voluntary code of practice it approved asks for a short spoken disclaimer at the start of an audio deep fake. A clone of your colleague that listeners could mistake for her fits that definition. A realistic stock narrator can count too: the Commission says resembling a person who could plausibly exist is enough, and the test is whether listeners would take the voice for a real person's. And audio you made before 2 August 2026 does not need to be labelled retroactively. Either way, one line settles it.

Microsoft lets the disclosure be light. It requires its customers to disclose synthetic voices, but says the disclosure "may be minimal, momentary, or even unnecessary" when the context already makes it obvious.

The simplest habit: say it at the start, and write it in the description. Here is how short it can be:

Keep these records

Nothing inside a downloaded audio file proves your plan, your licence or the source of your script. If a client, a marketplace or a YouTube dispute ever asks, these five things answer:

1. The receipt for the plan you were on when you made the audio.
2. A dated copy of the terms and the plan page you relied on. Pages change, and some platforms disappear.
3. Your script and its sources, with any licence for words you borrowed.
4. Signed consent from every real person whose voice you cloned.
5. The date you generated each file, in your project or your file names.

When to hire a narrator or a lawyer instead

An AI voice is the right tool for most lessons and training. Four cases are not most cases.

• An audiobook for Audible. ACX requires a human narrator unless Audible authorizes otherwise.
• A brand voice nobody else may use. A library voice is licensed to everyone who pays for it. Exclusivity needs an actor under contract, or a custom voice under a written licence.
• A specific real person's voice: a colleague, an actor, your founder. Have a lawyer write the consent. Cloning takes a recording; undoing it may not be possible.
• Someone else's words in something you sell: a textbook chapter, a TED transcript, song lyrics. Only their owner can license them. No voice plan covers the text.

Check yourself. Three quick questions on what you have read:

Frequently asked questions

Can I use AI voices commercially without copyright issues?

Yes, if three things hold. Your plan grants commercial use. You hold the rights to the script. And the voice is not a clone or a sound-alike of a real person who has not agreed. Copyright problems come from the words more often than from the voice: a voice cannot be copyrighted, but a script can.

Who owns the copyright to an AI voiceover?

In the US, you own the copyright in your script, and in how you edit and arrange the work. The synthetic performance generated from typed text is very likely not protected for anyone, per the U.S. Copyright Office's January 2025 report. In the UK, the recording itself can be protected as a sound recording, owned by whoever arranged its making, though no court has yet applied that to AI audio. Your platform's licence decides where you may use the audio.

Is AI voiceover illegal?

No. Using a licensed AI voice for narration is legal. The risk is mostly civil, a lawsuit or a takedown, though some misuse is a crime: in Tennessee, using a person's voice without consent can be prosecuted as a misdemeanour. It comes from three places: words you have no rights to, a real person's voice used without consent, and breaking a platform's terms. One US rule comes close to a ban: an AI voice in a call you place to someone's phone counts as an "artificial voice", so they must have agreed to the call first.

Can AI-voiced videos be monetized on YouTube?

Yes. YouTube does not ban AI voices. It refuses to pay for templated, mass-produced or reused content, however it was made, and for AI personas presented as human experts on health, law, money or politics. An original lesson with an AI narrator can earn. A video needs a label when realistic AI content could pass for real, such as a real person saying what they never said or altered footage of a real place, and when AI music is its main focus.

Can I use a celebrity's AI voice?

Not without the rights holder's permission. A voice cannot be copyrighted, but imitating a famous voice to sell something has cost advertisers millions (Waits v. Frito-Lay, 1992), and laws such as Tennessee's ELVIS Act now cover simulated voices. Parody and commentary have some protection, but only as far as free-speech law reaches.

How do I license a voice actor's voice for AI?

Get written consent that names the uses, the languages, the term, the pay and what happens at the end, including deletion of the voice model. Check the actor's existing contracts for a rider that forbids AI use. In California and New York, a clause that lets a digital replica do work the actor would otherwise have done in person, without a reasonably specific description of the uses, can be unenforceable, unless a lawyer negotiated it for the actor in a signed writing, their union's agreement covers replicas, or the uses fit the original contract.

Can you copyright your voice?

No. A voice is not a "work" under copyright law. Your recordings are protected, and so is a script you wrote. Your voice itself is protected in other ways: by publicity laws in many US states, and by new digital-replica laws such as Tennessee's ELVIS Act.

Do I have to credit or disclose the AI voice?

Sometimes. Free plans often require a credit line (ElevenLabs: in the title). OpenAI's text-to-speech guide asks that listeners be told its voices are AI. Udemy requires one sentence in the course description. The EU requires disclosure of deep fakes. A spoken line at the start and a line in the description cover most cases. Udemy's sentence and YouTube's "AI use" setting are separate steps.

Is internal training commercial use?

Usually, yes, for the vendors that define it. NaturalReader says even internal training videos need a commercial licence, and TED counts employee learning as commercial use of its talks. On Kukarella, audio for your own staff's training needs a paid plan, and you must hold the rights to the text you turn into speech.

What happens to my rights if I cancel my subscription?

It depends on the vendor. ElevenLabs keeps the licence for audio made while you paid "forever". Murf calls it lifetime, but stops new downloads. On Kukarella, audio you made while on a paid plan keeps its commercial rights after you cancel, and new audio on the free plan is for personal use only. Read the clause before you publish anything that will keep earning.

How we checked

We read the laws, court opinions and platform pages linked here on 3 October 2026. Kukarella's own statements come from our live Pricing page, home page FAQ and Terms of Use, and from the product itself, read the same day. An earlier version of this page named the U.S. Patent and Trademark Office where it meant the U.S. Copyright Office, said a comic's copyright was revoked when it was narrowed, and said a company's AI recording of Jay-Z violated his right of publicity. All three were wrong, and this version corrects them.

About the author. Nazim Ragimov has been building Kukarella since 2019, first for himself and his team, then for the users whose feedback shaped almost everything in it. Most of them teach, and he wants Kukarella to delight them and their students: a lesson that makes the class say "Now we get it." He started these articles himself, with help from his wife, Anna; today he sometimes uses AI for research and first drafts.

Sources

All pages read on 3 October 2026 unless stated.

Copyright and AI

• U.S. Copyright Office. Copyright and Artificial Intelligence, Part 2: Copyrightability. 29 January 2025. PDF
• U.S. Copyright Office. Copyright and Artificial Intelligence, Part 1: Digital Replicas. 31 July 2024. PDF
• U.S. Copyright Office. Letter on Zarya of the Dawn, 21 February 2023. PDF
• Thaler v. Perlmutter, No. 23-5233 (D.C. Cir., 18 March 2025). Opinion; certiorari denied 2 March 2026. Docket
• 17 U.S.C. §101, "work made for hire". Cornell LII
• UK Intellectual Property Office. Exceptions to copyright. gov.uk
• UK Government. Report on copyright and artificial intelligence, 18 March 2026. gov.uk

A person's voice

• Lehrman v. Lovo, Inc., No. 24-CV-3770 (S.D.N.Y., 10 July 2025). Opinion
• Midler v. Ford Motor Co., 849 F.2d 460 (9th Cir. 1988). Opinion
• Waits v. Frito-Lay, 978 F.2d 1093 (9th Cir. 1992). Opinion
• Tennessee ELVIS Act (HB 2091), in force 1 July 2024. Tennessee General Assembly
• NO FAKES Act of 2026 (S.4591), on the Senate calendar since 24 June 2026; floor objection 30 September 2026. Congress.gov
• California AB 2602 (Labor Code §927). leginfo
• New York S7676B (General Obligations Law §5-302). NY Senate
• National Association of Voice Actors. Synth/AI and the AI rider. navavoices.org
• SAG-AFTRA and Narrativ, digital voice replica agreement, as reported by Silicon Republic, 14 August 2024. Article
• Quebec Civil Code, article 36. LégisQuébec; Manitoba Privacy Act, s. 3. Manitoba Laws; Newfoundland and Labrador Privacy Act, s. 4. House of Assembly
• European Data Protection Board. Guidelines 05/2020 on consent. PDF
• Microsoft. Disclosure for voice and avatar talent. Microsoft Learn

Stories

• Twenty Thousand Hertz. TikTalk (Bev Standing), 24 August 2022. Episode; The Verge, the settlement, 29 September 2021. Article
• NBC News. Drake pulls "Taylor Made Freestyle", 26 April 2024. Article
• NBC News. Scarlett Johansson and OpenAI's Sky voice, 20 May 2024. Article
• Andy Baio, Waxy.org. Jay-Z and the Vocal Synthesis takedowns, 28 April 2020. Article

Vendors' terms

• ElevenLabs. Can I publish the content I generate on the platform? Help; What happens to my content after my subscription ends? Help; Studio terms, 7 May 2026. Terms; Studio guide. Docs; Iconic Marketplace. Page
• Murf. Do I have commercial rights over the voice-over created? Help; Will my Commercial Rights and Business License be valid after my subscription has ended? Help; What happens to my account and projects after canceling? Help
• NaturalReader. Commercial version. Help
• OpenAI. Text-to-speech guide. Docs; Service Terms. Terms; Services Agreement. Terms
• Amazon Web Services. Service Terms. AWS; Google Cloud. Service Specific Terms. Google Cloud
• Speechify. Terms of Service for Speechify Studio. Terms
• Play.ht home page, archived 6 January 2026. Wayback Machine
• Microsoft. Text to speech transparency note. Microsoft Learn; disclosure guidelines. Microsoft Learn
• WellSaid Labs. Acceptable use policy. Policy
• VOICEVOX. Terms of use. Terms
• TED. TED Talks Usage Policy. Policy
• Teachers Pay Teachers. Seller basics: copyright and trademark on TPT, 12 September 2024. Seller Blog
• Kukarella. Pricing, home page FAQ, Terms of Use

Platform rules

• YouTube. Channel monetization policies. Help; Disclosing use of GenAI content. Help
• Google. Quickstart guide to add a new book, Google Play Books Partner Center. Help
• Apple. Content guidelines, Apple Podcasts for Creators, rule 1.11. Apple
• Udemy. Course quality checklist: use of AI. Help; Course description: rules and guidelines. Help
• ACX. Audio submission requirements. Help
• Federal Communications Commission. Declaratory ruling FCC 24-17, 8 February 2024. PDF
• European Commission. Guidelines on transparency obligations for providers and deployers of AI systems, C(2026) 5054, 20 July 2026. European Commission; Code of Practice on Transparency of AI-Generated Content, 10 June 2026. European Commission
• Regulation (EU) 2024/1689 (the AI Act), Articles 3(60) and 50. EUR-Lex; the Commission's AI Act Service Desk, Article 50 and Article 3; Transparency obligations under Article 50, FAQ. European Commission