How to Protect Your Music From AI Training

AI music generators learn from recordings, and AI chatbots learn from web pages, which can include the lyrics and bios musicians post online. This guide is for independent musicians who want a say in how AI companies use their work.

It shows how to opt out on your own website and still appear in Google Search and in ChatGPT's and Claude's search answers. It also covers what to check on the platforms you upload to, how to prove what's yours, what to look for in contracts and where the law stands.

TL;DR

• Opt out of AI training on your own website by naming the main AI companies' training bots in its robots.txt file, or with your website builder's AI setting.

• Block only the training bots if you want ChatGPT and Claude to keep finding you and linking to your pages.

• Check every platform you upload to. YouTube only shares your videos with outside AI companies if you turn that on. SoundCloud's terms say it needs your opt-in to train AI that imitates your voice or music.

• Register your songs and recordings, and keep your session files, so you can prove what's yours.

• Read the AI clause in any label, publishing or distribution deal before you sign.

Table of contents

• Can you protect your music from AI training?

• How AI companies get music

• Block AI training bots on your own website

• What a robots.txt opt-out can't do

• Should you block AI search bots too?

• Check the AI settings on every platform you use

• Make your rights easy to prove

• Read the AI clause before you sign

• Where the law stands

• Common mistakes to avoid

• Common questions

Can you protect your music from AI training?

You can protect it in part. Opt out of AI training on your own website, check the AI settings on each platform you upload to, and register your songs and recordings. A website opt-out applies once each AI bot reads your robots.txt file. It can't undo past training, and it relies on AI companies respecting it.

How AI companies get music

In a 2024 court filing, Suno said its model was trained on tens of millions of recordings. Its training data, the filing said, includes "essentially all music files of reasonable quality" on the open internet. Suno also said it abided by "paywalls, password protections, and the like".

GEMA, the German music collecting society, sued Suno over six songs. Suno had admitted training on them, and the songs had been copied from YouTube by stream-ripping, which gets around YouTube's protection against downloads.

For text, AI companies send crawlers across the web. OpenAI's GPTBot and Anthropic's ClaudeBot, for example, collect pages that may be used to train their AI models. The internet company Cloudflare reported in August 2025 that 80% of the AI crawling it saw over the previous 12 months was for training.

Lyrics, bios and news pages on musicians' websites can end up in that crawl. GEMA has also taken OpenAI to court over song lyrics. A Munich court ruled for GEMA in November 2025, and that ruling has been appealed.

Block AI training bots on your own website

On your own website, you decide what AI bots are asked to do. The big AI companies say their crawlers follow a file called robots.txt, which tells each bot which pages of a site it may visit.

Each opt-out takes two lines: User-agent: followed by the bot's name, then Disallow: /, which asks that bot to stay off the whole site. These are the main names to list:

• GPTBot: OpenAI's crawler for AI training data

• ClaudeBot: the crawler Anthropic uses to collect training data for Claude

• Google-Extended: tells Google not to use your pages to train Gemini or in the Gemini app's answers

• Applebot-Extended: tells Apple not to use your pages to train its AI models

• meta-externalagent: Meta's crawler, which it uses to train AI models

• Amazonbot: Amazon's crawler, which Amazon says may be used to train its AI models

• CCBot: Common Crawl's crawler, which builds a free archive of the web that anyone can use

The two "Extended" names don't take you out of search. Google says Google-Extended doesn't affect whether you appear in Google Search. Apple says pages that block Applebot-Extended can still show up in its search results.

Microsoft has no bot name to list. Bing said in 2023 that it won't use pages with a noarchive tag in their code to train Microsoft's AI models. Those pages stay in Bing's search results but drop out of its chat answers.

How you add the lines depends on your website builder. If it has an AI or crawler setting, use that. If it lets you edit robots.txt, add the lines yourself, and if it offers neither, ask its support team.

On Noiseyard, every website opts out of AI training unless you change it under "AI and your website" in Website Settings. The help article on the AI setting explains each option and what it doesn't cover.

Whatever you use, check the result. Add /robots.txt to the end of your website's address and open it in a browser. Each AI name you blocked should have a Disallow: / line under it, sometimes after a few more names.

What a robots.txt opt-out can't do

A robots.txt opt-out has three limits:

• Past training. An opt-out can't undo training that already happened. Anthropic says blocking ClaudeBot signals that "future materials" should be left out of its training data. Google-Extended covers the training of "future generations" of Gemini.

• Bots that ignore it. The robots.txt standard says its rules "are not a form of access authorization", so they can't lock a bot out. TollBit, a company that tracks AI bots on publishers' sites, reported that about 15% of AI scrapes there ignored a robots.txt block in the first half of 2026.

• Embedded players and copies elsewhere. Your robots.txt covers its own web address and nothing else. Players you embed from Spotify, YouTube, SoundCloud or Bandcamp, files loaded from another address and copies of your songs on other sites all fall outside it.

Should you block AI search bots too?

Block them if you want chatbots to stop showing your pages. OpenAI and Anthropic use separate bots for training and for search. OpenAI's GPTBot collects training data, and OAI-SearchBot finds websites for ChatGPT's search answers. OpenAI says the two settings are independent, so you can block training and still appear in ChatGPT search.

At Anthropic, ClaudeBot collects training data and Claude-SearchBot indexes pages for search. Claude-User opens pages when someone asks Claude a question.

Blocking the search bots can take you out of those answers. OpenAI says sites that block OAI-SearchBot "will not be shown in ChatGPT search answers," though they can still appear as navigational links. Anthropic says blocking Claude-SearchBot can make you less visible in Claude's search results.

Google uses one name, Google-Extended, for training Gemini and for using your pages in the Gemini app's answers, so blocking it keeps your pages out of both.

AI Overviews and AI Mode in Google Search have their own setting in Google Search Console, called Search generative AI. If you choose Exclude in that setting, you won't get "any traffic or impressions" from those features. That setting doesn't change your normal search results, and it doesn't affect AI training. For ranking in those normal results, see SEO for musicians.

Check the AI settings on every platform you use

Music you upload to a platform follows its settings and terms, so check each one you use. The details below are from November 2026.

YouTube

YouTube lets you allow outside AI companies to train on your videos, and the setting is off by default. If you don't want your videos used that way, you don't need to do anything. Channel owners can check the setting in YouTube Studio, under the channel settings.

The setting is about companies separate from YouTube and Google. Google told CNBC in June 2025 that it trains its own AI models on a subset of YouTube videos and honors specific agreements with creators and media companies. CNBC reported that creators can't stop that training.

If a distributor claims your music through Content ID, its own setting applies to the videos it claims. YouTube shares a video only when every rights holder on it allows that company. For everything else about your channel, see how to use YouTube as a musician.

SoundCloud

SoundCloud's terms say it won't use your content to train AI that recreates your voice, music or likeness, unless you opt in. It revised its terms to say this in May 2025, after a backlash from users over an earlier AI clause. The promise covers AI that imitates you, so read the full terms if other uses matter to you.

Bandcamp

Bandcamp's robots.txt already lists GPTBot, ClaudeBot, Google-Extended, CCBot and other AI bots with Disallow: /, on the main site and on artist pages. Since January 2026, it has also banned music made "wholly or in substantial part by AI". Using AI tools to impersonate other artists or styles is banned too. If you're weighing other places to sell your music, Bandcamp alternatives compares them.

Your distributor

If you release music on your own, Spotify, Apple Music and other streaming services get it from your distributor, so its terms count too. In September 2026, Believe and TuneCore, its platform for self-releasing artists, partnered with Suno. Suno says their artists can choose whether to take part, and the companies say artists who consent will be paid.

Read what your distributor's terms say about AI, and check whether it can license your music without asking you. Our guide to releasing music on your own has a step on picking one.

Make your rights easy to prove

To take part in a lawsuit or a licensing deal, you'll need to show which songs and recordings are yours, and these habits make that easier.

Keep your session files, stems, dated demos and lyric drafts, and back them up somewhere safe. If you write with other people, a split sheet lists everyone who contributed to a song and the share each person owns.

In the US, register your songs and recordings with the Copyright Office. For a US work, the Office has to register or refuse it before you can sue over it. Register before anyone infringes the work, or within three months of its first publication, and you may also be able to claim damages set by law and legal fees.

For music, the Copyright Office says offering a recording to several streaming or download services usually counts as publication. So the three months can start when your distributor sends out a release, and registering early is the safe choice. If you wrote and recorded a song yourself, one application can cover both the song and the recording.

Wherever you live, join your country's collecting society for songwriters and register your songs with it. A society like GEMA can take an AI company to court for its members.

Read the AI clause before you sign

A label or publisher that owns your recordings or songs can license them to AI companies. Warner Music Group signed an AI licensing deal with Suno in November 2025, which also settled Warner's lawsuit against Suno.

In June 2026, the American Federation of Musicians (AFM), a musicians' union, sued Warner and Universal over their AI licensing deals. The union says the labels' deals with Udio, and Warner's deal with Suno, licensed recordings its members played on without paying or crediting those musicians.

Before you sign a label, publishing or distribution deal, ask whether it lets them license your music for AI, whether you can say no and how you'd be paid. Ask the same about sync and library deals, which the guide to getting your music in film and TV covers.

Where the law stands

As of November 2026, the big US cases are still open. Universal and Sony are suing Suno, in cases expected to test whether AI training counts as fair use. Independent musicians have also filed proposed class actions against Suno and Udio. In August 2026, a federal judge in Massachusetts let one of the cases against Suno go ahead on all but one claim.

In the EU, the AI Act requires companies that build general-purpose AI models to respect rights holders' opt-outs. Companies that signed the EU's voluntary code of practice for these models also promised to use crawlers that follow robots.txt. The signers include Amazon, Anthropic, Google, Microsoft, Mistral AI and OpenAI, and Meta said it wouldn't sign.

In Germany, a Munich court ruled largely for GEMA against Suno on July 31, 2026. It found that both Suno's training and the music it put out broke copyright law. The ruling isn't final and may be appealed.

For questions about your own songs, talk to a music lawyer or your collecting society.

Until the courts decide these cases, the steps above are what you can do now, starting with your own website. Fans who find you there through Google or a chatbot can also hear more of your music and join your mailing list. How to build an email list as a musician shows how to set up the signup.

Common mistakes to avoid

• Blocking every bot, or Googlebot. A User-agent: * group with Disallow: / applies to every bot without its own group, and rules for Googlebot affect Google Search itself. Name the AI bots one by one.

• Checking only your main address. A shop or blog on another subdomain needs its own robots.txt. Anthropic asks site owners to add its rule "for every subdomain" they want to opt out.

• Misspelling a name. Capital letters don't matter, but a missing hyphen or a typo can stop a rule from reaching its bot.

Common questions

Was Suno trained on my music?

It may have been, if your music was online. Suno hasn't published a list of the recordings it trained on. In a new lawsuit in September 2026, Universal and Sony named 60,202 of their recordings. They say an analysis of Suno's training data found millions of them.

Does blocking AI bots hurt my Google ranking?

No. Google says Google-Extended isn't "used as a ranking signal in Google Search." A crawler follows the lines under its own name when there are any, so rules for GPTBot or ClaudeBot don't apply to Googlebot.

Does Spotify allow AI training on my music?

Spotify bans apps built on its platform from training AI on Spotify content. For its own AI music products, it said in October 2025 that it would work through upfront deals with labels, distributors and publishers. It added that artists and rights holders will choose if and how to take part.

Can I let some AI companies train on my music and not others?

Yes. On your own website, block only the bots of the companies you want to keep out. YouTube's third-party training setting lets you pick companies from a list. Choosing "All third-party companies" also allows companies that aren't on that list. YouTube says it isn't arranging payments from any of these companies for now.

Do I need a website to protect my music from AI?

No. Platform settings and copyright registration work without one. If you have a website, set its AI rules as well. The other reasons to have one are in why every musician needs a website.