Your Twitch Chat Trains Amazon's AI, Even If You Never Stream

A quick message in your favorite Twitch streamer’s chat could become training material for Amazon’s generative AI.

Twitch has added a setting that lets streamers stop Amazon from using their channel content for future AI training. The catch is that the setting is turned on by default.

“Channel content” also reaches beyond the streamer’s broadcasts. According to Twitch’s account settings FAQ, it includes the messages viewers post in stream chat.^[1]

That leaves regular Twitch viewers with surprisingly little control. If you post in someone else’s chat, that channel owner’s setting determines whether your message may be used for training. Turning off AI training on your own Twitch account does not protect what you write in another streamer’s chat.

Twitch also has no plans to show viewers which streamers have opted out, according to comments made by company leaders during a livestream about the change. Unless a streamer tells you, you may have no way to know what happens to your messages before you hit send.

Twitch turns Amazon AI training on by default

On August 12, Twitch announced that it had added a setting allowing users to opt out of having their channel content used to train generative AI models across Amazon.

The announcement focused on the arrival of an opt-out control. Amazon had already been using at least some Twitch data for AI development.

Twitch chief product officer Mike Minton acknowledged at a 2024 event that Amazon had used Twitch content for AI work in a “prototyping” capacity, according to reporting on the company’s earlier experiments.

During Twitch’s August 2026 livestream, however, Minton said he could not explain exactly which content Amazon had used or which models had received it.

“I don’t actually know the answer to that question because I don’t know what Amazon… has done in terms of model training and what they’ve used and not used,” Minton said during the livestream.

His explanation for making the new setting opt-out was considerably clearer.

“If it was opt-in, nobody would opt in,” Minton said. “That’s honestly the answer. So it’s going to be on by default.”

Twitch users’ resistance should not have come as a surprise. An All About Cookies survey found that 45% of Americans believe companies should only use personal data for AI training when users opt in.

Additionally, default settings can shape decisions before users realize there is a choice. This tactic, known as preselection, is a common type of dark pattern.

Naturally, the admission drew an angry response from Twitch users, many of whom argued that low participation ought to be a reason to seek clearer consent.

Minton said Twitch was following a common approach among content platforms while giving creators a way to refuse future training.

Twitch head of community Mary Kish acknowledged the mood during the same broadcast.

“We don’t expect you to be happy or excited about this,” she said.

What Twitch may use to train Amazon’s AI

Twitch says content from a participating channel may be used to improve future generative AI models developed by Amazon. That content can include:

• Livestreams and videos on demand
• Clips and highlights
• Stream chat
• Pictures and text published on the channel

These models may generate or synthesize text, audio, images, or video. Their use is not necessarily confined to Twitch.

As one example, Twitch says audio from a stream could help Amazon refine speech-to-text models. That might improve captions on Twitch while also improving captions elsewhere at Amazon.

The company has not publicly identified every Amazon model receiving Twitch content or provided a complete list of planned uses. During the livestream, Kish said Twitch knows the content is being used to train models, though she argued that listing hundreds of possible applications would not be helpful.

For viewers, “stream chat” is the crucial part of Twitch’s definition. Chat can contain reactions, jokes, personal stories, and conversations between viewers. All of it takes place on the streamer’s channel, so the streamer’s preference governs whether those messages are eligible for Amazon’s training.

Your own opt-out does not follow you into other chats

Imagine that you open your Twitch settings and turn off generative AI training. Content associated with your channel should no longer be used for future training covered by the setting.

Later, you visit a streamer who has left the setting on and post a message in their chat. Twitch considers that chat part of the streamer’s channel. The streamer’s choice applies to your message, even though you opted out through your own account.

“If you chat on someone else’s stream, their opt-out preferences govern if that chat can be used for training,” Twitch says.

There is currently no badge or warning that tells viewers whether a channel participates in AI training. Minton said Twitch does not plan to add that kind of labeling, according to coverage of the company’s question-and-answer stream.

You can ask a streamer whether they have opted out, but you must rely on their answer. Their settings may also change later.

The mismatch makes Twitch’s opt-out less useful for viewers than it first appears. You control whether Amazon can use content on your channel, but you cannot carry that preference into every community you visit.

How to turn off Twitch AI training

Even with those limitations, switching off the setting can protect content associated with your own channel from future generative AI training.

On the Twitch website:

1. Sign in and select your profile picture

2. Choose Settings.

3. Open Security and Privacy. It's the middle tab on the browser version.

4. Scroll down to Training for Generative AI. It's near the bottom.

5. Turn off the setting that allows your channel content to train generative AI models at Amazon.

In the Twitch mobile app, open your profile, go to Settings, then Security and Privacy. You only need to change the setting once, though. The preference should carry across the website and mobile app.

After switching it off, refresh the page and confirm that the toggle remains disabled.

If you regularly stream, consider telling viewers that you opted out. Twitch does not provide that information to them, and the setting affects their chat messages too.

What Twitch’s AI opt-out does not cover

The wording of Twitch’s policy is forward-looking. Turning off training means your channel content will not be used for “future” improvements to the Amazon models covered by the setting.

Twitch does not promise that opting out will remove content already used to train a model. Once information has been incorporated into model training, reversing that process is considerably more complicated than deleting a stored post or video.

The switch also does not disable every use of AI or machine learning on Twitch. The platform may continue using channel content for existing features such as AutoMod, captions, viewer recommendations, and sponsorship assistance. Twitch says systems such as AutoMod do not retain your content to produce new material in the way a generative model does.

Turning off the setting also does not prevent outside companies from collecting publicly available Twitch content. Minton argued during the livestream that other AI developers may already scrape public material without Twitch or Amazon’s permission.

That wider problem does not change the choice Amazon has made to train on Twitch content by default, though it does explain why no privacy toggle can guarantee that a public stream stays out of every AI dataset.

How Twitch viewers can limit what reaches AI training

Twitch has not given viewers a platform-wide way to keep their messages out of participating channels. Until that changes, your safest option is to treat public stream chat as public data.

Avoid posting details you would be uncomfortable seeing copied, analyzed, or retained elsewhere. That includes your full name, location, workplace, email address, and information about other people. A streamer opting out reduces one specific use by Amazon, though it does not turn Twitch chat into a private conversation.

If those details are already circulating online, start with our guide to removing your personal information from the internet.

You can also ask streamers to disable AI training and explain that the choice affects their viewers. Streamers can place that information in their channel description if they want community members to know where they stand.

Simply watching a stream without chatting does not add your words to its chat log. For viewers who cannot confirm a channel’s setting, staying out of chat is the clearest way to prevent a new message from becoming part of that channel’s training-eligible content.

Bottom line

Twitch’s new setting gives channel owners some control over future generative AI training. Viewers remain tied to choices made by each streamer they watch.

You can disable training on your own account and still have a message become eligible for Amazon’s models when you post it somewhere else. Twitch does not warn you when that happens, and its opt-out language offers no guarantee about content used before you found the switch.

If you use Twitch, turn off the setting for your own channel and be more selective about what you type in public chats. Until Twitch gives viewers a preference that follows them across the platform, someone else may be deciding whether your words help train Amazon’s next AI.