EFF guide: how to opt out of AI training on ChatGPT, Gemini, Claude, Copilot and other major platforms

What this guide covers

The Electronic Frontier Foundation (EFF) maintains a step-by-step guide, updated 1 October 2026, showing users how to opt out of AI training and disable AI features across major platforms. For chatbots, it links to each service's own documentation for stopping conversations being used for training: ChatGPT (OpenAI), Perplexity, Gemini (Google), Claude (Anthropic) and Copilot (Microsoft). For social networks it gives concrete steps, including: LinkedIn (Microsoft) — Settings & Privacy > Data Privacy > 'Data for Generative AI Improvement', turn off use of your data for training content-creation AI models; X — Settings and privacy > Privacy and safety > 'Grok & Third Party Collaborators', uncheck unwanted options (xAI); Twitch (Amazon) — Settings > Security and Privacy, uncheck 'Training for Generative AI'; and for Instagram and Facebook (Meta), setting the account to private as the main available control. It also covers browser and phone AI features and Google services such as Gmail and Workspace smart features.

Why it matters to creators

These are the actual controls that determine whether material a creator types, uploads or posts on each platform can be used to train that company's models. The guide makes the differences visible in one place: some services offer a direct training opt-out, some only offer indirect controls such as private accounts, and the steps and availability vary by platform and region. It is a practical reference for creators who want to limit training use of their work on services they already use.

What is not yet known

EFF notes the directions may only be necessary in the U.S., may not be available in all regions, and change often; the page states it is up to date as of 1 October 2026. An opt-out generally stops future training use of covered content; the guide does not establish that opting out removes material already used in training.