Opting Out of AI Training: The Settings Users Must Find First

Once your data trains an AI model, it is typically not possible to remove its influence on that model — and most AI tools collect that data by default, according to guidance originally published by the EFF and republished under CC BY 4.0. That asymmetry is the backdrop for a new platform-by-platform walkthrough showing where users can still say no.

Most chatbot and app vendors gather interaction data to "improve models," which often includes a human review step where a person manually checks AI outputs for accuracy. The controls exist, but they vary by platform and change frequently. Here is what the source found.

What are the chatbot opt-outs?

• ChatGPT (web, signed in): Settings > Data controls > turn off "Improve the model for everyone." Voice mode requires a separate request through OpenAI's Privacy Portal.
• Perplexity: The company says AI Data Retention is enabled by default for Free, Pro and Max users. Turn it off under Account > Preferences > "AI data retention."
• Gemini (web): Visit myactivity.google.com/product/gemini and switch off "Keep activity."
• Claude: Go to claude.ai settings > data privacy controls and turn off "Help improve our AI models."
• Microsoft Copilot: At time of testing, 10 November 2026, no visible opt-out existed in a personal account's settings on Copilot.com — despite Microsoft's own documentation referencing a "Training on conversation activity" control for some users.

The takeaway: four of five major chatbots offer a switch; Copilot was the exception in testing.

Which social networks let you say no?

Public posts fed much of the training data behind large language models, so social platforms are harder to police. On Instagram and Facebook, the practical move is setting your account to private. Other networks offer explicit toggles:

• LinkedIn: Settings & Privacy > Data Privacy > "Data for Generative AI Improvement" > disable "Use my data for training content creation AI models."
• Snapchat: Settings > Privacy Controls > "Manage My Information" > disable "Snap using your Public Content to train generative AI models."
• Twitch: Settings > Security and Privacy tab > uncheck "Training for Generative AI."
• X: Settings and privacy > "Privacy and safety" > "Grok & Third Party Collaborators" (x.com/settings/grok_settings) > uncheck options, including "Allow your public data as well as your interactions, inputs, and results with Grok and xAI to be used for training and fine-tuning."

What changed on phones?

Android and iOS keep adding AI features that depend on personal data, and turning them off now costs something.

Android. Until September 2026, users could roll a Pixel back to Google Assistant; that option is gone. Users can only disable Gemini as a voice assistant entirely or find an alternative. In Google Messages, disable "Show Gemini button" under Messages settings. In Google Photos, turn off "Use Gemini in Photos" under Preferences.

iOS. Since iOS 27, removal requires rolling back Siri: Settings > Screen Time > Content & Privacy Restrictions > Siri > "Allowed Siri Version" > select "Siri Classic." The same screen disables "Writing assistance," "Math Assistance" and "Image creation."

What does Google's wider ecosystem cost you?

Gmail bundles its AI into one "Smart features" toggle under Settings > General. Turning it off also kills smart compose, automatic filters and spellcheck. Google Workspace users can disable "Smart features in Google Workspace" via Drive's Privacy tab — losing Gmail-to-Calendar event syncing and personalized search results in the process.

Zoom, Adobe Acrobat and Notion all carry their own AI settings, and the source urges users to check every app they use daily.

Is running a local model the real answer?

For users who object mainly to big tech holding their data, open-weight models run entirely on a desktop or laptop via inference engines such as KoboldCpp, LM Studio, Ollama or Llamafile — slower and less sophisticated than server-side models, but fully under user control. Purpose-built local tools include the transcription app OATS and Off Grid AI Desktop.

For marketers, the signal is blunt: consumers are being handed, and using, the means to pull their data out of the training pool — and the trade-offs they accept to do it tell you how much friction they will tolerate.