Which AI Tools Train on Your Data (2026), and How to Opt Out

Disclosure: LabFin is reader-supported. If you sign up for a tool through links on this page, we may earn a commission at no extra cost to you. Commissions vary between providers and our ratings do not — here’s our full disclosure.

Which AI tools train on your data? On personal plans, most of the assistants we review do by default: ChatGPT, Claude, Gemini, Perplexity, Grok, DeepSeek and GitHub Copilot among them. Each offers an opt-out, but you have to find it.

Business plans, by contrast, mostly skip training by default. This guide lists each tool’s rule, where to switch it off, and how long companies keep your chats, from their own pages read between September 30 and October 2, 2026.

Which AI Tools Train on Your Data: The Short Answer

By default, the personal plans of ChatGPT, Claude, Gemini, Perplexity, Grok, DeepSeek and GitHub Copilot train on your chats or code. Notion AI and Gemini Notebook, by contrast, do not, and Cursor does not once Privacy Mode is on.

Jasper’s Pro terms use your content to develop its models with no general opt-out, and we could not confirm Copy.ai’s terms. Lex, Sudowrite and Mem, however, state that they do not train on your writing. On business plans from OpenAI, Anthropic, Google, GitHub and Microsoft, meanwhile, training is off by default.

Training Defaults on Personal Plans

Tool	Trains by default?	How to stop it
ChatGPT	Yes	Settings, Data controls
Claude	Yes	Turn off Help improve Claude
Gemini	Yes, while Keep Activity is on	Turn off Keep Activity
Perplexity	Yes	Turn off AI data retention; new data only
Grok	Yes	Turn off Improve the model
DeepSeek app	Yes; stored in China	Turn off Improve the model for everyone
GitHub Copilot	Yes, since April 24, 2026	Settings, Privacy
Cursor	Only with Privacy Mode off	Turn on Privacy Mode
Notion AI	No	Nothing to change
Gemini Notebook	No, unless you send feedback	Do not rate answers
Jasper Pro	Yes, under its terms	No general opt-out; EU and UK may object
Mem	No, for general models	Nothing to change
Lex	No	Nothing to change
Sudowrite	No	Nothing to change
Poe	Bot developers may store and train	Keep private material off Poe
LM Studio, Ollama	No, when run locally	Turn off cloud features
Bolt.new	No opt-out documented	Keep client code out
Lovable	Opt-out only on Business	Business plan setting
Replit	Yes, under its policy	No opt-out documented
Gamma	Yes, on personal plans	Settings, Data Controls; Team excluded
Otter.ai	Yes, on de-identified recordings	No opt-out documented
Fireflies.ai	No	Nothing to change
Fathom	Sources disagree	Check current terms
Grammarly	Yes, on individual plans	Turn off Product Improvement and Training
QuillBot	Yes, for account holders	Opt out in settings; Team excluded
Wordtune	Yes, without a written agreement	Enterprise agreement
Zapier	Yes, de-identified data	Opt-out form; Company and Enterprise excluded
Make (Maia)	No	Nothing to change
Descript	Sources differ	Turn off Share data with Descript
Canva AI	Not by default, Canva says	Check privacy settings
Adobe Firefly	No, Adobe says	Nothing to change
Writer	No, reviews report	Nothing to change

From each company’s privacy, help and terms pages, read September 30 to October 2, 2026. We leave Copy.ai out because third-party readings of its terms conflict. Mem, Lex, Sudowrite, Poe, LM Studio and Ollama also come from our reviews of each, as do the rows from Bolt.new to Writer, read July to October 2026. ElevenLabs, Ideogram, ClickUp AI, Surfer SEO, Clearscope, Frase and Beautiful.ai, however, state no rule we could find.

ChatGPT

ChatGPT’s Free, Go, Plus and Pro plans train on your chats by default. You can opt out in Settings, under Data controls, or through OpenAI’s Privacy Portal.

Two details matter, though. Rating a response can still send the whole chat for review. OpenAI also keeps Temporary Chat for up to 30 days, though it does not train on it. Our ChatGPT review covers both.

Claude

Claude’s Free, Pro and Max plans train on your chats unless you turn off Help improve Claude. With training on, Anthropic keeps that data for five years; with it off, for 30 days.

Anthropic does not train on incognito chats even with the setting on, and it keeps rated feedback for five years. Our Claude review explains the settings in full.

Gemini

On personal accounts, Gemini trains on your chats while Keep Activity is on. With it off, Google keeps chats for 72 hours, but some connected Workspace features need it on.

Google can also keep human-reviewed chats for up to three years, even after you delete them. Our Gemini review covers Keep Activity, and Gemini Notebook follows a stricter rule, as our NotebookLM review explains.

Perplexity, Grok and DeepSeek

Perplexity’s Free, Pro and Max plans train by default, as our Perplexity review notes, and turning off AI data retention covers only data from then on. Grok, similarly, trains unless you turn off Improve the model, as our Grok review covers, and xAI deletes its Private Chat within 30 days.

DeepSeek’s app also trains unless you opt out, and it stores data in China for as long as the account exists. For private work, therefore, self-host its open weights instead, as our DeepSeek review explains.

Coding Tools

GitHub Copilot has trained on interaction data from Free, Pro and Pro+ since April 24, 2026, unless you opt out under Privacy. Cursor, by contrast, does not train with Privacy Mode on, which teams get by default. Our GitHub Copilot review and Cursor review cover both.

Windsurf, now Cognition’s Devin Desktop, similarly stops training once you opt out. Our guide to using AI for coding projects covers safe settings for code.

Writing and Workspace Tools

Notion AI does not train on your content by default, as our Notion AI review explains, and its AI providers keep data for 30 days or less. Jasper’s Pro terms, on the other hand, use inputs and outputs to develop its models, with an objection right only in the EU and UK.

Copy.ai’s rules remain unconfirmed: third-party readings disagree. Our Jasper vs Copy.ai comparison covers both.

Which AI Tools Train on Your Data at Work

At work, the answer to which AI tools train on your data is mostly none. ChatGPT Business and Enterprise, Claude Team and Enterprise, Google Workspace with Gemini, GitHub Copilot Business and Microsoft 365 Copilot Business all skip training by default. Perplexity Enterprise also adds zero data retention with its model providers.

APIs follow similar rules. In addition, OpenAI and Anthropic do not train on API data by default. Google, meanwhile, may use free-tier Gemini API content to improve its products but not paid-tier content, guides citing its pricing page report.

Habits That Protect Your Data

Knowing which AI tools train on your data is only the first step. Next, turn off training in each tool you use before you paste anything private. Also avoid rating answers in chats that hold sensitive material, since feedback can send the whole chat for review.

Also use temporary or incognito chats for one-off private questions, and keep client work on a business plan. Our guide to choosing the right AI tool for a task covers privacy as one of the first checks.

How We Research

Each rule comes from the company’s own privacy, help or terms pages, read between September 30 and October 2, 2026, and recorded in our reviews. One example is Anthropic’s privacy center.

Where official pages are silent, we say so; Copy.ai is the main gap. Our about page explains how we work, and our list of the best AI tools sets data rules beside features.

AI Training on Your Data: FAQ

The Verdict

Which AI tools train on your data is mostly a question of defaults. Nearly every personal plan we review trains unless you act, while nearly every business plan does not.

So spend five minutes on each tool’s settings before you trust it with private work. For client material, moreover, use a business plan.