Every AI you use - ChatGPT, Gemini, Copilot, Grok, Perplexity, and yes, my pick, Claude - is made by a company with a privacy policy and a track record. Most people never read a word of it. You should. Not to be paranoid - to be awake. "Free" AI is rarely free; a lot of the time you're paying with your conversations. Here's how to check any of them yourself. Including mine. Especially mine.
Updated July 8, 2026 - always double-check, these policies change fast.
The 5 questions to ask about ANY AI
- Do they train their AI on what I type? (Your chats become part of the machine.)
- Is that ON by default, or do I have to opt in? (Almost always ON - you have to turn it off.)
- Can actual humans read my chats? (Sometimes yes - "human reviewers.")
- Do they sell or share my data? (Look for "we do not sell" - then read the exceptions.)
- What will they hand to a government or in a lawsuit? (The "lawful purposes" question.)
The decoder: what the wording really means
Policies are written to sound harmless. Here's the translation:
- "to improve our models / to train our models" = they use your chats to build the AI.
- "human reviewers / reviewed by our team or service providers" = real people can read what you typed.
- "unless you opt out / you can turn this off" = it's ON right now. Go turn it off.
- "de-identified / anonymized / aggregated" = they take your name off but still use the content.
- "service providers / affiliates / third parties" = other companies that also get your data.
- "we do not sell your data ... except" = read everything after "except."
- "retention / how long we keep it" = 30 days is good; 3 to 5 years is a long memory.
- "for legal, safety, or security reasons" = the carve-out where opting out may not fully apply.
- "lawful purposes / law enforcement requests" = what they'll do when a government asks.
Where to actually look (2 minutes)
- Search "[tool name] privacy policy" and "[tool name] data training opt out."
- In the app: Settings -> Data Controls / Privacy. That's where the training toggle lives.
- Find the AI-specific "privacy hub" or "how your data is used" page - clearer than the fine print.
The honest scorecard (July 2026 - verify, this moves fast)
Short version: almost all of them train on your chats by default and make you go turn it off. Here's each, plus where to shut it off.
Claude (Anthropic) - my pick, same standard
Does NOT sell your data. But consumer plans train on your chats by default (since 2025), and chats flagged for safety review can still be used. Why I trust them anyway: when the Pentagon demanded they allow mass domestic surveillance and fully autonomous weapons, they refused - and took real punishment for it.
Turn it off: Settings > Privacy > "Improve Claude for everyone." Check the policy
ChatGPT (OpenAI)
Trains on consumer chats by default. Business, Enterprise, and Edu accounts aren't trained by default. No evidence they sell your data.
Turn it off: Settings > Data Controls > "Improve the model for everyone." Check the policy
Gemini (Google) - the one I personally avoid
By default, HUMAN reviewers can read a sample of your chats, and Google can keep them up to THREE YEARS even after you delete your activity. Their own advice: don't type anything you wouldn't want a stranger to read.
Turn it off: myaccount.google.com > Data and Privacy > Gemini Apps Activity > off. Check the policy
Copilot (Microsoft)
Personal/consumer Copilot trains on your conversations, voice, and uploaded files by default. Good news: work / Microsoft 365 organization accounts are NOT used for training - so if your Copilot is through a company or Office 365 work account, you're already covered.
Turn it off: in the app, Account > Privacy > the "Training on conversation activity" toggles. Check the policy
Perplexity
Trains on your searches by default. Says it doesn't sell your info - but a 2026 lawsuit ALLEGES it quietly shared data with Meta and Google through hidden trackers. That's an allegation, not proven - but it's exactly why you read the policy AND watch the track record.
Turn it off: Settings > Preferences > AI data retention. Check the policy
Grok (xAI / X) - the most aggressive
Trains on ALL your public X posts by default, plus your Grok chats - and as of January 2026, what you type to Grok is treated like a public post. It now sits inside the combined SpaceX / xAI / X company. An opt-out toggle exists, but experts call the protection incomplete.
Opt-out toggle is in your X/Grok privacy settings. Check the policy
Do this ONE thing today
Whatever AI you actually use: go find the training toggle and decide, on purpose, whether it's on or off. Two minutes. That single habit puts you ahead of almost everyone.
Why I chose Claude
I didn't pick Claude because it's flashy. I picked it on ethics. When the Pentagon demanded Anthropic remove its guardrails and allow mass domestic surveillance and autonomous weapons, they said no and paid for it. That's the backbone I want behind the tool I teach on. And no company here is a saint - which is exactly why I just showed you how to check them all, mine included.
The bigger lesson
Don't take my word, or theirs. The tool matters less than the habit: read the policy, find the setting, decide for yourself. That's not paranoia - that's being awake. And you're not behind for just now learning this. You're right on time.
Get the next one
Plain-English AI, no hype, no countdown clocks. One useful thing at a time.