I work with these tools every day, and until I prepared this series I didn’t know there’s a switch that decides whether my chats are used to train the model. I’ve never flipped it. At three of the five it’s on by default. That’s the most honest place to start this part, because if I didn’t know, the reader this series is for doesn’t either.
This is part 4 of 6. Part 3 covered everything an assistant remembers about you. This part is about where that goes: who reads it, how long it stays, and where the switch is.
Four questions per tool
The first question is whether your chats are used to train the model, and what the default is. Gemini says it outright: “Keep Activity” is on by default if you’re eighteen or over, and saved activity is used to train the models. So does Perplexity: “AI Data Retention” is on by default, on free and paid alike. ChatGPT says your chats may be used “unless you opt out”, and its pricing page lists “opt-out available” for every personal plan. Claude puts “none by default” on its pricing page and makes you choose when you create an account; what the switch in that screen defaults to isn’t published anywhere. Grok lists a switch called “Improve the Model” on grok.com, without saying how it’s set.
Important for the question this series asks: none of the five distinguishes between free and paid here. Someone paying for ChatGPT Plus or Perplexity Pro sits on exactly the same setting as someone on free. The difference sits with the business plans, and those come in part 5.
The second question is what a temporary chat does. All five have one: “Temporary Chat” in ChatGPT, incognito in Claude, temporary in Gemini, “Private Chat” in Grok, incognito in Perplexity. Those chats aren’t used for training and don’t create memories. They just don’t vanish right away: ChatGPT, Claude and Grok keep them for up to thirty days, Gemini for seventy-two hours. Perplexity says incognito searches are never stored.
The third question is who reads along. Google is the clearest: a subset of chats is reviewed by human reviewers, and Google itself asks you not to enter confidential information. Reviewed chats are disconnected from your account and kept for three years. Grok says a limited number of authorised staff may review conversations, including to improve the model. Claude only reviews chats flagged by a safety filter, and then keeps them for up to two years. ChatGPT mentions review only for temporary chats, to monitor for abuse. Perplexity says nothing about it.
The fourth question is where it sits. Grok says it plainest: everything is processed in the United States. Claude names servers in the US, ChatGPT the US as one of its locations. Google and Perplexity name no location. For anyone living in Europe, the responsible party at OpenAI, Anthropic and Google is an Irish company; at Grok it stays the American company, with a representative in Dublin. EU storage exists at ChatGPT for business customers only.
The table
Checked on 6 September 2026, from the makers’ privacy and help pages. The menu paths come from those pages; I haven’t walked each one in the app.
| Trains on your chats | Where the switch is | Temporary chat | Deleted after | |
|---|---|---|---|---|
| ChatGPT | Yes, unless you opt out; free and paid alike | Settings, Data Controls, “Improve the model for everyone” | Not for training, kept thirty days | Thirty days |
| Claude | “None by default” per the pricing page; you choose at sign-up | Settings, Privacy, “Help improve our AI models” | Incognito, not for training, thirty days | Thirty days; flagged chats up to two years |
| Gemini | Yes, on by default from age eighteen | Gemini Apps Activity, “Keep Activity” | Not for training, seventy-two hours | Eighteen months automatically; human-reviewed three years |
| Grok | Switch exists, default not published | Settings, Data, “Improve the Model”; on X under Privacy and safety, Data sharing | “Private Chat”, thirty days | Thirty days |
| Perplexity | Yes, on by default, Pro and Max included | Settings, Preferences, “AI Data Retention” | Incognito, never stored | Account thirty days; individual threads no automatic term |
What I see myself
I had never switched anything off at any of the five, and I didn’t know I could. That’s the whole note for this part, and I’m leaving it as it is because it probably applies to most readers.
What I do see after reading those pages is that I’ve connected Gemini to my Gmail and Drive, with the activity setting on, because the connection doesn’t work otherwise. That was the convenience from part 3. It’s also exactly the combination Google itself attaches a warning to: don’t enter confidential information.
Caveat: the switch isn’t the whole story
It’s tempting to flip the switch at all five after this part and call it done. Three things remain. A chat flagged by a safety filter still gets reviewed and kept at Claude, switch or no switch. At ChatGPT, a thumbs up or down releases the whole chat for training regardless, switch or no switch. And the switch is about training, not storage: what you type today sits in your history at four of the five until you delete it, and for thirty days after that.
The second caveat is the direction this is moving in. Everything that made the assistant more useful in part 3, memory, projects, a connection to your mail, raises what’s at stake in this part. The personal assistant and the privacy question are the same switch, seen from two sides.
What to do tomorrow
Open the settings of the assistant you use most and find the line from the table. Look at how it’s set. Flip it if you want to, but look either way; that’s the step I skipped myself all along.
Frequently asked questions
Is paid safer than free?
No. At all five, the training setting on the paid personal plans is the same as on free. Different rules only apply on the business plans, and part 5 is about those.
If I switch it off, is my old chat safe too?
Usually not retroactively. The switch is about what happens from that moment on. If you want old chats gone, you have to delete them yourself, and at most tools that still takes thirty days.
Sources: ChatGPT data controls, how ChatGPT uses your data, ChatGPT temporary chats, ChatGPT retention, OpenAI business data, Claude model training, Claude privacy setting, Claude retention, Gemini Apps Privacy Hub, Gemini activity, Grok FAQ, Grok privacy policy, Grok on X, Perplexity data collection, Perplexity account settings. Checked 6 September 2026.