Does ChatGPT, Claude, Gemini, or Grok Train on Your Data?
Paying $20 a month does not make your chats private. Here is what each of the four big AI tools does with your conversations, taken from their own policy pages. Plus the one setting that changes it.
Last updated: September 2026
Quick Answer
All four big AI tools can use your chats to train their models. Paying for a personal plan does not stop it. The line that matters is personal plan versus business plan. Business plans leave your chats out of training by default. On a personal plan you flip the setting yourself, and it takes about two minutes.
You pay $20 a month for ChatGPT. That feels like it ought to buy you some privacy.
It buys you speed and features. On its own it does not keep your conversations out of the company’s training data. Almost every owner we talk to assumes the opposite. Here is what all four tools do, and how to change it.
The Short Answer: Free Versus Paid Is the Wrong Question
The question that matters is personal plan versus business plan.
Your $20 ChatGPT Plus account is a personal account. So is Claude Pro. So is the paid Gemini plan. Paying moves you up a tier of speed and features. It does not move you into a different privacy bucket.
Business plans are the different bucket. On a team or business account, these companies keep your chats out of model training by default. That is the real reason to put a company on one, and it usually matters more than the per-seat price.
| Tool | On a personal plan (free or paid) | On a business or team plan |
|---|---|---|
| ChatGPT | Trains on your chats unless you turn it off. OpenAI says ChatGPT “improves by further training on the conversations people have with it, unless you opt out.” | No training by default. OpenAI says it does not train on inputs or outputs from business users, including Team, Enterprise, and the API. |
| Claude | Depends on the choice you made. Anthropic asked personal users to pick. Saying yes also lets them hold chats for five years instead of 30 days. | No training. Anthropic leaves Claude for Work, the API, and its cloud versions out of model training. |
| Gemini | Trains on your chats while Keep Activity is on. Google also says a subset of chats get read by human reviewers. | No training. Google says it does not use Workspace customer data to train the models behind Gemini. |
| Grok | Trains by default. SpaceXAI says it may use your content and interactions with Grok, plus its responses, to train its models. | No training. SpaceXAI says it does not use content from business and enterprise customers to improve its models. |
If you cannot remember ever making a choice about this, you have the default. Go look.
ChatGPT: Where the Setting Lives
On a free or Plus account, OpenAI trains on your conversations unless you say no.
To turn it off, open Settings and find Data Controls. There is a switch there for improving the model. Switch it off. OpenAI also has a privacy portal with a “do not train on my content” option. There is a Temporary Chat mode too, which keeps one conversation out of training.
Team and Enterprise accounts are the opposite. Training is off by default and you get one admin view of who has access.
Claude: The One That Changed
Anthropic used to leave personal chats out of training entirely. That changed, and now Free, Pro, and Max users make the call themselves.
The setting sits under Privacy in your settings. Two things ride on it. Allowing training also stretches how long Anthropic keeps your chats, from 30 days out to five years. Declining keeps you at 30 days.
Here is the part most people miss. Anthropic says that if you delete a chat from your history, it will not be used to train future models. Deleted chats leave your view right away and clear their systems within 30 days. So the five-year window is not a five-year sentence.
Team and Enterprise accounts stay out of training. So does the API, and the versions that run inside Amazon and Google cloud services.
Gemini: Two Very Different Google Accounts
Gemini on a personal Google account trains on your chats while the Keep Activity setting is on. The control lives in your Google account activity settings, not inside Gemini itself.
Google is also unusually direct about people reading chats. Its own help pages say a subset of conversations get reviewed by human reviewers to help improve the service.
Gemini inside Google Workspace is a different animal. Google states that it does not use Workspace customer data to train or improve the models behind Gemini. If your business already pays for Workspace, you may be sitting on the private version and not know it.
Grok: On by Default, and X Posts Count Too
SpaceXAI trains on Grok conversations by default. Its own FAQ says it may use your content and interactions with Grok to train its models. That covers Grok’s replies too.
The switch is called “Improve the model.” In the mobile app it sits under Settings, then Data Controls. On grok.com it sits under Settings, then Data. SpaceXAI says that once you choose not to share, your new conversations will not be used to train its models.
There is a second thing to know if your business posts on X. Grok can also draw on public X posts. That control is separate, over in your X privacy settings under the Grok section.
Grok also has a Private Chat mode, marked with a ghost icon. Those chats stay out of your history and get deleted within 30 days.
What “Turned Off” Actually Means
Switching training off is worth doing. It is not the same as nobody ever seeing a chat, and you should know the difference.
All four companies still hold conversations for a while, often around 30 days. Staff can look at them when they are chasing abuse or answering a legal request. Content their safety systems flag gets held longer. Anthropic, for example, keeps flagged material up to two years and the related safety scores up to seven.
Google says the practical version out loud in its own help pages. “Please don’t enter confidential information that you wouldn’t want a reviewer to see.”
That is normal across the industry. It is also why the list of things you never type in matters more than any switch. Settings lower your risk. The list is what keeps private things out of the machine to begin with.
Check Your Own Settings Right Now (Two Minutes)
Open whichever tool you use and look. Here is where to go.
| Tool | Where the training switch lives |
|---|---|
| ChatGPT | Settings, then Data Controls |
| Claude | Settings, then Privacy |
| Gemini | Your Google account activity settings |
| Grok | Settings, then Data Controls in the app, or Settings, then Data on grok.com |
These menus move around. Trust what you see on your own screen over any article, this one included.
One thing to put your mind at ease. Turning training off does not delete your chat history or break anything. Your saved chats, projects, and files stay right where they are. It will not make the tool worse at answering you. Training helps build future versions for everybody, not a private version for you.
What This Means If You Run a Business
Three moves cover almost every small business, in order of how much they buy you.
- Flip the switch today. Two minutes, no downside, and it applies to whatever account you are already using.
- Move the company onto a business plan. Training off by default, one place to see who has access, and you can shut an account down when somebody leaves. For a team of any size this is the real answer.
- Write down what never gets typed in. Customer card and bank details, Social Security numbers, health information, passwords, and anything under a confidentiality agreement. Our one-page AI policy template gives you that list to hand your team.
None of this is a reason to slow down on AI. It is the opposite. Set the settings right, tell your team the short list, and you can push AI into more of your business. No more quietly wondering whether you should.
I have bought a lot of software in 30 years, and here is what I have learned about privacy settings. Nobody turns them on for you. The default is whatever suits the company selling it, and that is not a scandal. That is just how it works.
Take the two minutes. Then go use the tools.
Want your AI set up right the first time?
Most consultants hand you a list of settings and wish you luck. We look at how your business actually works and set the accounts and rules up properly. Then we build the system your team will use every day.
Frequently Asked Questions
Does paying for ChatGPT make my chats private?
Not by itself. ChatGPT Plus is still a personal account, and OpenAI trains on personal chats unless you turn that setting off. Business plans like Team and Enterprise keep your chats out of training by default. Paying more buys speed and features, not privacy.
Is Claude safer than ChatGPT for business information?
Your plan and your settings matter far more than the brand. All four companies keep business and team accounts out of model training by default. All four let personal users switch it off. Pick the tool that fits your work, then set it up properly.
If I turn training off, can anyone still see my chats?
Possibly, and it is worth knowing. These companies still store conversations for a period, often about 30 days. Staff can review them when chasing abuse or answering a legal request. Google says it plainly in its own help pages. Do not enter confidential information you would not want a reviewer to see.
Do I need a business plan, or is flipping the setting enough?
If it is just you, the setting is usually enough. Once you have employees, a business plan is worth it. Training is off by default and you control the accounts. That control is what lets you shut off access the day somebody leaves.
Could a competitor see something I typed into AI?
Not directly. These tools do not show your conversations to other users. Training data also gets stripped of identifying details first. The real risk is more ordinary. Your information sits in an account you do not control, like an employee’s personal login.
What should I never type into any AI tool?
Customer payment and bank details, Social Security numbers, and health information. Also passwords, door codes, employee pay records, and anything under a confidentiality agreement. Past that list, take names and account numbers out when you can. That habit costs nothing and prevents most trouble.
Take Two Minutes, Then Get Back to Work
Open the tool you use most and find the training switch. ChatGPT keeps it under Data Controls, Claude under Privacy, Gemini in your Google account settings, and Grok under Data Controls. Flip it, and you have handled the part most owners never get to.
Then write down the short list of things nobody types in. Our free one-page AI policy template has that list ready to hand your team.
Want a hand getting your accounts and rules set up properly? Book a free 20-minute discovery call. We will walk through it with you, whether you work with us or not. No script, no sales tactics.
Sources:
1. OpenAI: How Your Data Is Used to Improve Model Performance
3. Anthropic: Updates to Consumer Terms and Privacy Policy
4. Anthropic Privacy Center: How Long Do You Store My Data?
5. Google: Gemini Apps Privacy Hub
