How does AI use my data?
Kahu team·Updated 2 Oct 2026·6 min read
AI uses data in two ways: the huge amount of text and images a model was trained on, and what you type into it. Your conversations are used to answer you, may be stored for a period, and on many personal plans can be used to train future models unless you turn that off. Business plans often have stricter terms.
Two kinds of data
An AI model is first trained on very large amounts of existing material, much of it collected from the internet. That shapes what it knows. Then, every time you use it, your prompts, uploads and its replies become a second kind of data: your conversation history. Most privacy questions for a business are about this second kind.
What providers say they do
Each provider sets its own rules, and they change. Here is what three of them say on their own pages today for their consumer products.
| Google Gemini | Anthropic Claude | Microsoft Copilot | |
|---|---|---|---|
| Used to improve models? | Yes, while Keep Activity is on | Only if you allow it in Model Improvement | Yes, unless you opt out |
| Off switch | Keep Activity setting | Model Improvement setting | Training opt-out in privacy controls |
| Also worth knowing | Activity auto-deletes after 18 months by default, adjustable | Incognito chats are not used to improve Claude | Conversation activity is stored for 18 months by default |
Google says a subset of Gemini chats are read by human reviewers, and that even with Keep Activity off, chats are kept for up to 72 hours to provide the service and keep it safe. Microsoft says some Copilot conversations get automated and human review, that it removes identifying details such as names and email addresses before training, and that people signed in with an organisational account, or under 18, are not used for training. Anthropic says chats flagged for safety review may still be used for safety work.
What this means for a business
- Assume a personal plan may keep and learn from what staff type, until you have checked the setting.
- Prefer business plans with written data terms for work use.
- Agree what may never be pasted in: passwords, financial details and customer personal information.
- Check where data is stored and whether you can export or delete it.
- In New Zealand, remember the Privacy Act applies to personal information you put into AI tools.
Frequently asked questions
Does AI learn from my conversations straight away?
No. Conversations used for training feed into future versions of a model, not the one answering you now. Memory features are different: they save details to personalise later chats, and can usually be switched off.
Can AI tools see my other files and emails?
Only what you upload or connect. If you connect email or a drive, the tool can read what that connection allows, so grant the least access you need.
Where do I find the setting?
In the tool's privacy or data settings. Look for words like activity, model improvement or training.
