You tell an AI assistant that you prefer short answers. Later, it seems to know your preference. Is it remembering you, reading the current conversation, or simply making a good guess? Those are different possibilities. Understanding the difference helps you decide what to share and how to get a useful answer.
Start with three different questions
- Current conversation: can the assistant use something you wrote earlier in this chat?
- Personalisation: does the product use saved preferences or previous chats in future conversations?
- Data storage: does the provider keep information, and what may it use that information for?
These questions overlap, but one setting does not necessarily control all three. An answer that sounds personal is not proof that a permanent memory exists. Equally, an assistant forgetting a detail is not proof that the provider has deleted it. Check the product controls and its official explanation.
What does this look like in a real product?
Google’s Gemini Privacy Hub describes separate controls for activity, saved information, personalisation and connected apps. It also explains that turning activity off does not stop all processing: conversations are still used to respond and protect the service. Temporary chats are retained for 72 hours. These are Gemini-specific rules; another service or a work account may have different policies.
Source: Gemini Apps Privacy Hub
A useful preference is different from a private detail
For a beginner, a helpful instruction might be “Use plain New Zealand English and explain unfamiliar terms.” You can include that at the start of a chat without providing your full name, address or personal history. Give the assistant the smallest amount of context needed for the task.
Imagine you want help planning a community workshop. “The audience is new to AI and the session lasts 30 minutes” is useful context. A spreadsheet of attendees’ private details usually adds nothing to an outline. Being specific about the task does not mean being personal about everyone involved.
Try a harmless memory experiment
Notice whether the second answer follows the instruction. This tests how the assistant uses the current conversation. It does not test retention or training. You can start a new chat and compare the behaviour, but use the settings page to establish which personalisation features are enabled.
Your five-minute settings check
- Find the official privacy, activity and personalisation controls for your exact app and account.
- Review any saved preferences; correct or remove ones you do not want.
- Check whether deleting a chat also removes saved information, or whether separate controls are required.
- Read what temporary mode changes and what it still retains.
- At work, follow your organisation’s rules before supplying workplace information.
Keep a small reusable introduction for tasks you repeat: your preferred tone, audience and format. You can paste it when needed and update it yourself. That makes your instructions easy to review, even when memory features change. The aim is to understand the controls well enough to make a deliberate choice each time.

