What you will prepare
Result: A small, non-sensitive preference card and a comparison of how it changes a fictional task.
- Level
- Beginner planning
- Time
- About 10 minutes to prepare and compare a preference card
- Device
- Phone, tablet or computer for reading and notes
- Tools
- Paper or a local note for drafting; an AI text chat is optional for comparing answers. No connected mailbox, calendar or personal-history upload is needed.
Before you begin: A preference card does not change the service’s storage or training settings. Review those separately in the actual product.
Times are planning estimates, not measured reader results. About this October 20, 2026 presentation update.
You can make an AI assistant more useful by sharing a few relevant preferences. Start with the format you want, the task you are doing, and any practical constraints. You do not need to begin by uploading your life story.
The aim is to reduce unnecessary sharing while getting useful help. No personalization technique creates a guarantee of zero privacy risk.
Personalization is not always model training
Giving context in a conversation, saving a preference for later chats, and training a model are different processes. A service may offer some or all of these, with different controls and data policies.
For example, “Use short paragraphs” is an instruction about the answer you want. It does not, by itself, tell you whether the provider will retain that instruction or use eligible interactions to develop its models. Check its documentation.
Build a small preference card
Try a fictional, non-sensitive card like this:
I am learning a new skill. Explain unfamiliar terms, give one worked example, and finish with a short practice exercise. Ask about missing details instead of inventing them. Keep the first answer to five steps.
Test the card on an ordinary task, such as organizing a hobby project. Keep instructions that improve the result and remove ones that add little value.
A preference card can also describe tone, preferred units, available time, or a broad experience level. Include only what matters for the current purpose.
Generalize the situation
Instead of sharing your exact employer, team roster, and internal deadline, practice with “a small team preparing a fictional project update.” Instead of sharing a family member’s identifiable story, ask for a general list of questions you can adapt privately.
Changing names alone may not protect someone if the surrounding facts remain distinctive. For practice, change the whole scenario rather than relying on cosmetic edits.
Keep lasting preferences separate from temporary facts
Ask yourself whether a detail is useful just for this task or appropriate to retain longer. Your preferred explanation style may remain useful. A one-time sensitive problem may not belong in saved personalization at all.
Review any memory or saved-information feature through the product’s actual settings. An assistant saying “I forgot that” is not a substitute for checking the relevant control and documentation.
Google’s Gemini privacy documentation, for example, distinguishes saved information, activity, and connected services. Other products and account types can use different arrangements.
Try a comparison before adding more data
- Ask for help with a fictional task without a preference card.
- Repeat the task using three or four useful preferences.
- Compare clarity, relevance, and the editing you still need to do.
- Add another detail only if you can explain why it should help.
You may discover that a better instruction matters more than more personal information. If extra detail does not improve the result, leave it out.
Be deliberate about connected accounts
Before granting access to a calendar, mailbox, or document service, read what the connection can access and do. Consider whether you can complete the task using a smaller, non-sensitive input instead.
For client or employer information, follow an approved process. Our client-data checklist explains the questions to resolve before using a tool for confidential work.
Useful personalization is a selective habit: share a clear purpose, test the result, and revisit what the service retains.
Reviewed October 2, 2026.
Report a correction (this guide’s address is included automatically).
AI disclosure: This article was drafted and revised using AI. Read our full disclaimer.