Jump to a section
The short answer

Use custom instructions for how you want answers (tone, length, format) and memory for stable facts that genuinely help (your household, your hobbies, your goals). Keep anything sensitive out of both, review what's been remembered every few months, and turn memory off for chats you'd rather it forgot.

Two different things

Most assistants now offer two ways to stop you repeating yourself, and they're worth keeping separate in your head.

Custom instructions compared with memory
FeatureCustom instructionsMemory
What it isStanding rules you write onceDetails the assistant keeps from your chats or that you ask it to save
Best forHow you want answers: tone, length, formatFacts about you that rarely change
Who decides what goes inYou, directlyYou, or the assistant as you chat

Connecting AI to your email, calendar and files: do it safely

How to schedule recurring prompts in Gemini

Review the memory the service actually holds

ChatGPT's memory settings are in Settings, Personalization, Memory. OpenAI says deleting a chat does not necessarily remove a separate saved memory; review the remembered item as well as its sources.

Anthropic's current Claude help uses Settings, Your Claude, with Settings, Memory in an older layout. Settings differ across rollouts, so use the official help linked here when your screen looks different.

In Gemini, start with instructions for your UK preferences

UK readers can save preferences such as UK spelling in Gemini instructions. Google's current UK-English help uses Personal context, then Your instructions for Gemini. On a computer, start with Settings. In the mobile app, open Menu and your profile picture or initial. Add your instructions and choose Submit. Follow the help for your device if its menu differs.

Those instructions are separate from asking Gemini to save facts during a chat. Google currently limits that chat-memory feature to US personal accounts over 18 with a Google AI plan and Memory on.

Google's separate Memory summary is not currently available in the UK. Use the features your account offers; you do not need a new plan for the UK-spelling example below.

What to put in custom instructions

Write rules about the answers, not your life story. A good set is short enough to read in 30 seconds.

An example set of custom instructions
I'm in the UK, so use UK spelling, pounds and UK examples unless I say otherwise. Give me the answer first, then the detail. Keep answers short unless I ask for more. Use plain English and avoid jargon. If you're not sure about something, say so rather than guessing. When I ask for options, give me three at most, with your recommendation.

How to make ChatGPT, Gemini and Claude answer for your country

What's worth remembering, and what isn't

Memory is most useful for stable, low-risk facts that save you typing. It's least appropriate for anything you'd regret being stored.

  • Worth it: how many people are in your household, dietary preferences, hobbies, the kind of car you drive, projects you're working on.
  • Keep out: passwords, account numbers, ID numbers, health details you'd rather keep private, and other people's personal information.
  • In Claude, decide on purpose whether 'Include sensitive topics in memory' should be on. It covers things like health conditions or religious beliefs, and you may prefer to keep those out.
  • Every few months, open the memory list and delete anything out of date or that you're not comfortable with.

AI chatbots and your privacy: what they keep, who sees it, and how to take control

When to switch it off

For one-off sensitive chats, use a temporary or private chat if your app has one, or pause memory first. And if answers start feeling oddly specific or stuck in a rut, check what's been remembered: old memories can steer new answers in ways you don't expect.

How to stop AI chatbots using your chats for training

ChatGPT Projects and Gemini Gems: set them up for the jobs you do most

Check memory with one harmless preference

A preference such as 'use UK spelling' makes a better first example than health, money or identity details. Training controls are separate from memory controls.

  1. Open the memory or personal-context settings for the assistant you use.
  2. Review the saved items and any available sources. Correct or delete anything inaccurate or too personal.
  3. Add one harmless preference, start a new chat and check whether it helps.
  4. If you want to remove information, check both the memory entry and the original chat or file. Turning memory off alone is a different action.

Sources

These sources support the guide. Tool settings, prices and policies change, so dated source boxes show when those details were checked.

AI helps with research and drafting. Dan checks the sources and is responsible for what is published.

How we check