Published September 11, 2026
Custom Instructions and Memory: How to Set Up ChatGPT and Gemini to Actually Know You
Repeating the same background info about yourself in every new conversation? That's exactly what memory and custom instructions were built to end. Here's how to set them up correctly the first time.
On This Page
If you find yourself typing things like "I'm a beginner developer" or "I prefer short answers" at the start of every conversation, you're manually re-explaining yourself every time to a tool that's fully capable of remembering it automatically. Custom instructions and memory features exist precisely to fix this.
The Difference Between Custom Instructions and Memory
Custom instructions are fixed text you write once in settings, automatically applied to every new conversation (like your field of work, preferred reply style, or language). Memory is different: it's information the model extracts automatically from your conversations over time (like the fact you're working on a certain project, or that you're sensitive to a particular writing style), without you ever stating it explicitly each time.
What Belongs in Your Custom Instructions
About me: I work in [your field], my experience level in [topic] is [beginner/intermediate/advanced]. How would you like responses? Keep answers [concise/detailed], in a [formal/casual] tone, and avoid [what you don't want, e.g. over-explaining, emojis, long intros].
Fill each section with information that's genuinely stable and won't change weekly. Avoid putting temporary important details (like a project ending next month) here — those fit memory or the conversation itself better.
How to Control Memory Instead of Leaving It Random
Don't assume the model will infer everything on its own. When you mention something you want remembered going forward, ask for it explicitly instead of assuming it'll stick automatically. Periodically review your saved memories in settings and delete anything outdated — stale, inaccurate memory produces answers built on a wrong assumption about you.
Remember this for future conversations: [the exact information you want saved].
Good memory isn't about how much the model remembers — it's about how accurate what it remembers is.
A Note on Privacy
Avoid saving very sensitive information (precise financial data, private medical details) into persistent memory, even if the feature seems secure. Only share what actually improves the quality of responses, not every detail of your life.
Conclusion
Spend 5 minutes today setting up custom instructions, and build the habit of correcting memory whenever you notice it's wrong. This small difference saves you dozens of repeated introductory sentences over months of use.