Every new AI chat starts from nothing. You explain your business, your tone, your three jobs, your goals — again — and by Thursday you're bored of typing the same paragraph into ChatGPT or Claude. Memory features exist to fix exactly this. Most people just never turn them on properly, so they never notice the difference — and end up quietly convinced that "AI doesn't really get me" when the truth is they never gave it the chance to remember anything in the first place.
Why AI Keeps "Forgetting" Everything About You
By default, a chat window is a blank context. The model only knows what's in that conversation, so the moment you open a new tab, it has no idea you're a teacher, that you run a small e-commerce store on the side, or that you always want bullet points instead of essays. Some tools now ship with memory switched off, others switch it on but scope it narrowly to one chat thread. Either way, the fix isn't a better prompt each time — it's a one-off setup that sticks.
Turn On Native Memory the Right Way
In ChatGPT, go to Settings → Personalization → Memory and make sure it's enabled, then fill in Custom Instructions properly: one box for what it should know about you, one box for how you want it to respond. In Claude, memory and project-level context live under your account settings and inside Projects, where anything you pin stays available to every chat in that project. The setup takes under five minutes and it's the single highest-leverage thing you can do before writing another prompt.
Write a Memory Brief Instead of Hoping It Picks Things Up
Don't rely on the AI to infer who you are from scattered messages. Paste a short, deliberate brief once, and let memory carry it forward from there.
Adjust the details, but keep the structure: identity, tone, output format, and how you want uncertainty handled. That last line alone will save you from a lot of confidently wrong answers.
Give Each "Job" Its Own Space
If you juggle more than one thing — teaching and a side business, say — one shared memory gets confused fast. Use separate Projects or custom GPTs for each area of your life, each with its own brief. This is genuinely the first thing I set up when I build an AI workflow or a website for a client through Tafuma Web Studio — it's a ten-minute job that saves hours of re-explaining every single week.
What to Keep Out of AI Memory
Memory is convenient, not a vault. Don't store passwords, bank details, National Insurance or passport numbers, or health information about specific people — yours or anyone else's. Treat it the same way you'd treat notes left on a shared desk: useful context, never anything you'd mind someone else reading. If you wouldn't say it out loud in an open-plan office, it doesn't belong in a memory setting either.
The Five-Minute Weekly Memory Check
Once a week, open your memory settings and skim what's stored. Delete anything that's gone stale — an old job title, a project that's finished, a preference you've changed your mind about. Add one new fact if something's shifted. Five minutes, once a week, and the AI actually stays useful instead of quietly working off last month's version of you.
None of this is complicated. It's just a setup step almost nobody bothers with, which is exactly why it's worth doing today.