Grokbot memory not working? Here's what's actually happening

What it actually stores

Grok works from three different pools, and people conflate them:

• The current chat. Everything you've said in this thread. Rich, precise — and gone the moment you start a new one.
• Stored facts. A selective list it keeps across chats. Useful, but short, and it decides what earns a slot.
• Your public activity. If you're signed into X, it can read the version of you that you publish.

Notice what isn't in that list: the reasoning behind a decision you made six weeks ago, the constraint that shapes your whole project, the thing you're deliberately not doing and why. That's the material that makes an assistant feel like it knows you, and it lives in conversations that no longer exist.

Things worth trying first

1. Ask it point blank what it has stored about you. The list is usually thinner than expected.
2. Check memory settings in the app — memory can be off, or scoped differently than you assumed.
3. Keep one long thread per project instead of starting fresh daily. Chat context is the strongest memory it has.
4. Say "remember this" explicitly for the handful of facts that must survive.

Those help. None of them fix it, because none of them change the size of the container.

Why this gets worse as Grokbot gets better

Grokbot is good enough now that people are running more of their week through it — research, drafting, decisions, code, agents doing real jobs. The more that's true, the more of your thinking ends up accumulating inside a place you don't own, in a form you can't take anywhere. Cancel, switch, or lose a model, and the version of you it built goes with it.

The durable fix

Stop asking the assistant to be the archive. Keep your own record of what you've decided, what you're building and how you think — then connect it, so Grok reads from it at the start of any conversation and answers like it already knows you.

Step-by-step: how to give Grokbot your memory.