You've been using the same AI assistant for months. It calls you by name. It knows your job. It formats things the way you like. It feels like it really gets you.
Then you ask it to list everything it knows about you, and the response is... underwhelming. Maybe it has your name and a few preferences, but it's missing your biggest project, your preferred tools, half the instructions you've given it, and that correction you made last week about not using semicolons in your JavaScript.
This is more common than you'd think.
The illusion of understanding
AI assistants are very good at seeming like they know you, even when their actual stored memory is thin. There are two reasons for this.
First, they're good at picking up context clues from the current conversation. If you mention you're working in Python, the AI adjusts on the fly. That feels like memory, but it's just in-session adaptation. Close the tab and it's gone.
Second, the stored memory they do have goes a long way. If all your AI knows is your name and your job title, it can already frame responses in a way that feels personalized. But "personalized" and "actually understands your preferences" are very different things.
What your AI probably gets wrong
Based on what people find when they export their AI memories (we see a lot of these through ImportMemory), here are the most common gaps:
Outdated information. Old job titles, finished projects, tools you've stopped using. AI memory doesn't have a built-in expiration date. If you told ChatGPT you were learning Rust six months ago and you've since moved on, it still thinks you're learning Rust.
Missing instructions. You told it to keep answers under 200 words. You told it to always use metric units. You told it not to start responses with "Absolutely!" But when you check the memory, half of those corrections never got saved. They worked in that one conversation and then disappeared.
Wrong inferences. You asked about React Native once for a friend, and now your AI thinks you're a mobile developer. Inferred preferences are powerful but blunt. The AI doesn't always know the difference between a one-off question and a genuine interest.
Gaps in personal context. Your AI might know your name and job but not that you're in a different timezone, that you have a team of five, or that you're preparing for a board presentation next month. The things that would actually make it useful on a daily basis are often the things it hasn't stored.
How to check for yourself
This takes about 30 seconds. Open your AI tool and ask:
Compare the response to what you'd expect. I'd bet there's at least one thing that's wrong, a few things that are outdated, and several important preferences that are missing entirely.
This is your AI profile. It's what shapes every response you get. And for most people, it's a pretty rough sketch.
How to fix it (in 2 minutes)
Once you've seen your memory, you have two options.
Clean it up in place. Delete wrong entries, correct outdated information, add the missing preferences. Every tool lets you do this, though the interface varies. For ChatGPT it's Settings > Personalization. For Claude it's Settings > Privacy > Memory preferences. For Gemini it's Settings and Help > Instructions for Gemini.
Export, fix, and re-import. If your memory needs a lot of work (or if you want to switch tools anyway), it's often faster to export everything, edit it as a text document, and then import the cleaned-up version into whichever tool you want. Our migration guides walk you through this step by step.
Think of it like cleaning out your email filters. You set them up ages ago, some are outdated, some are wrong, but you never look at them because everything "mostly works." Taking 2 minutes to audit and fix your AI memory is the same kind of low-effort, high-impact cleanup.
The real issue
The deeper problem is that AI tools don't make this visible. Your memory should be something you see and manage regularly, not something buried in a settings page you visit once. Until the platforms fix that, you have to be proactive about checking it yourself.
The payoff is real though. A clean, accurate memory profile makes your AI dramatically more useful. It's the difference between an assistant that kind of knows you and one that actually saves you time.
Go run the audit. It takes 30 seconds, and you'll probably be glad you did.