How to remember what you decided in ChatGPT.

ChatGPT’s memory holds things like: prefers concise answers. Works in operations. Has three kids. Mentioned a move to Denver in March. That’s a profile, built up across every conversation you’ve ever had, and it’s genuinely useful for what it’s for — making the next answer fit you without you repeating yourself.

A decision is a different shape entirely: we chose the smaller vendor, because price mattered more than the extra features, which ruled out the two we’d shortlisted before. That’s not a fact about you. It’s the outcome of a specific comparison, with reasoning attached and other options explicitly set aside. Calling both of these “memory” hides that they’re built for different jobs.

Ask ChatGPT to remember a decision, and it can, in the same list as everything else it knows about you — but it sits there as one more fact, not as a decision with its own shape. There’s nowhere for the options you ruled out. Nowhere for the reasoning, beyond whatever fits in the summary. And when you change your mind three months later, the update usually just replaces the old note — so the fact that you once decided the opposite, and why, is the part that quietly disappears.

That last part matters more than it sounds. The reasoning behind a decision is usually more valuable later than the decision itself — it’s what tells you whether anything has actually changed enough to reopen it.

Five things, and most decisions only need this much: what you chose. The reason you chose it. What it replaced, if anything, kept rather than deleted. When it was decided, and by whom — you, or whichever assistant wrote it down. And, separately, any related question you haven’t settled yet, so it doesn’t get mistaken for a decision that was actually made.

Say you spent a few evenings working out how to replace a car. ChatGPT compared buying against leasing — you keep cars a long time, and every lease you priced had a mileage cap your commute would blow through, so: buy, not lease. Then hybrid against electric — no home charger made daily electric charging a hassle, so hybrid, for now. Then two models — a three-year-old hybrid SUV was about $26,000 against $34,000 new, and it fit three child seats across the back where the other one you liked didn’t. One thing stayed open: finance through the dealer, or wait on the credit union’s rate.

Kept as a decision record rather than a chat: three decisions, each with its reason attached, sit there permanently. Six months from now, if someone asks why you didn’t go electric, the answer is right there — not reconstructed, just read. The one open question is still marked open until the credit union actually answers, not silently forgotten. See the full record this example is drawn from.

You don’t maintain this separately from the conversation. Once you land on something, you say so — “save the decision: buying, not leasing, and why.” — and it’s kept as its own record rather than folded into your profile. If you reverse it later, you say that too, and the earlier decision stays visible, marked reversed, instead of vanishing. This works the same for Claude, and for Gemini through Gemini Spark, if that’s where the conversation happens to be. Here is how Giraffe works.