How to Edit What AI Remembers About You (or Your Company)
AI tools build up memory about you over time — and sooner or later, some of it goes stale or just plain wrong
AI tools build up memory about you over time — and sooner or later, some of it goes stale or just plain wrong. A finished project it still thinks is active. An old preference you've changed. A company detail from a year ago. The good news: editing is getting easier, and some platforms now let you update memory rather than just wipe it. The catch: how you do it differs in every tool, and none of it carries across the AI tools you use. Here's how editing works across the major platforms — and why managing memory tool-by-tool isn't the same as having one editable source of truth.
How do you edit what AI remembers about you?
It depends on the platform. Most let you view what's stored and delete it; some now let you update or correct entries rather than only wipe them. What none of them offer is editing that carries across tools — every platform is its own separate settings panel. So "editing your AI memory" is really several different chores, one per tool, with no shared source of truth.
We'll get to why that fragmentation matters. First, the practical steps for each major platform.
How to edit ChatGPT's memory
ChatGPT gives you the most direct controls of the three. In Settings → Personalization → Manage memories you can see saved memories, delete individual entries, clear them all, or toggle memory off. Beyond the settings panel, OpenAI says you can also ask ChatGPT directly to update, combine, or remove what it remembers — so correcting a fact in conversation is supported, not just deletion.
Two things worth knowing. ChatGPT's memory has shifted toward automatic background synthesis, so alongside the saved-memories list it also draws on your past chats (a separate "reference chat history" control) and curates memory on its own over time — which means not everything it "knows" appears as a tidy editable entry. And memory is a separate control from model training; managing one doesn't change the other. Exact menus and behavior change often, so verify against your current version.
How to edit Claude's memory
Claude offers more genuine editing, including correction in conversation. You can manage stored memory under Settings → Capabilities → View and edit your memory, and delete or modify specific items under Settings → Memory. Claude also separates memory by project, so each project has its own isolated memory space you can manage independently.
The most useful difference: you can correct Claude mid-conversation. Tell it "actually, I've moved off that framework" or "forget what I said about the old timeline," and it updates, and Claude is generally transparent about when it's using stored context. That's genuine in-place correction — though it still leans on you to catch stale entries as they come up, and exact menu paths vary by plan and rollout.
How to edit Gemini's memory
Gemini leans on memory of your past chats to personalize responses for eligible users, rather than a prominent list of saved facts you edit one by one. In Google's framing, you can review and manage what it retains and turn personalization off, but the editing model is less about revising discrete entries and more about controlling whether it draws on your history. The pattern across all three platforms: deletion and on/off control are well supported; granular, portable editing is not.
Why deleting isn't the same as editing
Even where platforms now let you update entries, deletion is still the blunt fallback — and your context isn't made of independent facts. Fixing one outdated detail by deleting its entry often takes useful context with it, like tossing a whole photo album because one picture is out of date.
But the bigger problem isn't delete-versus-update within one tool. It's that editing is platform-specific and trapped. Each tool has its own settings, its own way of correcting memory, its own picture of you. Fix something in ChatGPT and Claude still has the old version. Update Claude and Gemini never hears about it. Your context isn't just occasionally awkward to edit; it's impossible to edit consistently, because there's no single place where a correction takes hold everywhere at once.
How to actually edit your context (not just delete it)
Treat your context as a living document you revise, in one place, that every tool reads from.
This is the model a context layer uses, and it's where the editing problem actually gets solved. Instead of hunting through each platform's settings to delete entries, your context lives in one owned place you can revise directly. In Unabyss, that includes editing conversationally through the context chat — you tell it what changed ("this project wrapped," "we've repositioned around X," "I've moved teams") and it updates the structured context, the way you'd revise a note rather than shred and rewrite it. You're correcting in place, not deleting and hoping.
And because that one context layer serves all your AI tools through MCP, a single edit propagates everywhere. Update a fact once and Claude, ChatGPT, and anything else you connect read the corrected version — no per-platform cleanup, no three copies drifting out of sync. That's the difference between managing memory (deleting entries, tool by tool) and editing context (revising the truth, once, for everything).
So: use each platform's delete controls to clear out what's clearly stale — but if you're tired of editing-by-deletion across five tools, the fix is owning a context layer you can actually revise.
→ How context delivery works across tools: How to Deliver Personal Context to AI Tools
→ Owning and controlling your context: How to Give AI Access to Your Data Without Giving It Away
→ Edit your context in one place with Unabyss →
