How to Actually Get Real Value Out of Claude's Memory (Without Turning It Into a Junk Drawer)
Claude will now remember you across chats. Six habits that separate the people getting a smarter assistant from the people whose memory profile is quietly slowing them down.
Here's the uncomfortable truth about Claude's memory feature: most people flipped it on in March, forgot about it, and now carry a memory profile that looks like a browser tab graveyard. Half-finished projects. A client who fired them in April. A tone preference they set once, as a joke, that Claude has been earnestly obeying ever since.
Anthropic pushed memory out to every Claude account, free and paid, in March 2026, and it's on by default. The upside is real. You stop re-explaining who you are every chat. The downside nobody warned you about is that a bloated memory profile actually eats into your usage limit and makes Claude's answers worse, not better. More context isn't automatically more signal.
I've spent the last few months treating Claude's memory like a real tool instead of a novelty toggle, and the gap between "genuinely useful" and "junk drawer" comes down to six habits. None of them are technical. All of them take fifteen minutes to set up and will save you hours a month.
1. Open the settings page and actually look at what’s in there
This is the step everyone skips, and it’s the one that changes the most.
To toggle Claude’s memory on or off, navigate to Settings > Memory and turn on “Generate memory from chats.” On the new memory experience, that’s where you live. On the legacy view, it’s Settings > Capabilities > Memory, and “Search and reference chats” and “Generate memory from chat history” are on by default, click “View and manage memory” to see what Claude has stored.
Do it right now, before anything else. I’ll wait.
What you’re going to find is embarrassing. A note that you were “preparing for a Q2 kickoff,” six months ago. A preference for “concise bullet points” that you set during one specific project and have been fighting ever since. The name of a coworker who left the company. Delete all of it. Claude builds memory as a set of individual entries organized into categories, reading, writing, and updating them in real time as you chat rather than on a fixed schedule. Which means every stale entry is quietly showing up in every new conversation you have.
If you’ve never audited yours, that first cleanup is the single highest-ROI thing you’ll do this week.
2. Save preferences, not projects
Here’s the rule that separates a useful memory profile from a bloated one: memory is for things that will still be true in six months. Everything else belongs in a Project.
Preferences that belong in memory:
- Your role and what you actually do all day
- Formatting rules (“no em-dashes,” “always answer in one paragraph, then the caveats,” “give me the code before the explanation”)
- Tools you use (“we’re a Postgres shop, don’t suggest Mongo”)
- Tone (“skip the throat-clearing and get to the point”)
Things that do NOT belong in memory:
- The client you’re working with this quarter
- The manuscript you’re editing
- The launch date for a specific feature
- Your kid’s soccer schedule
Anything scoped to a particular piece of work goes in a Claude Project. Each project has its own separate memory space and dedicated project summary, so the context within each of your projects is focused, relevant, and separate from other projects or non-project chats. That’s the entire point of Projects. Use them.
The mental model: memory is who you are. Projects are what you’re doing right now. Don’t mix them.
3. Use the “remember this:” prompt on purpose
Claude picks up context automatically, but the automatic mode is a magpie. It grabs whatever caught its eye. If there’s something you actually want stored, tell it flat out.
The phrase I use is literally “Remember this:” followed by the fact I want on file. “Remember this: I write in AP style. Serial commas off. No em-dashes.”
That’s it. Claude adds it to memory, and you can go verify in Settings > Memory that it landed. It’s dramatically more reliable than hoping the auto-summarizer will pull the right signal out of a long chat about something else. The more specific and concise your instructions, the more consistently Claude follows them.
The corollary: if Claude has picked up something wrong, and it will, say “Forget that I [x]” or just go delete the entry yourself. Don’t let a wrong memory sit there polluting every future chat.
4. Use Incognito for anything you don’t want remembered
Not every conversation should feed your memory profile. Salary comparisons. A draft of a resignation letter. A weird medical question at 2am. A brainstorm for a company you might start on the side. None of that belongs in the running summary Claude uses to decide who you are.
Incognito chats are available to all Claude users on free, Pro, Max, Team, and Enterprise plans. When starting a chat outside of a project, you’ll see a ghost icon in the upper right of the screen; clicking it enables incognito chats. In this mode, Claude won’t remember your chats, so they won’t be saved to Claude’s memory or your chat history.
That ghost icon is the most underused button in the app. Make it a habit. If a chat is a one-off, sensitive, or off-topic from the person Claude thinks you are, flip incognito on before you type. Your memory profile stays clean, and the assistant stays sharp.
5. Prune monthly like you prune your inbox
The dirty secret nobody wanted to lead with: a fat memory profile isn’t just messy, it’s slower and worse. Every entry is context Claude has to carry into every chat, which means more tokens spent on you before the model even reads your question. If you’ve noticed your usage limit running out faster than it used to, this is why.
Put a recurring calendar reminder on the first of every month. Five minutes, open Settings > Memory, delete anything that no longer applies. Old projects. Preferences from a role you don’t have anymore. Names of people you don’t work with. Facts that were true in April and aren’t now.
If you want the nuclear option, “Reset memory” permanently deletes all memories including project memories, once you click “Reset memory,” it cannot be undone, and upon re-enabling the feature, you’ll start from scratch. That’s overkill for most people. Monthly pruning is the sustainable rhythm.
Think of it like your email inbox. Memory isn’t a vault, it’s a working list. It only earns its keep if you weed it.
6. Audit the memory before big work, and ask Claude what it thinks it knows
The last habit is the one that will save you the most embarrassment.
Before you start any high-stakes chat, a piece of writing you’re going to publish, a strategy doc for your boss, code that’s going into production, do a five-second sanity check. In the chat itself, type: “What do you remember about me?”
Claude will list what it’s currently loading into context. If any of it is wrong, wrong tense, or wrong project, fix it before you start. Ten seconds of your time, and it saves you the experience of Claude confidently referring to your “current role at [company you left last year]” in a document you’re about to send to a client.
The habit compounds. Every time you catch a stale memory, you’re not just fixing today’s chat, you’re fixing every future chat until the next time you would have caught it, which might have been months from now.
The one habit that ties it all together: treat Claude’s memory the way you’d treat a very literal-minded new assistant who takes notes. Everything you say near them, they write down. If you want them to be useful in six months, you have to be intentional about what makes the notebook and what doesn’t. The people getting a genuinely smarter assistant out of this feature aren’t lucky and they aren’t power users. They opened the settings page, deleted the junk, and set a reminder to do it again next month. Start doing that, and Claude stops being a chatbot with amnesia and starts being the assistant it was supposed to be all along.