You know the feeling of explaining your workflow to an AI for the fiftieth time? Claude memory is here to end that. Anthropic’s update lets Claude actually learn about you as you chat, instead of waiting for you to hand-write custom instructions. The result: an assistant that stops asking “what do you do again?” every single session.
Why “hi, it’s me again” is the worst AI experience
Chatbots are weirdly forgetful. You tell ChatGPT or Claude how you like your emails phrased, what projects you’re juggling, and which tools you use. Then you open a new conversation the next day and it greets you like a stranger at a party who’s already forgotten your name.
That’s not just annoying. It wastes real time. Every re-explanation costs you a few minutes, and if you use AI heavily, those minutes add up to hours across a month. Static custom instructions were the old fix, but they only work if you actually maintain them. Most people don’t. They update them twice a year and wonder why the AI still gets everything wrong.
The better fix is memory that builds itself. And that’s exactly what Anthropic rolled out in late August 2026.
What Claude memory actually does (and where it lives)
Claude memory works in the background while you talk. Mention that you prefer concise replies, that you run a small agency, or that you hate jargon, and Claude can fold those details into what it remembers. No form to fill out. No settings to configure up front.
The useful part is that you control the whole thing. Head to Settings > Memory in Claude, and you’ll find everything it knows organized by topic. Each entry shows what Claude recorded, and you can edit or delete any single one. Don’t like that it saved your lunch order preferences? One click and it’s gone.
Memory across Chat and Cowork
Here’s the part that impressed me: memory follows you between products. Tell Claude something in a regular chat, and it carries into Claude Cowork projects too. That’s a big deal if you use Cowork for deeper work, because previously, context lived and died inside individual conversations.
Claude can also update its memory while you’re mid-conversation. It doesn’t wait for the end of a session to save things. So if you correct it mid-task, that correction can stick for next time.
How to use Claude memory (step by step)
Claude memory is on by default, so there’s no switch to hunt for. The real work is making sure it remembers the right things and forgets the wrong ones.
Turn it on and see what Claude already knows
Open Claude, then go to Settings > Memory. You’ll see a list of topics Claude has recorded. Scroll through it. This is the fastest way to discover what it’s been quietly learning about you.
Most people find a few surprises here. Some entries are spot-on. Others are confidently wrong. That’s fine, because the next step fixes them.
Edit or delete a single memory
Click any memory topic and you can edit the text or delete the entry entirely. Do this for anything outdated or flat-out incorrect. That finished project from three months ago? Delete it. The preference you changed last week? Update it.
The goal is a memory file that reflects who you are now, not who you were in January.
5 prompts that put Claude memory to work
Knowing Claude has memory is one thing. Using it well is another. These five prompts turn the feature from a gimmick into a daily time-saver.
- Find out what Claude already knows. “Based only on your saved memory about me, tell me the 5 most useful things you know about how I work, what I prefer, and what matters to me. Don’t use information from this conversation unless it’s already in memory.” This is better than a generic “what do you remember about me?” because it forces Claude to separate genuinely useful context from random trivia. It also surfaces mistakes while they’re still cheap to fix.
- Stop repeating yourself. “Look at what you remember about me and identify 3 pieces of context I probably shouldn’t need to explain to you again. For each one, show me an example of how you could use that information in a future conversation.” If Claude knows how you like emails written, you shouldn’t have to restate your tone preferences on every request. This prompt makes the payout explicit.
- Put memory to work on a new project. “I’m starting a new project. Before I give you any other instructions, use what you already remember about my preferences and the way I work to suggest how we should approach it. Separate what you’re basing on memory from anything you’re assuming.” This tests whether memory actually improves the answers, or just sits there looking pretty. When Claude separates memory from assumption, you can see exactly what’s driving its advice.
- Let Claude identify the gaps. “Based on the kinds of things I ask you to help with, what are 5 things about my preferences, routines, or goals that would make you more useful to me if you remembered them? Ask me one question at a time to fill in the gaps.” You’d be surprised what Claude knows it doesn’t know. Maybe it has your job title but not your formatting preferences. This prompt turns memory-building into a two-minute Q&A instead of a guessing game.
- Run a memory audit. “Audit what you currently remember about me. Put each memory into one of four categories: useful, outdated, unclear, or unnecessary. Explain briefly why, and tell me which memories you think I should review or delete.” Memory decays. Projects end, preferences shift, people change. A monthly audit with this prompt keeps your memory file from becoming a museum of outdated facts.
What Claude won’t remember (privacy, explained)
Here’s the part that spooks people, and it’s worth being clear about it. Anthropic says Claude won’t save certain sensitive info by default, including details about your health, religious beliefs, and political views. You can choose to allow sensitive topics through the settings, but the default is no.
That’s a sane default. It means the AI isn’t quietly building a dossier on your medical history because you mentioned a doctor’s appointment in passing. If you deliberately want Claude to remember something sensitive, you opt in. Otherwise, it stays out.
Is Claude memory better than ChatGPT’s?
Both ChatGPT and Claude have shipped memory features, and they take different approaches. ChatGPT’s memory is also automatic, but Claude’s is more transparent and easier to audit, since everything lives in one organized settings panel you can edit line by line.
Claude’s approach also wins on control. You see exactly what it recorded, and you can delete individual entries rather than nuking everything. That visibility is worth a lot when you’re trusting an AI with how you work.
The takeaway
Claude memory is a small feature with an outsized payoff. Set it up once, clean it occasionally, and you’ll stop re-introducing yourself to your own assistant. If you haven’t checked what Claude already knows about you, do that first. You might be surprised, and you’ll definitely save yourself some typing.
For more ways to squeeze value out of Claude, check out our roundup of seven underused Claude features, learn how to double your Claude Pro usage without upgrading, and get familiar with persistent AI coworkers if you want to take this whole idea further.
