Most small business owners who try Claude hit the same wall in the first week. You explain your business, your tone of voice, your price list and your awkward supplier, and you get a useful answer. Then you open a new chat the next morning and start from scratch. The Claude memory feature exists to remove that wall. This guide covers what it stores, what it refuses to store, how to keep client work separate, and the checks a UK business should run before letting a whole team rely on it.
What is the Claude memory feature?
According to Anthropic's help centre, Claude saves memory as individual topics while you chat, covering things like your role, ongoing projects and working preferences. Those topics then flow into new chats. They also reach Cowork tasks that run in the cloud, though local Cowork sessions do not use memory.
Say a florist in Leeds once told Claude "we close on Mondays, our minimum delivery order is set in the shop policy, and we never use the word 'cheap'". Claude should respect those details next week without anyone pasting them in again.
Which Claude plans include memory?
The rollout happened in stages. PCWorld reported in March 2026 that memory, which launched for paying users in August 2025, was being extended to free users.
The distinction that catches people out is memory versus past-chat search. Anthropic's help centre lists past-chat search as available only on paid plans: Pro, Max, Team and Enterprise. Memory gives Claude a running record of what matters. Search lets it go back and read an old conversation in detail. If you want Claude to pull up the exact wording of a quote you drafted in June, you need search, not memory.
What does Claude remember, and what does it refuse to store?
You can switch off the sensitive-topic exclusion with an "Include sensitive topics in memory" option, according to Anthropic's help centre. Most businesses have no good reason to turn that on. A care agency or a physiotherapy clinic that handles health information should leave it off.
Memory is good at storing:
- Your brand voice rules and banned words
- Your service list and how you describe it
- Recurring formats, such as how you lay out a quote or a weekly staff update
- The names of your own projects and what stage each is at
Do not use memory as your record of prices, stock levels, contract terms or anything with legal weight. Those change, and a memory that was right in April can be wrong by September.
How do projects keep client work separate?
This setting matters most if you do client work. An accountant, a marketing freelancer or a recruitment consultant should set up one project per client. That way details from a Bristol restaurant client cannot drift into advice for a Cardiff café.
If you already write down how your business runs, those documents make excellent project instructions. Our guide to documenting business processes with AI covers how to turn messy know-how into something Claude can follow.
How do you turn memory on, pause it or reset it?
Anthropic's help centre lists these controls:
- Pause memory. Keeps existing memories but stops Claude using them or creating new ones. Useful while you test whether memory is helping.
- Reset memory. Permanently deletes all memories, project memories included.
- Per-chat off switch. In a new chat, open the plus menu and turn memory off before your first message. You cannot change it afterwards.
- Incognito chat. Use the ghost icon. Nothing is saved to memory or history.
Some accounts still show an older layout, where the setting sits under Settings, then Memory, labelled "Generate memory from chats".
Is the Claude memory feature safe for UK GDPR?
The practical risk is less about Anthropic and more about your own team storing customer details somewhere you do not track. Three habits help:
- Ban customer names, phone numbers and addresses from chats with memory on
- Review the stored memory list monthly and delete anything personal
- Note Claude in your record of processing activities if staff use it for work
On Team and Enterprise plans, check how your organisation's data exports and retention settings treat incognito chats. Incognito keeps a chat out of memory, but you should not assume it removes the chat from your audit trail. For a wider check, work through our AI data protection checklist for UK businesses before rolling Claude out.
Where does memory save real time in a small business?
UK businesses are adopting AI widely but not deeply. ONS figures published in July 2026 show that around a third of larger firms now use AI, but few of them describe their use as extensive. Memory is aimed squarely at that gap, because it makes everyday use less of a chore.
Take a plumbing firm that writes its job confirmations, review requests and late-payment nudges through Claude. It gets the most from memory, because the context barely changes. For customer-facing replies specifically, see our guide to AI customer service for UK small businesses.
What are the downsides of letting Claude remember?
The failure mode is subtle. You tell Claude your day rate in spring, raise it in autumn, and forget to mention the change. Every proposal Claude helps draft afterwards leans on the old number. Nothing errors. The answers just drift.
Two more trade-offs to know about:
- Shared logins. If two people share one account, memory merges their preferences. Give each person their own seat.
- Imported memories. Anthropic describes its import and export tool as experimental. Read everything it brings in from another assistant before trusting it.
How should a team roll out Claude memory?
A simple rollout for a business of five to twenty people:
- Week one. The owner or office manager uses memory alone, on internal tasks only.
- Week two. Review the stored memories. Delete anything inaccurate or personal.
- Write the rules. One page: what can go into Claude, what must use incognito, one project per client.
- Open it up. Members switch on memory individually and get a short walkthrough.
- Monthly check. Each person skims their memory list for stale facts.
If you are deciding whether Claude should take on admin that a human assistant handles now, our comparison of an AI admin assistant versus a virtual assistant sets out where each one fits. For setup guides, prompts and plan comparisons, the Claude hub collects everything in one place.
Claude Memory Feature — FAQ
Is the Claude memory feature free?
Yes. In March 2026, PCWorld reported that Anthropic was extending memory to free accounts. Paying users had first got it in August 2025. Anthropic's help centre now lists memory as on by default for Free, Pro and Max users. Searching past chats is a separate feature, and that one is still limited to the paid Pro, Max, Team and Enterprise plans. If you run a small business on the free plan, you get memory but not past-chat search. Check which of the two you actually rely on before you decide whether to upgrade.
Can I stop Claude remembering a specific conversation?
Yes. Anthropic's help centre describes two options. You can open an incognito chat using the ghost icon. It is not saved to memory or chat history, and Claude will not search it later. Or, in a new chat, open the plus menu and switch memory off before you send your first message. That chat still stays in your history, and you cannot change the setting once the chat has started, so decide before you type. For anything involving a named customer's complaint, a staff grievance or payment details, incognito is the simpler habit to teach your team.
Does Claude memory mix up information between different clients?
Not if you use projects properly. According to Anthropic, each project has its own memory. Inside a project, Claude uses that project's memory plus your general account memory, but not memory from other projects. If you want a client project fully sealed off, turn off 'Use account memory' on that project's page. Deleting a project also deletes its memory. A sensible setup for a bookkeeper or agency is one project per client, with account memory switched off for any client whose work is confidential.
Do Team plan users get memory automatically?
No. On Team and Enterprise plans, Anthropic says the organisation owner has to enable memory first in the organisation's capability settings. It then stays off for each member until that person switches it on. Turning memory off at organisation level permanently deletes memory data for every member, so an owner should not toggle it casually. Owners also cannot view or edit individual members' memories. Some Team and Enterprise organisations are still on the older memory system, which builds a summary every 24 hours rather than saving topics as you chat.
Can I move my ChatGPT memory into Claude?
Anthropic's help centre describes a way to move memory to and from other AI assistants, and says the feature is experimental and still in active development. Treat whatever it brings across as a draft. After importing, open your memory settings and read what came across. Delete anything out of date or anything that names customers or staff. An imported memory full of old prices or a former supplier's details will quietly skew every answer Claude gives you afterwards.
