The Problem: Claude Has No Natural Memory
Every time you open a new conversation with Claude, it starts with no recollection of anything you’ve discussed before. The Claude you’re talking to right now has no idea what you worked on yesterday, what your business does, or what tone you prefer — unless that information is fed into it somehow.
This is a deliberate architectural decision, not an oversight. But for anyone using Claude as a real business tool, it creates a recurring problem: you spend the first part of every conversation re-explaining context that hasn’t changed.
Memory and Preferences are the two systems designed to solve that. They work differently, sit in different places, and serve different roles.
What Are User Preferences?
User Preferences are a free-text field you write yourself. Think of them as standing orders — instructions that get loaded into every single conversation before you type your first message.
They live at Settings → Profile → User Preferences in claude.ai.
Whatever you write there, Claude reads it at the start of every chat. It’s permanent until you edit it manually. Claude never writes to it — only you do.
What belongs in Preferences:
- Your business context (who you are, what you do)
- Technical defaults (default programming language, document format, tone)
- Routing rules (“always use my work email for business tasks”)
- Credentials and frequently-referenced data (API endpoints, account handles)
- Hard rules about how you want Claude to behave (“never use bullet points”, “be direct and skip the pleasantries”)
What doesn’t belong in Preferences:
- Information that changes frequently — Preferences require a manual edit every time
- Project-specific context — that belongs in a Project instead
- Sensitive credentials if you have a better place to store them
One important note: changes to Preferences only take effect in new conversations. If you edit Preferences mid-conversation, the current chat doesn’t pick up the changes.
What Is Memory?
Memory is Claude’s automatic note-taking system. As you use Claude over time, it synthesizes information from your conversations — your role, working style, recurring projects, preferences you’ve expressed — and stores it as a set of discrete entries that carry forward into future conversations.
Unlike Preferences, Claude writes to Memory. You can read it, edit it, add to it, or delete entries — but it also updates itself based on what you tell it in conversation.
You can find it at Settings → Capabilities → View and edit memory.
Claude’s memory is not a recording. It’s closer to the notes a good assistant would keep between meetings: key facts, preferences, and ongoing context that make future interactions more useful. Once a day, Claude synthesises your recent standalone conversations and updates a profile in the background.
How They Work Together (and Where It Gets Confusing)
Here’s the part most guides skip: both systems are loaded into every conversation simultaneously. Claude doesn’t check one before the other — they’re both present from the first message. Think of it as two documents sitting on the desk before the meeting starts.
| User Preferences | Memory | |
|---|---|---|
| Who writes it | You | Claude (auto) + you (manual edits) |
| When it updates | When you edit it manually | Every ~24 hours (auto) + immediately when you instruct Claude |
| What it’s for | Standing orders, hard rules, credentials | Evolving context, corrections, learned facts |
| Where it lives | Settings → Profile | Settings → Capabilities |
| Limit | Not publicly documented | 30 editable entries |
| Takes effect | New conversations only | Immediately |
The conflict problem: If the same information lives in both places with different values, Claude will try to reconcile them — but it can’t always tell which is correct. One of them will be stale and you won’t know which one Claude is acting on. Pick one location per type of information and be consistent.
A Practical Setup for Business Users
Put in Preferences:
- Your name, business name, location
- Default email accounts and routing rules
- Hard behavioural rules that never change
- Frequently-used credentials and technical defaults
- Your two or three most important contextual facts (industry, tools, clients)
Put in Memory (or let Claude learn):
- Active projects and their current status
- Decisions you’ve made (“we dropped X — don’t reference it again”)
- Workflow corrections
- Agent-specific rules and exceptions
- Anything that tends to change month-to-month
Remove from Memory:
- Anything that duplicates what’s already in Preferences
- Outdated project context
- Information that’s no longer true
Keep out of both:
- Sensitive passwords and API keys (use a password manager)
- Confidential client data
- Anything you’d be uncomfortable with sitting in a cloud-accessible profile
The 30-Entry Memory Limit
This catches power users off guard. Memory has a hard cap of 30 editable entries. Once you hit it, Claude can’t add new ones without replacing an existing entry.
I hit this limit myself recently — mid-morning briefing, Claude trying to log a new instruction, nowhere to put it. When I pulled up my Memory entries to make room, I found WordPress credentials already stored in Preferences taking up 13 slots, a note that I drink Coke instead of coffee, and a handful of other entries Claude had helpfully learned from passing mentions in conversation. None of it needed a permanent slot in a 30-entry cap. The audit took five minutes and freed up enough space to store things that actually matter.
Everyone’s Memory has a version of the Coke entry. Claude is good at picking up context — sometimes too good. If you haven’t reviewed yours lately, there’s a decent chance it’s holding onto things you said once in passing that have been quietly occupying real estate ever since.
If you’re running a serious workflow — multiple clients, multiple agents, multiple tool integrations — you’ll hit this limit. The workaround is periodic maintenance: review your Memory entries every few weeks and consolidate or delete anything that’s redundant, outdated, or already covered in Preferences.
The biggest space-wasters are usually credentials and technical configs that got auto-added to Memory when they already lived in Preferences. Delete those duplicates first.
Viewing and Editing Your Memory
You can ask Claude directly what it has stored at any time:
“What’s in your memory about me? Show me everything.”
Claude will list its current entries. You can then instruct it to update, correct, or delete specific items in that same conversation — changes take effect immediately, not at the next 24-hour synthesis cycle.
You can also manage entries directly in the Settings UI: Settings → Capabilities → View and edit memory. The settings view and the conversation view should match — if they don’t, the settings view is the authoritative record.
One More Thing: Incognito Mode
When you start a conversation in Temporary Chat (Incognito) mode, Claude does not read your existing memories and does not create any new memories from that conversation. Use it for sensitive client work, confidential discussions, or any conversation where you want a clean slate without your standing context influencing Claude’s responses.
The toggle is in the top-right corner of a new conversation in claude.ai.
The Bottom Line
Memory and Preferences aren’t interchangeable — they’re complementary layers designed for different types of information. The businesses that get the most out of Claude are the ones that treat this setup intentionally: Preferences for what never changes, Memory for what evolves, and a habit of periodic cleanup to keep both accurate.
It takes about 30 minutes to set up properly. After that, you stop re-explaining yourself at the start of every conversation — and Claude starts showing up already knowing the basics.
Questions about AI workflows for your business? Book a call.