Practical. No hype.
You spend twenty minutes giving Claude the context it needs — your role, your department, what you’re working on, how you prefer to communicate. It gives you exactly what you need. You close the tab. Next time you open it, the slate is blank. Claude knows nothing about you. You start again.
This is the thing that frustrates most surgeons when they first use AI tools seriously. It looks like the tool has a short memory. The reality is more specific: AI has no memory between conversations by default — but it has a very good memory within a well-structured setup. The difference between those two states is worth understanding.
What actually persists — and what doesn’t
Within a single conversation, Claude holds everything. All context, all instructions, all documents you’ve shared — it’s all available throughout that session. This is the context window, and for Claude it currently runs to 200,000 tokens (about 150,000 words — roughly a 500-page book). Most clinical writing tasks fit easily within that.
Between conversations, nothing persists — unless you set it up to. That’s the gap most people fall into. They use Claude like a search engine, a fresh question each time, and never build the scaffolding that makes the tool genuinely useful.
There are three mechanisms for getting context to persist. Each solves a different problem.
1. Claude’s memory feature
Claude’s built-in memory feature automatically synthesises key facts from your conversations — your name, your role, preferences you’ve mentioned — and applies them in future sessions. It’s passive: you don’t need to do anything to activate it. As of 2026, it’s available across all tiers including Free.
The limitation is that memory is broad and slow-updating. It captures the basics but it doesn’t hold granular context — the specifics of a project you’re working on, a patient cohort you’ve described, or a set of instructions about how to draft a particular type of letter. For that, you need Projects.
2. Projects
Projects are the feature that changes how useful Claude is for professional work. A Project is a persistent workspace: it has its own system prompt (instructions that apply to every conversation within it) and its own document library (files Claude can reference throughout). When you open a conversation inside a Project, Claude already has your context. You don’t start from scratch.
The practical setup for a surgeon: one Project for clinical correspondence, with your voice fingerprint and a template for referral letters in the system prompt. One Project for research and literature review, with the guidelines or papers you reference regularly. One Project for admin, with your rota and department information. Each one is a different working environment with Claude already briefed for that context.
A comparison published in 2026 found that Claude Projects and ChatGPT Projects handled persistent context differently — Claude’s approach prioritises document access and instruction fidelity, which suits structured professional workflows rather than casual use (Unmarkdown, 2026). For the kind of work most surgeons want to do — consistent writing, document analysis, structured tasks — that distinction matters.
3. The briefing document
A briefing document is a simple text file you keep updated and paste into any new conversation where you need full context quickly. Think of it as a one-page professional profile: your role, your department, the kind of work you typically ask Claude to help with, your preferences, and any standing instructions. It takes ten minutes to write once and solves the context problem for any conversation where a Project isn’t set up.
A briefing document looks something like this:
I am a consultant orthopaedic surgeon at [hospital]. My subspecialty is [shoulder / trauma / arthroplasty]. I use Claude primarily for: drafting correspondence, summarising literature, preparing presentations, and working through clinical governance questions. My writing style is direct and concise — short sentences, no hedging, clinical detail in the first paragraph. When drafting letters, use a reading age of around 12 and avoid passive constructions.
Update it when something significant changes. Paste it at the start of any important conversation. It takes five seconds and saves fifteen minutes.
A note on what “context rot” means in practice
Even within a long conversation, accuracy can degrade as the context window fills — a phenomenon sometimes called context rot. If you’re working through a very long document or an extended multi-turn task, things said early in the conversation may be weighted less than things said recently. The practical fix is to restate key instructions periodically, or start a fresh conversation in a Project where those instructions live in the system prompt rather than the conversation history.
The next post in this series covers how to set up a Project properly: Setting up Claude properly — a 15-minute guide for surgeons.
Not sure where your context is disappearing to? Logged-in members can book a session — 30 minutes (£75) to diagnose the issue and get the right setup in place, or 60 minutes (£150) to build out your full Project structure so the context that matters never gets lost again.
References
- Anthropic. Context windows documentation. platform.claude.com. Accessed May 2026.
- Anthropic. Claude memory FAQ. support.claude.com. Accessed May 2026.
- Unmarkdown. Claude Projects vs ChatGPT Projects: Which AI Knowledge Base Is Better in 2026? unmarkdown.com. 2026.