The Claude feature most people scroll right past
Most people who use Claude open a new chat, re-explain their context, paste the same background files, and start from zero. Every single time.
There is a feature that ends this loop, and most users never touch it: Projects. A Project is a persistent workspace that remembers your instructions and reference files across every conversation inside it.
This article walks you through setting up a Project, what to put in it, and a realistic example you can build in the next twenty minutes.
What is Claude Projects, and why does it matter?
Claude Projects is a workspace feature that bundles custom instructions and uploaded reference files into one container. Every chat you start inside a Project automatically inherits that context, so you stop re-explaining yourself and get consistent, on-brief answers from the first message.
Think of a normal chat as a stranger who forgets you the moment you close the tab. A Project is more like a colleague who already knows your style, your files, and your standing instructions.
Projects is available on paid Claude plans, and the exact controls evolve over time, so treat the steps below as the general pattern rather than a fixed menu.
Why does a Project beat starting a fresh chat every time?
A fresh chat forgets everything, so you re-paste your context and re-explain your standards on every task, which wastes time and reintroduces inconsistency. A Project stores that context once, so every conversation starts already aligned to your files and your rules. You do the setup a single time instead of daily.
The hidden cost of fresh chats is drift. When you re-explain your brand voice from memory each morning, you describe it slightly differently, and the outputs wander.
A Project pins the definition in one place. Ten chats over two weeks all read the same instructions, so the tenth answer matches the first in tone and format.
The practical payoff is speed. Your first message inside a Project can be the actual task, not three paragraphs of "here is who I am and what I need".
How do you set up a Claude Project step by step?
Create a new Project, give it a clear name, write its custom instructions, and upload the reference files you reuse most. Then open a chat inside the Project and confirm it already knows your context before you rely on it for real work. Setup takes roughly ten minutes.
Start by naming the Project after the job it does, for example "Weekly Client Reports" rather than something vague like "Work".
Next, fill in the custom instructions field. This is the standing brief Claude reads before every chat in the Project. Here is a copy-paste starting point you can adapt.
Try this Project instruction block:
You are my reporting assistant. I am a marketing manager at a mid-sized Hong Kong company.
Voice: clear, direct, no marketing fluff. Use British English spelling. Prefer short paragraphs.
When I paste raw data or notes, produce a report with these sections in order: Summary, Key Numbers, What Changed, Recommended Next Steps.
Always flag any figure you are unsure about rather than guessing. If information is missing, ask one focused question before writing.
Finally, upload the files you reach for repeatedly, such as a brand style guide, a report template, or a glossary of your internal terms. These become Claude's reference knowledge for every chat in the Project.
What should go into a Project's instructions and knowledge?
Put stable, reusable context in a Project: your role, audience, tone rules, output formats, and reference documents you use across many tasks. Keep one-off details out. A good test is simple. If you would explain it at the start of most chats, it belongs in the Project.
Strong instruction content includes who you are, who the audience is, the tone and spelling conventions to follow, and the exact structure you want outputs to take.
Strong knowledge files include a style guide, a few gold-standard past examples, product or service descriptions, and any glossary of terms the model would otherwise get wrong.
Leave out anything that changes per task, such as today's specific data or a single client's one-time request. That belongs in the individual chat, not the shared Project brief.
A realistic example: a Project for your weekly reports
Imagine you write a client performance report every Monday. With a Project, you set the instructions and upload last quarter's best report once, then each week you paste only the new numbers and get a finished draft in your exact format within seconds.
In week one you build the Project, add the instruction block above, and upload one polished past report as the gold standard example.
In week two your entire prompt becomes: "Here are this week's figures, draft the report." Claude already knows the sections, the voice, and the standard to match.
By week four you have a reliable system. The reports look consistent because they are all generated against the same fixed brief, not against whatever you happened to type that morning.
Where does Claude Projects fall short?
Projects struggles when instructions grow bloated, when knowledge files go stale, or when you force unrelated tasks into one container. It is a context store, not magic. If your instructions contradict themselves or your uploaded files are outdated, Claude will faithfully follow the wrong guidance.
Bloated instructions are the most common failure. If you cram twenty rules into the brief, the important ones get diluted. Keep instructions tight and prioritised.
Stale knowledge is the quiet failure. An outdated price list or an old template will keep shaping answers long after it should. Review your Project files on a schedule.
Finally, resist the urge to build one giant Project for everything. Separate Projects for separate jobs keeps each brief focused and each output sharp.
Try it now
Pick the one task you repeat most often this week. Create a Project for it, paste in the instruction block above with your own details swapped in, and upload a single strong past example.
Then run your next real task inside it and compare the first draft to what you normally get from a blank chat. The difference in consistency is usually obvious immediately.
Master this and AI stops being a clever stranger you brief from scratch each day, and becomes a workspace that already understands your work.
We understand AI. We understand you better. With UD by your side, AI doesn't feel cold.
Turn One Project Into a Full AI Workflow
A single Project is a great start. The next step is designing a whole set of AI workspaces that run your recurring work reliably. We'll walk you through every step, from structuring instructions to organising knowledge and rolling it out across your team.
