How to Use Claude Projects: 5 Steps to Get Started
Every new Claude chat starts from zero. It has no memory of your business, your style, or the five files you uploaded last week. That’s fine for a one-off question. It’s a real time sink if you’re doing the same kind of work over and over. This guide covers exactly how to use Claude Projects to fix that, in five steps.
In 2026, SMB employees saved an average of 5.6 hours a week using AI tools, according to a Business.com and Dialog survey of small-business leaders and employees. Managers saved even more: 7.2 hours a week, against 3.4 for individual contributors. However, part of that saved time leaks right back out. It happens every time you re-explain context Claude should already have.
Claude Projects fixes that. It’s now available on every Claude plan, including Free. It bundles your instructions, your files, and your chats into one workspace that actually remembers. This guide covers the five setup steps, plus when skipping a Project entirely is the smarter move.
Key Takeaways
- A Claude Project bundles custom instructions, uploaded files, and chats so context persists automatically instead of resetting every conversation.
- Projects reached the Free plan in February 2026, capped at 5 projects; Pro, Max, and Team get unlimited projects.
- Setup takes five steps: create the project, write instructions, upload files, then start chats from inside it.
- Not every task needs one — one-off questions don’t benefit from the setup.
Part of my ongoing coverage of practical AI tools for solo operators — The Streamlined CEO newsletter covers breakdowns like this one every Tuesday.
Table of Contents
What Is a Claude Project?
A Claude Project is a workspace built from three parts: custom instructions, a knowledge base of uploaded files, and chats that inherit both. In February 2026, Anthropic brought Projects to the Free plan, capped at 5 projects; Pro, Max, and Team plans get unlimited projects.
Think of a regular chat as a conversation that forgets you the moment it ends. A Project chat is different. Open a new chat inside a Project, and it already knows your instructions. It’s already read your files. All of that happens before you type a single word.
That distinction matters more than it sounds. Have you ever pasted the same brand guidelines into Claude three times in one week? That’s the exact problem Projects solves. You write the context once. Every chat inside the project reads it automatically.
For comparison, Claude for Small Business layers pre-built connectors and workflows on top of a standard subscription. Projects is the free, build-it-yourself version of the same idea. You structure your own setup instead of buying a packaged one.
Step 1: How Do You Create a New Project?
Go to claude.ai/projects, or click “Projects” in the left sidebar. Then click “New project” in the top-right corner. That’s the entire first step.
You’ll be asked for a name and an optional description. Here’s the part almost every new user misses: Claude doesn’t actually read the project name as context. Neither does it read the description. Those fields are for you, not for the model. Name it something you’ll recognize at a glance, like “Q3 Client Reports” or “Blog Editorial.” Just don’t expect Claude to infer anything from that label alone. The real context has to live in your instructions and your files. That’s exactly what the next two steps cover.
Step 2: What Should You Write in Project Instructions?
Project Instructions are a permanent system prompt that every chat in the project reads first, before responding to anything you say. In other words, write it once, and it applies automatically to every future conversation.
So what actually belongs here? Your role for Claude to play. Your tone and formatting preferences. Any constraint that applies to every chat in this project, not just one. A content project, for example, might specify word count targets and a reading-level goal. A client-reporting project might specify a fixed structure every report has to follow.
What doesn’t belong here is anything specific to a single conversation. Keep those as regular messages instead. The test is simple. If you’d type the same instruction into every chat in this project anyway, it goes in Project Instructions. If it only applies once, it doesn’t.
A concrete example helps. For a client-reporting project, instructions might read: “Always structure reports as Summary, Findings, Recommendations. Use plain business language, no jargon. Flag any number you’re not fully confident in rather than guessing.” That’s specific enough to change every future chat’s output, and general enough to apply to any report the project generates.
Step 3: What Can You Upload to a Claude Project?
Claude Projects support a knowledge base of uploaded files. Style guides, product one-pagers, and past reports all count. Claude reads them automatically before every response in the project. That said, check Anthropic’s own Projects documentation and pricing pages at write time for the current exact file-count and size limits, since third-party estimates vary and the feature has changed more than once since launch.
The table below reflects limits reported by several 2026 guides at the time of writing. Treat these as directionally correct only. Confirm the current numbers directly on Anthropic’s site before relying on them for a specific project.
| Plan | Reported file size limit | Reported files per conversation |
|---|---|---|
| Free | ~10MB per file | Fewer than paid tiers |
| Pro / Max / Team | ~30MB per file | Up to 5 per conversation |
In practice, the real constraint usually isn’t file count. It’s the context window. A knowledge base stuffed with dozens of long documents competes for the same space as your actual conversation. Upload what Claude needs to reference, not everything you have on the topic. When trimming a knowledge base, prioritize files that answer questions Claude can’t infer on its own, current pricing, exact product names, house style rules, over background material Claude could reasonably guess or generate itself.
In my experience running this blog, the same pattern applies outside Claude’s UI entirely. Standing instructions live in one file that every new article session reads first, similarly to how a Project’s instructions work. A knowledge base of past briefs and style rules sits alongside it. Each article gets its own scoped session that inherits both automatically. It’s the Projects model, just applied to running a one-person content operation.
Step 4: How Do You Start a Scoped Chat Inside the Project?
Click “New chat” from inside the Project. Not from the main Claude sidebar. That one click decides whether the conversation inherits your instructions and files, or starts from zero like any other chat.
This trips up more new users than anything else in this guide. Start a chat from the main sidebar instead, and Claude has no idea your project exists. It hasn’t read your instructions. It hasn’t seen your files. So if Claude seems to have “forgotten” a project you carefully set up, check where you clicked “New chat” first. Don’t assume something’s broken.
You can run several chats inside the same project at once. Each one still reads the same instructions and files independently, they just don’t share what happens inside each other. Use separate chats for separate threads of work within the same project, rather than one long chat covering everything.
Step 5: How Do You Manage a Project Over Time?
A Project isn’t a one-time setup you configure and forget. Treat the instructions like a living document. As your process changes, update them. If Claude keeps missing something, fix the instructions rather than repeating the correction in every new chat.
The same goes for files. Swap out a style guide once it’s outdated. Remove a report Claude no longer needs. A knowledge base that’s accurate and lean produces better responses than one stuffed with every document you’ve ever touched on the subject. A simple prune trigger: if you haven’t referenced a file’s source information in the last month and nothing in it has changed, it’s a candidate to remove rather than keep accumulating.
A Real Example: Running a Content Operation as a Claude Project
Here’s a concrete version of the pattern above. I run this site’s content pipeline, planning topics, drafting articles, checking quality, publishing, as a standing set of instructions plus a knowledge base of prior work. That’s exactly what a Claude Project provides natively.
My instructions cover the recurring rules: a readability target, a sourcing standard, a publishing checklist. The knowledge base holds the article queue and lessons learned from previous posts. As a result, each new piece benefits from what the last one taught. Every new article starts its own scoped conversation, but none of it starts from zero.
In 2026, 78% of US businesses said AI improved their productivity, up from 46% in July 2024, according to Intuit’s QuickBooks AI Impact Report. That gain compounds when the AI itself doesn’t need re-briefing every session. In particular, for a one-person operation with no team to onboard, that’s the entire value of setting this up properly once.
When You Don’t Need a Project
Not every task deserves a Project. Skip it for one-off questions. Skip it for tasks that won’t recur. Skip it whenever the full context fits comfortably in a single message. Setting one up adds a few minutes of overhead that only pays off with reuse.
The real signal to watch for is repetition. Have you pasted the same background information into more than one chat this week? That’s the trigger to make it a Project instead. Otherwise, a plain chat is faster and just as effective.
Two related questions come up often enough to deserve their own answers. When does a Project actually beat a regular chat for a specific task? And how does a Project differ from Claude’s Artifacts feature? Both get a fuller treatment in upcoming posts on this site.
Common Mistakes When You First Use Claude Projects
New users tend to hit the same handful of snags. Here’s how to use Claude Projects without falling into them.
Treating the name and description as context. Covered in Step 1, but it’s worth repeating: Claude doesn’t read the project name or description. If something matters, it belongs in the instructions or the files, not the label.
Starting chats from the wrong place. The single most common mistake. A chat started from the main sidebar never inherits your project’s setup, even if you’re staring at the project’s page a moment before. Always click “New chat” from inside the project itself.
Assuming a Project remembers everything you’ve ever discussed. A Project gives every chat inside it the same starting context: instructions and files. It doesn’t merge memory across every past conversation in that project the way some other memory-first tools do. Each chat still runs independently once it starts. It just starts from the same baseline instead of nothing.
Uploading everything instead of what’s relevant. More files isn’t automatically better. A knowledge base padded with outdated drafts or irrelevant reports competes for the same context window as your actual conversation. Curate it the way you’d curate anything you hand someone on day one: relevant and current, not exhaustive.
Never revisiting the instructions. Projects set up once and never touched again drift out of date fast, especially if your process changes even slightly. Treat the instructions like Step 5 describes: a living document, not a one-time setup.
Most of these aren’t really about Claude. They’re the same mistakes people make setting up any shared system for the first time: skip the boring setup step, assume the tool works like something familiar, forget to maintain it. Claude Projects just makes the cost of skipping them more visible, because the whole point is skipping repeated setup.
Frequently Asked Questions
What is a Claude Project?
A Claude Project is a workspace that bundles custom instructions, uploaded knowledge files, and a set of chats, all of which share context automatically. It removes the need to re-explain your background or re-upload files at the start of every conversation.
Are Claude Projects free?
Yes. Since February 2026, Projects are available on every Claude plan, including Free (Anthropic, 2026). Free-plan users are capped at 5 projects; Pro, Max, and Team plans get unlimited projects.
What’s the difference between a Claude Project and a regular chat?
A regular chat starts with no memory of anything outside that single conversation. A chat started inside a Project automatically inherits that project’s custom instructions and uploaded files. In other words, you don’t repeat yourself each time.
How much can you upload to a Claude Project?
Claude Projects support a knowledge base of multiple files, with size limits that vary by plan. Check the exact current limits directly on Anthropic’s site, since these figures have shifted before and third-party sources don’t always agree.
Can you share a Claude Project with someone else?
Sharing and collaboration options exist on some plans, though the details depend on your specific subscription tier. This is a big enough topic that it gets its own dedicated guide on this site, covering exactly what’s shareable and what isn’t.
Do files in one Claude Project show up in another?
No. Each project’s knowledge base is isolated to that project. Files uploaded to a “Client A” project aren’t visible inside a “Client B” project, and vice versa. If multiple projects need the same reference file, upload it separately to each one.
How to Use Claude Projects: Quick Recap
Five steps get you from a blank Claude account to a working Project. Create it, write instructions, upload files, start chats from inside it, and revisit the setup as your work changes. None of it costs anything beyond the time to set it up.
The bigger shift is in habit, not mechanics. A content pipeline, a client account, a research thread: each of these is a candidate for its own Project. One-off questions aren’t. Once you notice yourself repeating context across chats, that’s the signal to stop and build the Project instead.
Not sure where your own time is actually going before you start automating any of it? This free time audit is a good place to start. And if you want more of this kind of practical, hands-on breakdown as it’s published, The Streamlined CEO covers exactly that.
Sources
- Business.com, “2026 Small Business AI Outlook Report” (with Dialog), retrieved 2026-07-29, https://www.business.com/articles/ai-usage-smb-workplace-study/
- Intuit, “2026 AI Impact Report: How AI Is Impacting Business Revenue and Productivity,” retrieved 2026-07-29, https://www.intuit.com/blog/global-stories/ai-impact-report/
- Anthropic, “What are projects?”, Claude Help Center, retrieved 2026-07-29, https://support.claude.com/en/articles/9517075-what-are-projects
- Anthropic, “How can I create and manage projects?”, Claude Help Center, retrieved 2026-07-29, https://support.claude.com/en/articles/9519177-how-can-i-create-and-manage-projects
- File-size/token limits in the table above are third-party estimates, not confirmed on an official Anthropic page as of 2026-07-29 — re-verify before relying on them.
