Claude Projects

Claude Projects gives a recurring job a reusable brief: keep the client background, approved reference files and working instructions together, then start a fresh chat for each report, email or decision. You spend less effort rebuilding the setup every time you ask Claude for help.

Start with one client or one recurring deliverable. A small, maintained reference pack is more useful than a folder full of everything your business has ever written.

Interface note: this guide covers chat-based Projects at claude.ai. Anthropic is also rolling out a new Projects beta through Claude Code and moving some existing projects into it. If your interface shows an ongoing conversation with parallel threads and a Library, follow the new Projects documentation for its controls.

What a Claude Project Holds

Think of a Project as a client workroom. The reference shelf holds the facts, the standing brief describes how to work, and individual chats are the jobs you do at the desk.

These are the core building blocks of Anthropic's chat-based Projects.

How Claude Uses Project Knowledge

Files added to project knowledge provide context across that project's chats. On paid plans, larger collections can trigger retrieval augmented generation, or RAG: Claude searches the reference shelf for relevant passages instead of putting the whole shelf on its desk for every answer. This activates automatically near the context limit, using a project knowledge search tool. Naming the document you need helps focus the request. Anthropic's RAG guide

Knowledge and memory are different. Knowledge is the material you deliberately supply. Claude also has project memory and chat-search features. In the current memory experience, a Project can use account memory from outside projects; turning off Use account memory under the Project's Memory settings keeps its memory separate. Memory controls vary with the experience available to your account. Anthropic's memory guide

For business work, make approved files your reference point. A useful remark in an old chat should become a dated decision in your reference pack before you rely on it as policy.

Which Plans Include Projects?

You can use Projects on Free, Pro, Max, Ultra, Team and Enterprise. Free accounts can create up to five Projects. Enhanced knowledge capacity through RAG is available on the paid plans. Project sharing with colleagues inside your organization is a Team and Enterprise feature. Anthropic's Projects overview

For a solo client-reporting setup, begin with the plan you already have. Choose a shared workspace when several people need the same maintained brief.

Set Up a Project for Monthly Client Reporting

Use this example for an agency account manager preparing an SEO performance report. The proposed files and prompts are a starting template, not a claimed client result.

1. Create a Project With One Clear Job

Open claude.ai/projects, select + New Project, and name it something recognizable, such as Northstar Client Reporting. Add a description for your own organization. Claude does not receive the Project's name and description as context, so put the actual assignment in its instructions too. On Team or Enterprise, start with private visibility and invite the people who need access. Project setup instructions

2. Add a Small Reference Pack

Use the + control in the Project's knowledge area to add the standing files. Keep a current date and an owner inside each document. These are suggested filenames and contents:

Keep the latest month's exports separate from this standing pack until needed. Attach them to the report chat with filenames that identify the reporting period. That makes it easier to distinguish this month's evidence from last month's example.

3. Write Instructions You Can Reuse

Select Set project instructions, paste your working rules, then select Save instructions. These instructions apply across the Project's chats. Anthropic's setup guide

Here is a copyable starting point:

You help our account manager prepare monthly SEO reports for Northstar. Use the project files for approved client background, metric definitions and report structure. Use the reporting-period files attached to the current chat for performance evidence.

Start by confirming the reporting period and identifying missing inputs. Follow report-structure.docx. Treat approved-report.pdf as an example of tone and layout, never as current performance data.

Distinguish observed results, possible explanations and recommended actions. Do not invent metrics, targets, causes or client approvals. If sources conflict, show the conflict and ask which source is authoritative.

For each material claim, identify the source filename and the relevant row, section or page where available. Mark unsupported claims as needing review. Use concise, plain English for a busy client stakeholder.

Finish with next actions, proposed owners and open questions. Treat every output as a draft for the account manager to approve.

These rules give you a review standard. They do not guarantee correct citations or calculations.

4. Test the Brief Before Drafting

Start a new chat inside the Project and ask:

Using client-brief.txt and metric-definitions.csv, explain the client's business goal, the metrics we report and what would count as success. Name the source for each answer. List anything you cannot establish from the files.

Check the answer against the documents. If Claude cannot distinguish a business goal from a traffic metric, fix the reference pack before generating a polished report.

Then attach the current exports and request:

Prepare the report for the period covered by these exports. First check the date ranges, metric definitions and missing comparisons. Then draft the report using our project instructions. Show the evidence behind each headline finding and leave gaps visible.

Review totals, date ranges and explanations against the original exports. A fluent explanation for a change is still only a hypothesis until the evidence supports it.

5. Keep Knowledge Current After Each Report

Assign one person to maintain the pack. Before each reporting cycle, replace outdated uploaded copies, remove superseded versions and update the decision log. Keep a separate archive in your normal document system if you need historical versions.

After approval, add durable decisions to decision-log.txt. Update the example report only when the new version is worth copying. Avoid filling the reference shelf with every draft and abandoned recommendation.

The workflow changes from find the brief, paste the rules, attach evidence, draft, review to maintain the brief, attach current evidence, draft, review. The repeated setup is the part you target.

For planning only: if reassembling context takes 15 minutes for each of eight monthly client reports, that represents two hours of setup. Measure how much your Project actually removes; collecting data and checking the report still take work.

How Sharing Works on Team and Enterprise

Share the Project when colleagues need the same reference pack. Public means visible to your organization, rather than published on the internet. Private restricts access to invited members.

Open the Project's Share menu, add colleagues by name or email, and choose their permissions:

Your chats remain private by default when you share the Project. Sharing a chat is a separate action that creates a snapshot; later messages require an update to that shared snapshot. Enterprise also has group sharing in beta, subject to group settings.

Administrators can restrict sharing. Archiving a Project preserves its permissions, so remove someone's access explicitly when their involvement ends. Anthropic's visibility and sharing guide

For the reporting example, give the person maintaining the reference pack editing access. Colleagues who only need to draft from it can use viewing access.

File Limits and Knowledge Capacity

The Project file limit is 30 MB per file. Do not substitute the separate chat-attachment limit when preparing a project knowledge pack.

Additional limits can depend on the amount of extracted text. Embedded images in non-PDF documents are not interpreted. Anthropic's file-upload rules

Unlimited file count does not mean unlimited knowledge. The context window is the amount Claude can work with at once, and its size depends on the model. On paid plans, automatic RAG expands project knowledge capacity by up to 10 times. That is an expansion mechanism, not a promise that every uploaded page is read for every answer. Context-window documentation, Project RAG documentation

Do not plan around a universal document count, a fixed word allowance or the old blanket 200K-token description. Use the current model limits and the capacity shown in your Project.

My recommendation: use Projects for work with a stable brief and a human reviewer. The reporting setup above still requires someone to collect fresh exports, reconcile conflicting sources and approve the result. If you need a live dashboard or unattended report delivery, evaluate that workflow separately.

Five Jobs Worth Organizing This Way

The best candidates repeat the same background while producing different outputs. In order of where I would start:

Client reporting is my first choice because the next cycle gives you a clear comparison: how much briefing was repeated, how many corrections were needed, and whether the final report improved.

Two Services You Could Build Around This

The strongest opportunity is helping a team establish and maintain its working context. These are proposed services around Projects, not claims about a Projects automation API.

1. A client-reporting setup kit for small agencies. DataForSEO estimates 590 monthly US Google searches for “seo client reporting”, retrieved for this article in November 2026. A small sellable version could include one client's reference pack, project instructions, a reporting prompt and an acceptance checklist tested against an approved past report. The agency pays for repeatable reporting setup. The catch: exports and definitions differ between clients, so a generic prompt alone is easy to copy and insufficient.

2. An internal-knowledge maintenance service. The same data source estimates 260 monthly US searches for “internal knowledge base software.” A small version could organize one department's approved procedures, identify conflicting versions and assign an owner and review date to every document. A Team or Enterprise Project becomes a place to use that material. The catch: a department still needs a system of record, access decisions and accountable owners; adding an AI interface does not supply those.

These search estimates show interest in the underlying jobs, not a count of willing customers. The reporting kit is the stronger starting offer: it has a bounded deliverable and a recurring moment when the buyer can judge whether it helped.

Claude Projects, ChatGPT Projects and Claude Code

ChatGPT Projects serves a similar organizing job: related chats share uploaded files, instructions and connected sources. Choose based on the assistant and sources your team already uses, then compare one real deliverable. OpenAI's Projects documentation and our ChatGPT Projects guide cover that setup.

Claude Code has project features for a different working environment. Repository instructions such as CLAUDE.md guide coding sessions. Its newer Projects beta coordinates work through cloud threads, repositories, files and memory. That is a different workflow from the chat-based reference pack above, although Anthropic is moving some existing Projects into the new experience. Claude Code memory and instructions, Claude Code Projects

For the assistant's broader strengths and tradeoffs, see our Claude review.

What are useful Claude Projects examples?

Start with a recurring client report, product-update workflow or account-review preparation. Give each Project one coherent job and an owner for its reference files. Keep unrelated clients in separate Projects.

How do I make an SEO report for a client with Claude Projects?

Add the client brief, metric definitions and approved reporting structure to project knowledge. Attach the current period's exports to a report chat, ask for gaps before drafting, and check every headline finding against its source before sending it.

How do Claude Projects differ from Claude Code Projects?

The chat-based version organizes conversations around shared reference knowledge and instructions. The newer Claude Code Projects beta coordinates ongoing work through parallel threads. Check which interface your account has before following setup steps.

Your Monday move: choose the next client report, appoint its reference-pack owner, upload the standing files and test Claude's understanding before asking for the draft. Track setup time and corrections so you can decide whether to expand the approach.

If your team needs help turning this into a maintained reporting workflow, explore our AI automation service.