Claude Projects: Your Persistent Research Assistant

What This Builds

Instead of re-explaining your firm's report style, brand voice, methodology standards, and client context every time you start a Claude conversation, you'll build a Claude Project that retains all of that context permanently. Every conversation you start within the Project immediately "knows" your firm's standards, your client's background, and your current study's objectives. The first draft of an executive summary that normally takes 20 minutes of context-setting becomes 5 minutes of just pasting findings and asking for the draft.

Prerequisites

• {{tool:Claude.plan}} subscription at {{tool:Claude.price}} — Projects require a paid account
• A completed research report or two to use as style examples
• Your firm's report template (if you have one)
• Time to build: 1–2 hours

The Concept

A Claude Project is like a permanent briefing room. You walk in, and instead of introducing yourself every time, the room already knows who you are, what style you write in, what your client does, and what you're trying to accomplish. You set up the room once — by uploading reference documents and writing a project instruction — and then every conversation that happens inside that room starts from that foundation. When the project ends (or a new client starts), you create a new Project for them.

Build It Step by Step

Part 1: Create the Project

1. Go to {{tool:Claude.url}} and sign in to your {{tool:Claude.plan}} account
2. Look for Projects in the left sidebar — click it
3. Click New Project
4. Name it clearly: "Client: [Client Name] — [Study Name]" or "Firm Research Assistant — [Year]" for a general-purpose version
5. You'll land on the Project setup page

What you should see: A Project page with a description field, an "Instructions" section, and a "Knowledge" section for uploading files.

Part 2: Write the Project Instructions

The instructions are the most important part — they tell Claude who it is, how it should write, and what it knows. Click the Instructions field and write this (customized for your situation):

You are a research analyst assistant for [Your Name / Firm Name], a market research analyst at [firm type].

Your role:
- Help draft research report narratives, executive summaries, and section introductions
- Code open-ended survey responses into themes
- Review survey questions for bias and methodology problems
- Generate client-facing language from statistical findings

Style guide:
- Write in confident, assertive business language — not passive or hedging
- Lead with the insight, not the methodology
- Use plain language — no statistical jargon in client-facing content
- Report tone: [professional / conversational / formal — pick one]
- Always answer "so what does this mean?" in any implication you write

Current client context:
- Client: [client name or type]
- Industry: [industry]
- Business question: [what the client is trying to learn or decide]
- Primary audience for deliverables: [who reads the reports]
- Key terms to know: [any brand names, product names, or industry jargon to spell correctly]

Report structure we use:
- Executive Summary (300 words, 3 implications)
- Section introductions (100 words each)
- Key Insight callout boxes (2 sentences, assertive)
- Implications slide (4–5 action recommendations)

Click Save.

Part 3: Upload Knowledge Files

In the Knowledge section, upload documents that Claude should reference:

1. Report template — Upload a previous report (with client data removed) as a style example. Claude will match the structure and tone.
2. Methodology guide — If your firm has a standard methodology document, upload it so Claude can write accurate methodology sections.
3. Brand voice guide — If you have a style guide, upload it.
4. Prior reports for this client — Upload 1–2 previous deliverables for this specific client so Claude matches their preferred format.

What you should see: Each uploaded file appears as a named document in the Knowledge section.

Part 4: Test and Refine

1. Click New chat inside the Project — this opens a conversation that inherits all your instructions and knowledge files
2. Test with: "Here are 5 key findings from our current study. Write an executive summary." — paste made-up or real findings
3. Evaluate the output against your style standards
4. If it's off, go back to the Instructions and adjust — the most common fix is adding more specific style guidance

What you should see: An executive summary that sounds like it came from your team, not a generic AI.

Real Example: Client-Specific Research Project

Setup: You're working on a brand health tracker for a regional grocery chain. You've built a Project called "Client: Harvest Foods — Brand Health Tracker Q1 2026."

Instructions uploaded:

• Style guide: "Use straightforward, direct language. Executives read on mobile. Avoid passive voice."
• Client context: "Harvest Foods is a 45-store regional grocer competing against Kroger and Whole Foods. Their key brand attribute is 'community-connected.' Their CMO, Sarah Chen, prefers bullet-format implications, not paragraphs."
• Knowledge files: last quarter's brand health report (anonymized template) + their brand positioning document

Input: You paste 8 key findings from the latest wave of tracking data.

Prompt: "These are the Wave 3 findings from our Harvest Foods brand health tracker. Write the executive summary for this wave, noting any significant changes from Wave 2 (which showed awareness at 67% and consideration at 41%). Use Sarah's preferred bullet format for implications."

Output: A Wave 3 executive summary that references Wave 2 comparison data, uses their brand language, and formats implications as bullets — without you having to re-explain any of that context.

Time saved: 15–20 minutes of context-setting eliminated per session.

What to Do When It Breaks

• Output sounds too generic → Add a specific style example to the Instructions: "Write the way this sentence sounds: [paste example from your best report]"
• Claude ignores the knowledge files → Explicitly reference them in your prompt: "Using the report template I uploaded as a style guide, write..."
• Tone is wrong for the client → Add more client-specific context to Instructions: "This client dislikes technical hedging language. Never say 'may suggest' or 'could indicate' — say what it means."
• Project is mixing up clients → Create separate Projects for each client — don't try to use one Project for multiple clients

Variations

• Simpler version: Don't upload knowledge files — just write thorough Instructions. This alone saves significant context-setting time even without documents.
• Extended version: Create a Project for each major study type (brand tracker, customer satisfaction, concept test) rather than each client — then the methodology context is always current.

What to Do Next

• This week: Build one Project for your highest-priority current client
• This month: Build Projects for your top 3 recurring clients or study types
• Advanced: Ask Claude to suggest what additional context would make its outputs better — it will tell you what's missing from the Instructions

Advanced guide for Research Analyst professionals. These techniques use more sophisticated AI features that may require paid subscriptions.