Build the team you can't afford yet.

The play.

One Claude. Five roles. Each role gets a specific job, a specific prompt, and a specific output. They work in order. Researcher first, Marketer last. Skip ahead and you'll build a beautiful thing nobody wants.

Most creators use Claude like Google. Ask a question, get an answer, move on. The shift is treating Claude like a team you direct , not a search bar you query.

The setup.

Before the roles, set up one Claude Project . Not a chat. A Project.

A chat forgets. By the time you reach the Marketer, Claude won't remember what the Researcher found three hours ago. A Project keeps every output as shared context, so each role builds on the last one instead of starting from scratch.

Open Claude. Click Projects. Create a new one. Name it the working title of your offer. Then paste this into the Project's custom instructions:

You are operating as a five-person team for a solo creator. The roles are Researcher, Strategist, Copywriter, Builder, and Marketer. I'll tell you which role you're in for each request. Stay in character for that role. When I switch roles, reference what the previous role produced.

Every prompt below goes inside that Project. Not a new chat. That's the spine. Now the roles.

Role 01. The Researcher.

Job: Find the pain people will actually pay to solve.

Most creators skip this. They build something they think is cool, then try to convince people they need it. That's how you spend six months on a launch nobody opens.

You are the Researcher.

Turn on web search for this task. I need real evidence, not pattern-matching from your training.

List the top 5 urgent and painful problems faced by [target customer]. For each one, pull supporting evidence from Reddit threads, Amazon reviews, or public forums I can verify.

For each problem, show me:
- The exact language people use to describe the pain
- A link to one source where someone described it
- Where they're already trying to solve it
- What existing solutions are failing them.

Paste into your Project. Turn on web search before sending. Replace [target customer] with a real specific group. "Freelance designers who hate sending invoices" beats "creatives."

Follow-up move: Whatever Claude returns, push back. "These three feel generic. Give me five more that are weirder and more specific to [audience]. Look in smaller subreddits, not r/Entrepreneur."

You know it's working when: You can quote a real person describing the pain in their own words, and you can click through to where they said it. If everything sounds like a marketing brochure, push harder.

Role 02. The Strategist.

Job: Turn the pain into a free offer people can't say no to.

Forget building the full product. Forget the perfect funnel. Build a lead magnet so good your audience would pay $50 for it. Then give it away.

This is the free taste so good they wonder what you're charging for . Real value, no asking, no gate.

You are the Strategist now.

Based on the pain points the Researcher just found, suggest 5 lead magnet ideas so valuable my target customer would pay $50-$100 for them.

Rank them by:
- Fastest to build (under 4 hours)
- Cheapest to launch (under $50)
- Most likely to convert sign-ups.

For each one, name the format (PDF, Notion template, mini-tool, Loom video, etc.) and the one sentence that would make someone download it.

Same Project. Claude already has the Researcher's findings.

Follow-up move: "Now combine the top two into one offer. What would that look like?" The best lead magnets are usually a remix, not a single idea.

You know it's working when: You can describe the lead magnet in one sentence and your test audience says "wait, you're giving that away for free?"

Role 03. The Copywriter.

Job: Write the words that make people click.

Never ask Claude for "good copy." Too vague. Build it step by step. Hook first. Then bullets. Then page. Each piece earns the next.

Run these three in order. Don't paste all three at once. Wait for the output, react to it, then move to the next.

1. "You're the Copywriter. Write 5 one-sentence hooks for [insert lead magnet]. Each one should use a different angle: contradiction, specific number, curiosity gap, fear of missing out, identity."

2. "Pick the strongest hook. Now give me 3 benefit-driven bullets that make the offer irresistible. Each bullet names a specific outcome, not a feature."

3. "Now draft copy for a one-page site with the hook as the headline, the bullets, a short middle section addressing the biggest objection, and a CTA people can't ignore."

Same Project. Each prompt feeds the next.

Follow-up move: Read every line out loud. Anything that sounds like a press release, paste it back and say "rewrite this like you'd say it to a friend."

You know it's working when: A friend who's never seen your offer reads the page and tells you what it does, who it's for, and why they'd care. If they hesitate on any of those three, the copy isn't there yet.

Role 04. The Builder.

Job: Turn the copy into a live one-page site.

No dev team. No design degree. No 12-page funnel. One page, capture emails, deliver the lead magnet. That's the whole job.

Use Claude Code for this. It runs in your terminal, edits real files, and ships a working site. If you've never used it, install it once and you'll use it for everything else after this.

Build me a visually stunning one-page site for [insert lead magnet name].

Stack: HTML, CSS, vanilla JavaScript. No frameworks. One file if possible.

Follow the design style of [paste a link to a site you love]. Match its layout, type scale, and color restraint.

Use this copy: [paste the headline, bullets, middle section, and CTA from the Copywriter].

Requirements:
- Mobile first, looks great on a phone
- Clean email capture form (use Formspree or similar)
- One CTA, repeated only at the top and bottom
- No stock photos. Use type and color only.

When it's ready, run a local server so I can preview.

Open Claude Code in an empty folder. Paste the prompt above. Claude Code will write the file, install what it needs, and start a preview server. Iterate in the same session.

Follow-up move: Open the preview on your actual phone. Whatever feels wrong, screenshot it and paste the screenshot back to Claude with one sentence. "The hero text wraps weird on iPhone. Fix it."

You know it's working when: You can submit your own email through the form, get the lead magnet, and the page loads in under two seconds on cellular. Polish comes later. Proof comes now.

Role 05. The Marketer.

Job: Turn attention into paying customers.

Most creators post randomly and hope something works. The Marketer builds a real plan. Channels. Cadence. Metrics.

This prompt only works if you give Claude your real constraints . Skip this and you'll get a plan written for a person who doesn't exist.

You're the Marketer now.

My constraints:
- I have [X] hours per week to spend on marketing
- My budget is [$Y] per month for ads or tools
- I have [Z] followers on [platform], starting from near zero on the others
- My offer is the one the Strategist landed on
- My audience is the one the Researcher validated

Build a 30, 60, 90 day plan that:
- Drives attention to the lead magnet
- Converts downloads into paying customers
- Fits inside my real time and money budget

For each 30-day block, give me:
- The one channel I should focus on (not five)
- The posting cadence
- The one metric that tells me it's working
- The signal to kill it and pivot.

Same Project. Claude has every other role's output. Don't make it guess.

Follow-up move: "Now write the first week of content. Specific posts, specific hooks, specific days." Generic plans never get executed. Specific plans get shipped.

You know it's working when: You can open the plan on Monday and know exactly what to post, when to post it, and what number you're checking on Friday. If you have to think about it, it's not specific enough.

Avoid these.

1. Using a chat instead of a Project. Context evaporates. By role four, Claude has forgotten the audience. The Project is the whole point.
2. Asking which AI is best before validating the offer. Claude vs ChatGPT vs Gemini is the wrong question. The tool isn't the problem. The direction is.
3. Skipping the Researcher. Building a perfect landing page for a product nobody wants is the most common way to waste three months.
4. Accepting Claude's first answer. The follow-up is where the real value lives. Every role above has one. Use it.
5. Polishing the Builder before testing demand. A live ugly page that captures emails beats a beautiful one that doesn't exist.
6. Running the Marketer without constraints. Vague budget, vague time, vague audience. You'll get a plan that looks impressive and works for no one.

The difference between the creator who fails and the one who builds something real isn't access to Claude. It's how they direct it.