Stack of folders representing Claude Projects workspace for writing and drafting client documents

Claude Projects: 5 Solid Wins From a Solo Consultant Stack

by

After six months of running Claude Projects as the spine of my client work, the pattern is clear enough to put on paper. Claude Projects sit on the paid claude.ai tier ($20/month) and act as persistent workspaces — each one holds a system prompt, uploaded files, and a 200,000-token context that survives every conversation inside it. For a solo consultant juggling four to six clients in parallel, that persistence is the whole game. This is what actually paid off, where Claude Projects still wastes my time, and the small setup rules I’d hand a new solo user before they touch the + New Project button.

In this article

My Claude Projects setup, six months in

The current Claude Projects shape on my account is six active projects: one per retainer client (three of them), one cross-client “writing voice” project, one research project for new pitches, and one personal weekly-review project. Each project carries between 8 and 22 uploaded files — proposals, transcripts, pricing docs, brand notes — and a system prompt that names the client, the engagement scope, and the voice rules I want Claude to follow.

The 200K context window per project is the constraint I plan around. With RAG on the paid tier, I can load more files than fit in context and Claude retrieves what it needs, but I still trim aggressively: anything older than 90 days for retainer projects, anything outside the active scope for research. Uploaded files cap at 30MB each, and I’ve never come close — proposals are PDFs under 2MB.

What changed at the six-month mark was discipline. Early on I’d dump everything into a project and trust retrieval. Now I treat the project like a folder I’d hand to a thoughtful junior strategist — only what’s relevant, labeled clearly. Claude Projects rewards that hygiene more than the tool itself advertises.

Win 1: Per-client Claude Projects cut the briefing tax to near zero

The briefing tax is the 8–12 minutes I used to spend at the start of every chat re-explaining who the client is, what we’re working on, and which constraints apply. With per-client Claude Projects, that tax disappears. I open the project, type the request, and the first draft already knows the client’s positioning, the rejected directions from previous rounds, and the voice rules.

For one B2B SaaS founder I work with, the project holds the brand voice doc, the last three months of LinkedIn posts I’ve ghostwritten, and a one-page “what we don’t say” list. When she pings me Tuesday morning with a new post angle, I’m drafting within 90 seconds. The same request in a fresh ChatGPT window would burn 10 minutes on context, and the draft would still need a voice pass.

The compound effect is what justifies the subscription on this client alone. Eight to ten briefings a week across three retainer clients used to cost roughly 80 minutes of context-loading every week. Claude Projects compressed that to under 10 minutes — the small overhead of opening the right project and tweaking the system prompt when scope shifts.

Win 2: Long-form drafts that hold past 2,500 words

The single use case where Claude Projects clearly outperforms a fresh chat is long-form drafting past 2,500 words. Strategy documents, board memos, and proposal narratives over that length tend to drift in a vanilla chat — voice loosens, a key positioning phrase slips, an earlier section’s framing contradicts a later one. Inside a project that already holds the brand voice doc and the working outline, the draft stays coherent.

Last month I drafted a 4,200-word repositioning memo for a branding agency director. I ran it as three sequential turns inside the project — outline, full draft, voice pass — and the final version needed roughly 45 minutes of editing instead of the 90+ I’d have spent on the same document in a context-free chat. The agency’s voice phrases showed up correctly without my having to prompt for them. That’s the project’s system prompt and uploaded voice doc earning their keep.

The catch: Claude Projects don’t fix bad source material. If the brand voice doc is vague, the long draft inherits that vagueness. I’ve started writing one-page “voice cards” specifically to feed Claude Projects, separate from the marketing-facing brand decks. Those cards are blunt and rule-shaped, and they outperform polished brand documents by a wide margin.

Win 3: Research briefs anchored to client positioning

Research is where Claude Projects quietly changes the shape of the work. My pitch research used to live in a Perplexity tab and a scratch doc. Now it lives in a research-only Claude Project that holds my standard pitch framework, the prospect’s public materials, and notes from any intro calls. When I drop a fresh URL into the chat and ask Claude to extract competitive signals, the answer comes back already mapped to the framework I use — not a generic summary.

The honest constraint: Claude doesn’t browse the open web from inside a Project the way Perplexity does. I still run the source-citation pass in Perplexity, then paste the cleaned-up notes back into the Claude Project for synthesis. The split is roughly: Perplexity for “what’s true on the public web,” Claude Projects for “what does that mean for this specific prospect’s pitch.” Each does its job better than the other could.

For deeper context on the research half of that split, my Perplexity vs Google for Research note from earlier this year covers how source-cited search compresses pitch prep — the upstream step before a research Claude Project ever opens.

Win 4 & 5: Strategy iteration and a Friday review ritual

The fourth win is iteration. Strategy docs almost always need a second and third pass after stakeholder review, and the project context survives every round. When a client comes back with “what if we lead with the integrations angle instead?” I don’t reload the prior decisions, the rejected directions, or the voice constraints — Claude Projects already has them. Iteration that used to cost 30–40 minutes of context resets now costs 8–10 minutes of focused editing.

“Iteration that used to cost 30–40 minutes of context resets now costs 8–10 minutes of focused editing — the project context survives every round, including the rejected ideas.”

The fifth win is the personal weekly-review project. Every Friday I dump the week’s client notes, time-tracked tasks, and any half-finished drafts into one project and ask Claude to surface three threads: what compounded, what stalled, and what to drop next week. The system prompt names my three retainer clients by code and tells Claude to be blunt about scope creep. I read the output in fifteen minutes on a Friday afternoon, and roughly half the items become the next week’s plan. It’s the lightest version of an operations review I’ve found that I actually keep doing.

Where Claude Projects still falls short

Three honest limits before anyone signs up expecting too much.

First, the 200K context window sounds generous but fills faster than expected for clients with deep history. After eight months of work with one client, I had to split their project into “current-quarter” and “archive” because retrieval started missing recent decisions. Two projects per client is now my default for any engagement over six months old.

Second, there’s no native version history on uploaded files. If I update the brand voice doc, I need to remove the old one manually or Claude will pull from both. I’ve been burned twice by stale uploads contradicting a fresh strategy. A small “last updated” check before any high-stakes draft is now part of my workflow.

Third, Claude Projects don’t share state across teammates the way Notion or Slack threads do. For a solo operator this is fine, but the moment a client wants a shared workspace, Projects stops being the answer. That’s where my Notion AI vs ChatGPT for Client Briefs split takes over — different tool, different job.

For me, Claude Projects earns its slot on the strength of three patterns: per-client context that survives every chat, long-form drafts that hold their thread, and a research-to-strategy handoff that no single tool covers alone. Six months in, the question isn’t whether to keep using it — it’s how many client engagements I can run in parallel before the projects-per-client math gets unwieldy. That’s a healthier ceiling than most $20/month tools ever hit.

FAQ

Is Claude Projects worth the $20 Pro tier for a solo consultant?

Yes, if at least one of three things is true: you write long-form documents over 2,000 words for repeat clients, you juggle two or more retainer engagements at once, or you’ve felt the briefing tax of re-explaining context to a fresh chat more than three times a week. Below that volume, the free tier with manual file uploads usually clears the bar.

Can I use Claude Projects with sensitive client data?

It depends on the client’s data agreement. Claude’s Pro plan does not train on user inputs by default and offers data controls, but I still keep PII-heavy work (names, financial records, identifiable HR notes) off the platform and route it through a local LLM instead. For drafts, positioning, and de-identified research, the platform fits a solo consultant’s risk profile.

How many Claude Projects should I run at once?

Not yet a hard rule, but my working number is one per retainer client plus two cross-client projects (research, personal review). Six total is the most I’ve managed without losing track. Above that, the cognitive load of “which project holds what” starts to eat the briefing-tax savings.

Does Claude Projects work for code or technical writing?

Yes, but I keep code work in Claude Code or Cursor instead. Projects on claude.ai are stronger for prose and strategy than for repository-aware coding sessions. For a side-by-side on the coding split, my Claude Code vs Cursor piece covers the call I made for daily-driver work.

What’s the single setup rule that mattered most for me?

It depends, but one rule stood out: write the system prompt as if a thoughtful new freelancer was reading it on day one. Name the client, the scope, the voice constraints, and the three things to never say. That clarity costs ten minutes once and pays back on every chat inside the project.

Sources

AI-assisted research and drafting. Reviewed and published by ToolMint.