The single highest-leverage Claude setup move I've found. Fifteen minutes once, hours back every week.
Most marketing teams have heard of Claude Projects. Few have set one up properly. The ones who have are the ones whose Claude outputs sound like their brand. The ones who haven't are still pasting "we're a B2B SaaS company that values customer-centricity" at the start of every conversation.
This post is the exact setup I use. Steal it. The whole thing is ten minutes once.
If you're re-pasting your brand context at the start of every Claude session, you're wasting roughly 10-15 minutes per marketer per day. For a team of six, that's 60-90 minutes daily. Annualised: ~250-375 hours per year spent re-typing context that should live in a Project.
The setup below takes ten minutes once. The payback is immediate and compounds.
A Project is a persistent workspace. Two things live in it:
1. Instructions: the system prompt. This is the brand context that every conversation in the Project inherits. Voice, KPIs, tools, non-goals.
2. Knowledge, files Claude can reference. PDFs, docs, transcripts, screenshots. Pin your style guide, your last positioning doc, your customer-interview synthesis, your top-performing emails.
Every conversation you start inside the Project starts with all of that loaded. Outside the Project, you're back to generic Claude.
The most important section. Don't write adjectives ("warm, confident, direct"). Adjectives don't transfer. Instead, drop in the output of the Voice & Tone Extractor, which gives you three voice attributes with their opposites, sentence rhythm patterns, and on-brand vs. off-brand pairs. That's what Claude can actually pattern-match against.
If you haven't run the Voice Extractor yet, do that before this Project setup. Genuinely. It's the single highest-leverage thing.
What are you actually trying to move? Pipeline? Activation rate? Trial-to-paid? Newsletter subs? Be specific. Include the current baseline and the target. Claude will use these to weight recommendations, if you say "we care about activation rate, currently 31%, target 45%," Claude will recommend campaigns differently than if you say "growth."
What's in your actual stack? Webflow + Klaviyo + HubSpot + Mixpanel + Notion is a different Claude context than Squarespace + Mailchimp + GA4 + Trello. Claude can recommend implementation paths that match what you've got, but only if you tell it.
This is the section most teams skip. List what you're not going to do. "We don't do paid social. We don't do influencer marketing. We don't do gated content beyond a single landing page test per quarter." Claude will stop recommending those things, and your outputs will stop drifting toward generic best-practice advice that doesn't fit your operation.
Agencies and multi-product companies hit this constantly. The answer is: one Project per brand. Don't try to pack three brand contexts into one Project. Claude will average them and the output gets generic.
If you're running three brands, that's three Projects, each with its own voice, KPIs, tools, and non-goals. Yes, you'll duplicate some setup. The clarity is worth it.
The exception: a holding-company-level Project for cross-brand work (M&A research, category strategy, talent decisions). That one gets a different shape, less voice, more strategic positioning.
Projects rot. Three months in, your voice has evolved, your KPIs have shifted, your tools have changed, and the Project still reflects the world from before. Outputs start getting subtly wrong in ways that take you a beat to spot.
Fix: every quarter, run a "Project audit." Re-run the Voice Extractor on your latest 5 pieces. Update the KPIs. Add anything you've stopped doing to non-goals. It takes 20 minutes per Project. Do it on the same day you do quarterly planning.
Projects aren't always the answer. Don't use one when:
That's the whole setup. Ten minutes if you've already run the Voice Extractor. Thirty minutes if you haven't. The payback is immediate.
Run it once on 3-5 of your best pieces, paste the output into your Project's Instructions, and every prompt from then on inherits your voice. It's free, no email gate.
Run the Voice Extractor →