Use referrals when people already recommend the product but sharing, rewards, or tracking are fragmented. It separates customer-to-friend recommendations from ongoing creator promotion and connects sharing, qualifying actions, and reward delivery into a complete journey.
Install and invoke
Run this in your project terminal to install only this skill for Claude Code:
npx skills add coreyhaines31/marketingskills --skill referrals -a claude-code
Node.js and npx must be available. For Codex, replace -a claude-code with -a codex. In the agent conversation, explicitly ask it to use referrals and supply your material. The installation command belongs in the terminal, not in your content brief.
Identify who recommends to whom
Clarify whether referrers are customers, partners, or creators, and what the recipient must do to qualify. Inspect natural collaboration or sharing first. An internal infrastructure tool may not suit a public watermark, while a collaborative product may benefit from improving invitations.
Example brief:
Use referrals to design a two-sided pilot for current paying customers.
Read context.md for product economics and budget.csv for the monthly reward ceiling.
Reward only after a new customer completes a qualifying first payment. Specify timing, sharing copy, attribution, and support explanations.
Describe refunds, duplicate accounts, and self-referrals. Do not purchase or activate a platform.
Customer referrals usually rely on an existing usage relationship. Affiliates may have an ongoing commission arrangement with different review and support needs. Shared links do not make the programs interchangeable. Ask for recommendations after users experience value rather than immediately after registration.
Make reward rules testable
Specify eligibility, each party’s benefit, timing, and exclusions. One-sided, two-sided, and tiered incentives must fit actual economics. Do not import generic growth figures as expected performance. Explain cash and discount value clearly instead of using ambiguous framing.
Map the whole journey: see the prompt, share, visit, register, complete the qualifying action, and record the reward. Observable events distinguish low sharing from poor recipient conversion. Attribution and commission services are implementation dependencies; installing the skill does not create links, send invitations, or make payments.
Test the normal route with test accounts, then inspect duplicate invitations, different devices, refunds, and delayed rewards. Begin with a bounded pilot and record actual cost and customer quality. If clicks are plentiful but meaningful use is weak, investigate audience fit and the landing experience before increasing rewards. Longer-term assessment requires actual retention and customer-value evidence.
Define both the referral moment and reward fulfillment
Look for moments when existing customers have already received value: finishing a first meaningful task, receiving exceptional support, renewing, or upgrading. A share request then is more natural than one at registration. Ask the skill to draw the complete path: where the current customer sees the invitation, whom they can share with, the new visitor’s landing page, the action that counts, and when each party receives what. If duplicate or self-referrals cannot be distinguished, avoid a complex reward program for now.
Separate customer referrals from affiliate partnerships. The former use an existing product relationship; the latter may involve publishers, commission attribution, and disclosure. Supply acquisition cost, customer value, margin, reward budget, and current channels. An example brief: “Design a first invitation flow for paid customers who have used the product for four weeks. Use the ceiling in budget.md. Start with the trigger, invitation page, valid-referral definition, fraud checks, and support rules. Do not promise viral growth.”
Before launch, run three test paths: a normal invitation, duplicate registration, and a visitor who does not qualify. Check reward reversals, refunds, and whether both sides understand the terms. Track valid referrals, subsequent use by invited customers, and reward cost together. Many shares without real new customers are not a successful referral loop.
Example: separate “invite a teammate” from paid referral
A collaboration product already has an “invite teammate” button and the team is considering a cash reward. Give the skill the current invitation flow, billing rules, and whether multiple people in one organization can buy. Ask: “Use referrals to distinguish a collaboration invitation from a rewarded customer referral. Draw both paths, identify which step is ordinary product use and which creates a genuinely new paying customer, and list budget and attribution questions. Do not write promotional copy yet.”
Check whether a teammate joining an existing project is incorrectly counted as a new customer, whether buyer and user may differ, and whether the reward would encourage duplicate accounts. A proposal to pay for every click needs a cost and fraud review. Define a valid new customer and refund handling before writing the sharing entry point or public terms; otherwise a convenient invite flow can create billing disputes.
What referral examples and adaptations can teach
Newsletter creators Brennan Dunn used tiered rewards and a progress bar, while Aleyda Solis set different milestones for ordinary readers and larger influencers. These are referral-program examples, not performance evidence for this Skill. Borrow the idea only after defining reward thresholds, fulfillment timing, and eligibility in your own flow, and test whether people want to share with a small audience first.
For retail or ecommerce, a SaaS “invite a teammate” loop may not fit. Sush Padhye's independent fork adapts the Skill around gifting, unboxing UGC, compensated ambassador disclosure and attribution, while warning against paying for purchases that would have happened anyway. The public fork is a workflow design, not evidence of a merchant's revenue lift. Define qualified new buyers, refunds and repeat purchases before considering a small holdout.