如何使用 Claude 進行聯盟行銷：Gabi Skill 指南

For the outreach, the programme terms written in plain language, the content structure and the briefs, yes. Not for product facts. Ask it to compare five products and it will produce a clean table of prices, specifications and feature support with some cells confidently wrong, which in affiliate content is a misleading commercial claim about something you are paid to sell. The working method is that the model supplies the structure and the questions and you supply the facts.

The disclosure has to be clear and conspicuous, which means hard to miss. It sits with the recommendation rather than in a footer, it is not hidden behind a more link, it is not only in a profile bio or a pinned comment, and for video it appears in the video itself, in the audio and on screen, long enough to read, not only in the description. A platform paid-partnership tag helps but is not on its own the whole answer, and the disclosure should be in the same language as the content. This is operational guidance, not legal advice; confirm the rules for your own market.

Two instruments matter in the US. The FTC Endorsement Guides, 16 CFR Part 255, revised in 2023, say a material connection between endorser and seller must be disclosed clearly and conspicuously, and an affiliate commission is a material connection. The Rule on the Use of Consumer Reviews and Testimonials, 16 CFR Part 465, which took effect in October 2024, is a rule rather than guidance and violations can carry civil penalties. It prohibits fake or AI-generated reviews and testimonials, buying positive or negative reviews, undisclosed insider reviews, suppressing negative reviews to create a misleading picture, and buying fake indicators of social media influence.

Writing a testimonial from a customer who does not exist is specifically prohibited conduct under the FTC reviews rule, and a model will produce one on request without objecting. It is also the fastest way to destroy a channel whose whole value is that people trust the recommendation. An AI chat can help structure and tighten a review you genuinely wrote from your own experience; it cannot supply the experience.

On six clauses. Cookie window: the brand wants it short, the partner long. Attribution: last click is cheap for the brand and never credits the content that started the search. Brand bidding: the brand usually bans it, because paying commission on its own brand-name traffic is paying twice. Coupon and browser-extension partners: often excluded or on a lower rate, because they attach at checkout to a sale already happening. Returns: the brand claws commission back on refunds. And recurring versus one-off commission. Agreeing these vaguely in a rush is what turns a programme into a series of disputes.

Explain what is in it for them in the first two lines, in terms of their audience rather than your product, include the commission terms as a fact rather than a tease, keep it under about 120 words with one ask and one link, and skip the flattering opening line. The specific trap with AI here is invented personalisation: asked to personalise, a model will describe a video they never made or cite a subscriber count it does not know. That does more damage than a generic email, because it proves you did not look. Instruct it to use only facts you paste and to mark anything missing as VERIFY.

The commission rate is the least important number and the only one most people look at. Three questions decide it: does the product genuinely serve the people who trust you, does the traffic you can send actually convert, and can you verify the claims you would be repeating on the brand's behalf. A useful test is to write the honest review you would publish if there were no commission at all. If it comes out thin, the programme is wrong for you regardless of the rate.

The download is a ZIP with SKILL.md at the root of the archive, not inside a nested folder. In the Claude desktop app open Customize then Skills, upload the ZIP and toggle it on. Skills are available on the Pro, Max, Team and Enterprise plans with code execution enabled: on Pro and Max you turn on Code execution and file creation under Settings then Capabilities, while on Team and Enterprise that switch sits with an owner in organisation settings. In ChatGPT or Gemini, paste the contents of SKILL.md into custom instructions instead.