카피라이터로 Claude 사용하는 방법: Olivia Skill 가이드

It can produce a fast, competent draft, and competent is the problem. The default output is copy a competitor could honestly sign: fine sentences that cost a reader nothing to skip. Conversion comes from specifics the model does not have, which is your proof, your customer's actual objection and the one thing you can say that a rival cannot. Supply those in a brief and the draft is genuinely useful. Skip them and you get well-written filler at speed.

Two moves, both concrete. Ban the default vocabulary by name in the prompt: transform, elevate, unlock, seamless, effortless, revolutionary, game-changing, robust, leverage, empower, cutting-edge, plus every intensifier that does not change the meaning of its sentence. Then give it a voice card extracted from three things you actually wrote, rather than describing a tone. Friendly and professional is a request for the average of the internet, which is exactly the sound you are trying to avoid.

Take any sentence and imagine your closest competitor's logo above it. If they could honestly put their name on it, the sentence says nothing. Built for teams who demand more. Quality you can trust. All fine, all signable by anyone. The same test in another form: if the opposite of your claim is absurd, it is not a claim, because nobody advertises unreliable service. A line is doing work only when a reasonable competitor could not honestly say it.

Yes, if you ask for persuasive copy without giving it evidence. It will produce plausible percentages, confident numbers and quotes from customers who do not exist. Published as marketing those are misleading commercial claims, and fabricating a testimonial is specifically prohibited conduct under the FTC's rule on consumer reviews and testimonials. The fix is one line in the drafting prompt: if a sentence needs a fact you have not supplied, it writes PROOF NEEDED and never fills the gap itself.

It will write them cheerfully, which is the risk. Superlatives, comparative claims, guarantees, up-to figures and anything touching health, finance or income are regulated ground in most markets, and the rules differ by country. Treat those lines as something a qualified person approves rather than something a model drafts. The audit prompt in this guide flags them separately for exactly that reason.

Fewer than you think, and differing on the angle rather than the wording. Ask for five variants with no further instruction and you get one sentence phrased five ways, which tests synonyms and teaches you nothing. Name the leads instead: the outcome, the problem avoided, the objection answered, the proof point first, the comparison against what they use now, the cost of doing nothing. Then ask which one to test first and what result would prove the angle wrong rather than the wording.

No. A conversation that just produced the draft defends it, and asking whether the copy is good gets you a yes. Paste the copy into a fresh conversation with no context about who wrote it, and ask for quotes rather than a verdict: every sentence a competitor could sign, every claim with no evidence attached, every deletable intensifier, and the calls to action counted.

The download is a ZIP with SKILL.md at the root of the archive rather than inside a nested folder, which is the usual reason an upload fails. In the Claude desktop app open Customize, then Skills, upload the ZIP and toggle it on. Skills need code execution enabled, under Settings then Capabilities. Anthropic's help centre currently lists Skills on Free, Pro, Max, Team and Enterprise, while its Academy tutorial lists Pro, Max, Team and Enterprise, so if you are on the free plan check Settings then Capabilities for your own account rather than trusting either page. In ChatGPT or Gemini there is no upload step, so open SKILL.md, copy the contents and paste them into custom instructions.