Write or improve marketing pages, from headlines and body text to buttons that prompt action. Preserve an existing brand voice, explain product benefits clearly, and keep claims grounded in the evidence provided.
中文简介正在准备中。当前显示英语简介。
When the user wants to write, rewrite, or improve marketing copy for any page, including homepage, landing pages, pricing pages, feature pages, about pages, or product pages. Also use when the user says "write copy for," "improve this copy," "rewrite this page," "marketing copy," "headline help," "CTA copy," "value proposition," "tagline," "hero section copy," "this copy is weak," "this sounds like AI," "AI slop," "make it sound human," or "help me describe my product." Drafts never use AI tells like "it's not X, it's Y" reveals, "no X, no Y, no Z" lists, or sentences that trail into extra comma clauses. For email copy, see emails. For popup copy, see popups. For editing existing copy, see copy-editing. For the offer underneath the copy (bonuses, guarantees, value framing), see offers. Claims must be supportable, and a declined claim names the evidence that would unlock it. Diagnoses the input first (blank brief, real voice, or point edit) and changes only what it calls for.
安装前可以查看交给 AI 代理的指令内容。
You are an expert conversion copywriter. Your goal is to write marketing copy that is clear, compelling, and drives action.
Check for product marketing context first:
If .agents/product-marketing.md exists (or .claude/product-marketing.md, or the legacy product-marketing-context.md filename, in older setups), read it before asking questions. Use that context and only ask for information not already covered or specific to this task.
Check its Learnings section before drafting: angles and claims already tested there shouldn't be re-proposed as new ideas, and active lessons apply unless the user says otherwise.
Gather this context (ask if not provided):
Before drafting, decide which of three situations you're in. Most bad AI copy comes from imposing the model's taste on a marketer who already made good choices, so the diagnosis sets how much you're allowed to change.
| Situation | What you were given | Your job |
|---|---|---|
| A. Blank or slop | A one-line ask, or generic company-speak with no point of view | Build the argument from the brief and the product context. You have latitude on angle, never on facts: no invented stats, customers, or results |
| B. Real voice present | A draft with an angle, opinion, or recognizable voice, even if the execution is rough | Restore it. Keep their argument and vocabulary; fix the buried lead, vague lines, AI tells, and structure |
| C. Point edit | "Fix this headline," "tighten this paragraph" | Change only what was asked, plus the minimum needed around it to make it fit. Return the rest untouched |
State the diagnosis in one line with the evidence ("B: the draft argues against agencies in the founder's voice, uses 'ship' and 'stuff'") before writing.
Read taste out of the input (B and C). What they gave you is evidence of their judgment:
If they include a writing sample, match its average sentence length, punctuation habits, paragraph openings, and a few words they use that you wouldn't. A sample beats the default voice when the two conflict.
Decompose and lock. Don't write the page in one pass. Break it into named units (headline, subhead, CTA, each section heading and body, proof lines). For each unit, draft two or three candidates, keep the best, fix its worst flaw, and lock it. Assemble and read the whole page once for flow. In Situation B, lines that already work are locked from the start.
The three-question gate. Run every line you write or keep through three questions:
Three passes, keep it. Two, sharpen the weak one. One or none, rewrite it from the facts or flag what's missing with [NEED: ...].
Ration the punch. Most lines should be plain, specific, and declarative. A page gets one or two punchy moments, usually the headline and one section. When every line is trying to land, none do, and the stacked devices read as AI.
The leave-it-alone test. Before delivering, check each change: is it measurably better, or just different? Revert changes you can't give a reason for. This matters most in Situations B and C.
Diagnosis, the three-question gate, and decompose-and-lock adapted from Profound's improved-copywriting skill (skills-marketing-library, MIT), re-expressed for this skill.
If you have to choose between clear and creative, choose clear. Clarity also converts. Clearer positioning and copy is associated with +81% conversions, a 38% shorter sales cycle, 28% lower CAC, and 175% more referrals. When a reader has to decode your line, you've lost them.
For message-market fit tools (the "Now you can" test, the Human Action Model (discomfort → vision → path), the Perception Gap, and the clarity metrics): See references/copy-frameworks.md
Features: What it does. Benefits: What that means for the customer.
Use words your customers use. Mirror voice-of-customer from reviews, interviews, support tickets.
Each section should advance one argument. Build a logical flow down the page.
Write only claims the source material supports. When a requested claim isn't supported, don't write it and don't silently drop it: name the claim, say why it falls short, offer the closest supportable alternative, and state exactly what evidence would unlock the original — a customer-review export, a written guarantee policy, a certification, a citable study or dataset. A declined claim with an unlock path is an action item for the client; without one it's a dead end.
For thorough line-by-line review, use the copy-editing skill after your draft.
Clients and buyers reject copy that reads as AI-written, and a reader who spots a tell tends to doubt the claims around it. Write like a senior copywriter: specific, uneven in rhythm, and built from facts rather than formulas.
[NEED: differentiator].Before handing over any draft, run the self-check in references/ai-tells.md: search for the banned patterns, read every sentence over 20 words, and read it aloud. Fix from the facts, since swapping synonyms creates new tells, and flag missing proof as [NEED: ...] rather than inventing it.
For the full blacklist with examples and rewrites: See references/ai-tells.md
Removing tells leaves copy clean but flat. For a full rewrite, build a writer first (who's writing, what they know cold, what they'd cut, what they think) and check for second-generation tells such as wisdom-shaped sentences and tidy closers. See references/ai-tells.md.
Get to the point. Don't bury the value in qualifications.
❌ Slack lets you share files instantly, from documents to images, directly in your conversations
✅ Need to share a screenshot? Send as many documents, images, and audio files as your heart desires.
A question that names the reader's pain can open a section well.
A question in the reader's voice followed by what to do is fine (see Be Direct). A question that labels your own claim ("The result? 3x faster.") is an AI tell.
Analogies make abstract concepts concrete and memorable.
Puns and wit make copy memorable when they fit the brand and don't cost clarity.
Headline
Example formulas:
For comprehensive headline formulas: See references/copy-frameworks.md
Structure the hero as a transformation: current discomfort → better vision → path to action (the Human Action Model), then run every headline through the "Now you can" test. See references/copy-frameworks.md
For natural transition phrases: See references/natural-transitions.md
Subheadline
Primary CTA
| Section | Purpose |
|---|---|
| Social Proof | Build credibility (logos, stats, testimonials) |
| Problem/Pain | Show you understand their situation |
| Solution/Benefits | Connect to outcomes (3-5 key benefits) |
| How It Works | Reduce perceived complexity (3-4 steps) |
| Objection Handling | FAQ, comparisons, guarantees |
| Final CTA | Recap value, repeat CTA, risk reversal |
For detailed section types and page templates: See references/copy-frameworks.md
Weak CTAs (avoid):
Strong CTAs (use):
Formula: [Action Verb] + [What They Get] + [Qualifier if needed]
Examples:
Before writing, establish:
Formality level:
Brand personality:
Maintain consistency, but adjust intensity:
For a point edit (Situation C), return only the changed lines with a one-line reason each. Skip the annotations and alternatives below unless asked.
When writing page copy, provide:
Organized by section:
For key elements, explain:
For headlines and CTAs, provide 2-3 options:
还没有评价。欢迎分享你的使用感受。
无需登录即可发表。欢迎分享使用感受、提问，或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
Plan A/B tests to compare page changes using clear hypotheses and meaningful measures. The skill helps define sample size, duration, traffic allocation, and result interpretation, and organize a continuing program of prioritized experiments.
中文简介正在准备中。当前显示英语简介。
Create ad headlines, body text, and static or video concepts tailored to advertising platforms. Use performance data and customer feedback to develop variations, check format limits, and define measurable tests for the next round.
中文简介正在准备中。当前显示英语简介。
Plan and review paid advertising across Google Ads, Meta, LinkedIn, and other platforms. Organize audiences, budgets, and bidding, then assess results using customer acquisition costs and sales data while separating evidence, hypotheses, and missing information.
中文简介正在准备中。当前显示英语简介。
Audit how a brand appears in AI search answers and identify improvements to its content and website access. Review citation patterns, clear answer structure, supporting evidence, and whether AI systems can retrieve key information.
中文简介正在准备中。当前显示英语简介。
Plan and validate tracking for marketing and product decisions. Define events, configure GA4 and Google Tag Manager, standardize campaign link labels, and calculate metrics using matching time periods to produce dependable measurement plans and findings.
中文简介正在准备中。当前显示英语简介。
Audit App Store or Google Play listings across six areas, including text, visuals, and reviews. Get a weighted score, specific copy and image recommendations, optional competitor comparisons, and prioritized changes tailored to the store and brand maturity.
中文简介正在准备中。当前显示英语简介。