Copywriting

Good copy reads like a smart person explaining something they care about. No one notices they're being persuaded — they just keep reading because it's useful and specific. Every technique in this skill exists to serve that feeling.

Outcome

Persuasive copy saved to projects/mkt-copywriting/{campaign-name}/, scored on 7 dimensions, with multiple variants for testing. Headline variants (5-10), body variants (2-3 for long-form), and A/B test suggestions included.

Context Needs

Load if they exist. Proceed without them if not — this skill works standalone.

Step 1: Load Context

Check brand_context/ and load per the table above. Show a brief status:

• Voice loaded: "[tone summary]. All copy will match."
• Positioning loaded: "Building around '[angle]' frame."
• ICP loaded: "Writing for [audience]. Awareness level: [level]."
• Nothing found: "No brand context yet. I'll write solid copy — we can build your brand profile anytime to make it on-brand."

Read context/learnings.md → ## mkt-copywriting section. Apply any previous corrections.

Step 2: Establish Format

Ask or infer from context:

State which format before generating.

Step 3: Check for Existing Work

If projects/mkt-copywriting/{campaign-name}/ exists, read what's there. Show a summary and ask: "Revise existing, add a new piece, or start fresh?"

Step 4: Pre-Generation Checklist

Before writing, confirm these (ask the user if unclear):

1. Audience awareness level — Schwartz 1-5 (read references/classic-frameworks.md for the full model). Determines headline approach and copy length.
2. Core transformation — Not features. Run the "So What?" chain three levels deep until you hit something emotional or financial. Read references/persuasion-toolkit.md § The So What? Chain.
3. Proof inventory — What testimonials, data, case studies are available?
4. Positioning angle — From brand_context or stated by user.

Step 5: Write the Copy

Read references/persuasion-toolkit.md and apply the techniques that fit the format:

• Headlines: Generate 5-10 using different angles (outcome, curiosity, proof, contrarian, story). Star your recommendation.
• Openings: Pick the entry style that best matches the audience's awareness level. Scene-setting for cold audiences, data or challenge for warm ones.
• Body: Keep the reader moving — vary sentence length, bridge sections, plant unanswered questions. Make pain concrete with numbers. Dig three levels past features to find the real benefit.
• Evidence: Every claim needs backup. Structure results as mini-stories: who they were, what they did, what changed, how long it took.
• CTAs: Name the reward, not the action. Below the button, reduce friction: remove risk, show peers, emphasise speed.

For landing/sales pages, use the full arc from the persuasion toolkit — hook through final CTA — adapting which sections to include based on format and audience awareness.

Step 6: Generate Variants

Read references/variants-and-scoring.md for the full protocol.

• Headlines: 5-10 variants across different frameworks. Lead with a QUICK PICK summary.
• Body copy (landing/sales pages): 2-3 variants — Control (strongest primary angle), Contrarian (counterintuitive entry), Proof-Led (evidence first).
• Email subject lines: 5-7 variants when writing emails.

Step 7: Score the Copy

Use the 7-dimension rubric from references/variants-and-scoring.md:

Clarity, Specificity, Voice, Desire, Proof, Urgency, Flow, each 1-10, total out of 70. Apply the thresholds table in references/variants-and-scoring.md.

Self-score generated copy honestly. If any dimension is below 7, note what would fix it and offer to revise.

When the user asks to "score this" or "review my copy" — score first, show weaknesses, then offer to rewrite.

Step 8: Humanizer Gate

Before saving, run all generated copy through tool-humanizer in pipeline mode. Use deep mode if brand_context/voice-profile.md was loaded, standard otherwise. Read tool-humanizer/SKILL.md and apply its pattern detection + replacement process. Only show the score summary if the delta exceeds 2 points.

Step 9: Save Output

Save to projects/mkt-copywriting/{campaign-name}/:

Include YAML frontmatter: type, campaign, awareness_level, variant, platform, word_count, date, score.

After saving, show the user actual copy excerpts — not just file paths.

Suggest A/B tests from references/variants-and-scoring.md § A/B Testing.

Step 10: Collect Feedback

Ask: "Does this sound like you? Which headline/variant direction resonates most?"

Log feedback to context/learnings.md → ## mkt-copywriting. If the user flags voice issues, note them so next run adapts.

Suggest logical next skills based on what was created (e.g., after landing page → email sequences, content repurposing).

Troubleshooting

• No brand context: Proceed standalone. Note what would make it better.
• No proof/testimonials available: Write placeholder sections marked [ADD TESTIMONIAL]. Suggest the user gather 2-3 specific results.
• User can't articulate the transformation: Run the So What? chain with them. Start with features, dig three levels.
• Copy sounds generic after generation: Check if ICP was loaded. If not, ask 2 questions about who they're writing for.
• Voice doesn't match: Re-read voice-profile.md and samples.md. Adjust sentence length, vocabulary, formality.

Rules

Updated automatically when the user flags issues. Read before every run.

• LinkedIn posts must NOT end with a question. Close with a strong declarative statement, a CTA, or a punchy fragment instead. This overrides any "end with a question" guidance in reference files. (2026-03-13)

Self-Update

If the user flags an issue with the output, add a dated entry under ## Rules in this skill's SKILL.local.md (create it if missing) and log it to context/learnings.md under ## mkt-copywriting. Write the general rule the correction implies, not the one-off fix.

Included AI-OS extension

Read references/ai-os-extension.md alongside this file. It contains the shipped supplementary methodology and integration rules. User corrections belong in a new SKILL.local.md; they take precedence over the shipped files.