Ad copy ab generator

Ad Copy A/B Generator

A skill for generating two strategically distinct RSA variations using proven copywriting frameworks.

What this skill does

For a given ad group, this skill produces:

• Version A - Direct/rational angle (features, specifics, numbers, proof)
• Version B - Emotional/aspirational angle (outcomes, identity, transformation)

Each version contains:

• 15 headlines (≤30 chars each), distributed across 5 categories
• 4 descriptions (≤90 chars each), each with a different focus
• Pinning recommendations
• A hypothesis explaining why each variation should win
• Suggested test duration

Required inputs

1. Product/service description - what's being sold
2. Target audience - who buys it
3. Key differentiators - what makes this offering different
4. Top 3–5 keywords for the ad group - so headlines can include them
5. Current best-performing ad headlines (if available) - to learn what's already working
6. Competitor ads (if available) - screenshot text or pasted ads helpful for differentiation

If any of (1)–(4) is missing, ASK before generating. Don't guess.

Workflow

Step 1: Gather context

If the user just said "write me ads," respond:

Before I write, I need:

1. What you sell - one or two sentences
2. Who buys it - target customer profile
3. What makes you different - 2–3 specific differentiators (not "we're better")
4. 3–5 keywords for this ad group - so I can include them in headlines
5. Current best-performing ads (optional) - paste headlines you know are working
6. Competitor ads (optional) - paste 2–3 competitor RSAs you've seen for these keywords

Step 2: Generate Version A - Direct/Rational

The structure for each RSA variation: 15 headlines split across 5 categories, 4 descriptions each with a distinct focus. See references/copy-frameworks.md for the framework details.

Version A leans on: features, numbers, social proof, specific guarantees, tangible benefits.

15 headlines distribution:

• 3 with primary keyword insertion (literal keyword in the headline, e.g., "Buy CRM Software" if keyword is "buy crm software")
• 3 with unique value propositions (specific, differentiated)
• 3 with social proof / numbers ("Trusted by 10,000+ Teams", "4.8★ on G2", "$50M Saved")
• 3 with urgency / CTA ("Start Free Trial", "Get a Demo Today", "See Pricing Now")
• 3 with benefit-focused messaging (the literal benefit, not the feature)

4 descriptions:

• 1 feature-focused
• 1 benefit-focused
• 1 social-proof focused
• 1 CTA-focused

Step 3: Generate Version B - Emotional/Aspirational

Same structure but tone shifts to identity, outcomes, transformation. Less about "what it does" and more about "who you become / what changes."

For B2B: emotional angles include relief from frustration, professional pride, peer recognition. For B2C: include identity, lifestyle, status, security, belonging.

Step 4: Pinning recommendations

Don't pin everything - Google's RSA learning works best when it can rotate freely. Default pinning approach:

• Pin one strong headline to position 1 (highest visibility): pick the one with primary keyword insertion + clear value prop. Pin to ensure consistency across impressions.
• Pin one strong headline to position 2 (still visible): pick the strongest social proof or differentiator headline.
• Leave position 3 and beyond unpinned for Google to optimize.

For descriptions: pin the strongest CTA-focused description to position 1; leave position 2 unpinned.

Step 5: Hypothesis and test guidance

For each variation, write a 2–3 sentence hypothesis:

Version A should outperform with high-intent searchers who want specifics - pricing, features, proof. The numbers and CTAs reduce friction for ready-to-buy clicks.

Version B should outperform with mid-funnel searchers who haven't decided yet - the aspirational angle creates consideration intent.

Suggest test duration based on traffic volume:

Ad group monthly clicks	Test duration
1,000+	2 weeks
300–1,000	4 weeks
100–300	6 weeks
<100	8+ weeks (or skip A/B; combine into one ad)

If the ad group sees fewer than 100 clicks/month, recommend ONE strong RSA instead of A/B. Tell the user: "Below 100 clicks/month, A/B test won't reach significance in a reasonable time."

Output format

Markdown response with this structure:

## Ad Copy A/B Test - Ad Group: [name]

### Version A - Direct/Rational

**Headlines:** (each on its own line, with character count and pinning)
1. [Headline A1] (28) - PIN POS 1
2. [Headline A2] (29) - PIN POS 2
3. [Headline A3] (25)
... (15 total)

**Descriptions:**
1. [Desc A1] (89) - PIN POS 1
2. [Desc A2] (87)
...

**Hypothesis:** ...
**Test duration:** ...

### Version B - Emotional/Aspirational
[same structure]

### Implementation notes
- Set up both ads in the same ad group with rotation set to "Optimize"
- Don't change the ads during the test period
- Compare CTR, conv. rate, and CPA at the end of the test window

If the user requests a "deliverable" or "spreadsheet I can give my team," generate an Excel workbook (xlsx skill) with separate tabs for each variation.

What this skill must NOT do

• Don't exceed character limits. 30 chars/headline, 90 chars/description. Count exactly. If a headline is 31 chars, rewrite it.
• Don't generate generic copy. "Best [Product]" is a placeholder, not a headline. Every headline should include something specific to the user's offering.
• Don't pin every headline. Over-pinning kills RSA performance.
• Don't include emojis unless the user is in a B2C-friendly vertical and explicitly OK with them.
• Don't promise "this will increase your CTR by X%" - copywriting outcomes are unpredictable.
• Don't write ads that promise things the user didn't say they offer. "Free shipping" isn't a freebie unless they said they offer it.
• Don't include competitor brand names in copy (trademark risk).

Reference files

• references/copy-frameworks.md - PAS, AIDA, social proof, and the RSA category breakdown
• references/character-count-rules.md - exact rules for what counts as a character + style notes