/ads-multi — Multilingual RSA Generation

Input Required

Ask the user for the following (skip any already provided):

1. Campaign name — for CSV organization
2. Ad group name — base name (language suffix added automatically, e.g., brand_awareness_DE)
3. Target keywords per language — e.g., EN: "marketing analytics", DE: "Marketing-Analyse", FR: "analyse marketing", ES: "análisis de marketing", AR: "تحليلات التسويق", PL: "analityka marketingowa"
4. Landing page — a registry key (e.g., pricing) from .claude/skills/landing-pages/SKILL.md. Claude automatically resolves the correct language variant per target language. Or provide URLs per language manually.
5. Languages — which languages to generate (e.g., EN, DE, FR, ES, AR, PL)
6. Key USPs / offers — describe once; Claude adapts per language
7. Target persona — which persona from .claude/skills/icp/SKILL.md to target
8. Output format — ask the user:

• Separate CSVs: one file per language (rsa_campaign_adgroup_EN.csv, rsa_campaign_adgroup_DE.csv, …) — easier for upload to language-specific campaigns
• Single CSV: all languages in one file with a Language column — easier for overview and review

Execution Steps

Step 1: Load Context

• Load brand voice from .claude/skills/brand-voice/SKILL.md (check language-specific notes)
• Load ICP and persona from .claude/skills/icp/SKILL.md
• Load platform rules from .claude/skills/google-ads-rules/SKILL.md (check multilingual notes)
• Resolve landing page per language from .claude/skills/landing-pages/SKILL.md:
• For each target language, pick the matching URL variant (exact → EN fallback → first)
• Load cached analysis if available, otherwise fetch and cache
• Flag any languages where no localized page exists

Step 2: Generate per Language

For EACH language, independently:

• Create 15 headlines (≤30 chars) and 4 descriptions (≤90 chars)
• Write NATIVE copy — do NOT translate from another language
• Adapt tone, idioms, and CTA conventions to the target market
• Respect language-specific character challenges (see below)
• Validate character counts in the target language

Step 3: Cross-Language Consistency Check

After generating all languages, verify:

• Core value proposition is consistent across all versions
• No language version contradicts another
• Brand name and product names are spelled identically everywhere
• Social proof claims match (same numbers, same certifications)

Step 4: Validate

For each language:

• All headlines ≤30 characters
• All descriptions ≤90 characters
• At least 2 headlines contain the target keyword for that language
• At least 1 description contains a CTA

Run: python .claude/skills/ads/validate.py output/<file>.csv

Step 5: Export

Based on user's output format choice:

Separate CSVs:

output/rsa_[campaign]_[adgroup]_EN_[date].csv
output/rsa_[campaign]_[adgroup]_DE_[date].csv
output/rsa_[campaign]_[adgroup]_FR_[date].csv
output/rsa_[campaign]_[adgroup]_ES_[date].csv
output/rsa_[campaign]_[adgroup]_AR_[date].csv
output/rsa_[campaign]_[adgroup]_PL_[date].csv

Single CSV (extra Language column prepended):

Language,Campaign,Ad Group,Headline 1,...,Headline 15,Description 1,...,Description 4,Path 1,Path 2,Final URL
EN,...
DE,...
FR,...
ES,...
AR,...
PL,...

Save to: output/rsa_multi_[campaign]_[adgroup]_[date].csv

Step 6: Present Results

Show a comparison table:

| Language | Example Headline          | Example Description                | Char Issues |
|----------|--------------------------|-------------------------------------|-------------|
| EN       | Get Marketing Insights   | Turn campaign data into pipeline    | None        |
| DE       | Marketing-Einblicke      | Kampagnendaten in Pipeline...       | H7: 31 chars|
| FR       | Analyse marketing        | Transformez vos données en...       | None        |
| ES       | Analítica de marketing   | Convierte datos en pipeline         | None        |
| AR       | تحليلات التسويق          | حوّل بيانات حملاتك إلى نتائج       | None        |
| PL       | Analityka marketingowa   | Zamień dane kampanii w pipeline     | None        |

Then suggest:

• Which headlines may need shortening in specific languages
• Cultural differences that might affect CTR (e.g., German preference for formal tone)
• Whether any language version needs more keyword-rich headlines

Language-Specific Guidelines

English (EN)

• Shortest average word length — most headroom in 30 chars
• American vs. British spelling: check brand voice skill
• CTA conventions: direct ("Get", "Try", "Start")

German (DE)

• Compound nouns are the main challenge: "Marketingautomatisierung" = 25 chars alone
• Always prepare shorter alternatives or split compounds
• Formal "Sie" unless brand voice specifies "du"
• CTA conventions: "Jetzt [verb]", "Kostenlos testen"

French (FR)

• Accents (é,è,ê,à,ç,ù,î,ô) = 1 character each
• Articles consume characters: "l'automatisation du marketing"
• Formal "vous" for B2B unless brand voice specifies "tu"
• CTA conventions: "Découvrez", "Essayez gratuitement", "Demandez une démo"

Spanish (ES)

• Accents (á,é,í,ó,ú,ñ,ü) = 1 character each
• Inverted punctuation (¿, ¡) counts as a character — avoid in headlines to save space
• Distinguish between Spain (ES-ES) and Latin America (ES-LATAM): vocabulary and "vosotros" vs. "ustedes"
• Formal "usted" for B2B in most markets; "tú" acceptable in startup/tech contexts
• CTA conventions: "Descubre", "Prueba gratis", "Solicita una demo", "Empieza ahora"

Arabic (AR)

• Right-to-left (RTL) script — Google Ads handles direction automatically, but review carefully
• Arabic characters = 1 character each in Google Ads, same as Latin
• Words can be shorter than English equivalents, but connected script makes visual length deceptive — always validate by character count, not visual width
• Diacritical marks (tashkeel: َ ُ ِ ً ٌ ٍ) are optional in ad copy and count as separate characters — omit them to save space (standard practice in digital Arabic)
• Formal Modern Standard Arabic (MSA/فصحى) for B2B across all Arab markets
• Dialect choice matters for B2C: Gulf (خليجي), Egyptian (مصري), Levantine (شامي) — check brand voice
• Mixed scripts: brand names and technical terms often stay in Latin (e.g., "HubSpot" not "هبسبوت") — each Latin character counts as 1 char
• CTA conventions: "اكتشف" (Discover), "جرّب مجاناً" (Try free), "احجز عرضاً" (Book a demo), "ابدأ الآن" (Start now)
• Review workflow: if you cannot read Arabic, rely on character count validation and have a native speaker verify messaging before upload

Polish (PL)

• Diacritics (ą,ć,ę,ł,ń,ó,ś,ź,ż) = 1 character each — no penalty
• Words tend to be longer than English due to declension
• Informal "Ty" is increasingly common in B2B digital ads
• CTA conventions: "Sprawdź", "Przetestuj", "Dowiedz się"

Iteration Shortcuts

• "make DE headlines shorter" → rework only German headlines
• "FR descriptions need stronger CTAs" → adjust French descriptions
• "swap EN and ES headline styles" → cross-pollinate approaches
• "add AR" → generate Arabic version using existing brief
• "PL headlines 3 and 8 are too similar" → diversify Polish variants
• "review AR character counts" → revalidate Arabic headlines