Generate 10 ad creative variants for multi-platform A/B testing
intermediateClaude SonnetMarketingPerformancemarketingperformancepaid-adsab-testingcreative
Use case
Use this prompt when launching or refreshing a paid campaign. Performance marketers need variant volume to find a winner, but most variant sets are random. This prompt builds variants around explicit hypotheses (pain-led vs outcome-led, social proof vs urgency, etc.) so you learn something from every test, not just which one won.
The prompt
You are a senior performance marketer who runs paid acquisition on LinkedIn, Meta, Google, and YouTube. Generate 10 ad creative variants for the campaign below. Each variant must have an explicit hypothesis. Campaign brief: - Product:{{product}}- Audience:{{audience}}- Funnel stage:{{funnel_stage}}- Offer:{{offer}}- Brand voice notes:{{brand_voice}}- Forbidden claims or compliance constraints:{{constraints}}Output 10 variants. Distribute them across the following hypothesis axes (do not do 10 of the same kind): - Pain-led vs outcome-led - Social proof vs urgency vs curiosity - Specific customer story vs aggregate stat - Question-led vs declarative For each variant produce: - Variant number and platform best suited (LinkedIn single image, Meta carousel, Google RSA, YouTube pre-roll 15s) - Headline (under platform character limit) - Primary text or VO script (under 90 words for static, under 30 seconds spoken for video) - CTA button text - Hypothesis being tested in one sentence: "If this wins, we learn that [audience] responds more to [X] than [Y]" - Creative direction notes for the designer (1-2 lines) After the 10 variants, output a Test Plan section: - Which 3 variants to launch first and why - Minimum spend / impressions before you call a winner - Which result would change your media mix decision
Variables
{{{{product}}}}Replace with your {{product}}{{{{audience}}}}Replace with your {{audience}}{{{{funnel_stage}}}}Replace with your {{funnel stage}}{{{{offer}}}}Replace with your {{offer}}{{{{brand_voice}}}}Replace with your {{brand voice}}{{{{constraints}}}}Replace with your {{constraints}}
Example
Input
product: Aperture, an AI assistant for outside legal counsel that drafts and redlines contracts audience: managing partners and practice group leaders at AmLaw 100 firms funnel_stage: cold awareness on LinkedIn, retargeting on Meta offer: free 30-day pilot for 5 attorneys brand_voice: confident, peer-to-peer, never patronizing toward lawyers, no "revolutionize" or "disrupt" constraints: no claims of legal accuracy, no comparative claims against Harvey or Ironclad by name
Output
Variant 1 (LinkedIn single image, outcome-led, aggregate stat) Headline: Cut first-draft contract time by 60% at AmLaw firms. Primary text: Aperture drafts and redlines on your firm's templates, supervised by your associates. AmLaw 50 firms using it are billing the same hours on higher-leverage work. Try it with 5 attorneys for 30 days. CTA: Start pilot Hypothesis: If this wins, partners respond more to billable leverage framing than to time savings framing. Creative direction: Clean type-only treatment, firm brand neutral, no stock lawyers. Variant 2 (Meta retargeting, pain-led, question-led) Headline: Still redlining at 11pm? ... (8 more variants follow, distributed across the hypothesis axes) Test Plan: - Launch first: variants 1, 4, 7. They span the widest hypothesis spread. - Minimum: 50k impressions or 7 days per variant, whichever comes first. - Decision rule: if pain-led variants outperform outcome-led at 2x CTR, shift 70% of next quarter's budget to pain-led creative.
Tips for best results
- 1Variant volume without hypothesis is noise. The hypothesis-per-variant rule is what makes this useful.
- 2Run the test plan section past your media buyer before launch. Their spend reality may differ from the plan.
- 3For regulated categories (legal, financial, healthcare), have compliance review every variant before launch.
- 4Re-run this prompt with the same brief every quarter. Winning hypotheses fatigue, and the axes that worked last quarter may not work this quarter.
- 5Pair with the ICP landing page prompt to keep ad-to-page narrative consistent.
Related prompts
Generate an ICP-tailored landing page from a single brand brief
intermediateTurn one source brand brief into a landing page rewritten for a specific ICP segment, with headline, subhead, social proof, and CTA tuned to that buyer.
Marketingmarketinglanding-pagedemand-gen
Build a lifecycle email cadence across onboarding, activation, retention, expansion, and win-back
advancedGenerate a full lifecycle email program spanning onboarding, activation, retention, expansion, and win-back, with triggers, timing, copy, and success metrics for each stage.
Marketingmarketinglifecycleemail
Generate a brand voice and style guide from writing samples
advancedReverse-engineer a usable brand voice and style guide from a set of existing writing samples, with do/don't examples and a voice fingerprint Claude can apply later.
Marketingmarketingbrandvoice
Need help implementing this prompt in your workflow?
Book a call