Generate CTA variants with rationale for A/B testing
intermediateClaude SonnetDesignContent Designcontent-designctacopywritingab-testingconversion
Use case
Use this when planning a CTA test — landing page, signup button, in-product prompt, email link. Most CTA tests fail because all four variants are paraphrases of the same idea; this prompt produces variants that test meaningfully different motivations.
The prompt
You are a content designer writing CTA variants for an A/B test. The variants must test different motivations, not different phrasings of the same motivation. Context: - Product:{{product}}- Surface and moment (where the CTA appears, what just happened, what comes next):{{surface}}- Audience:{{audience}}- Action behind the click (what literally happens):{{action}}- Brand voice constraints:{{voice}}- Length budget (button or link, char count):{{length_budget}}- Current control variant (if any):{{control}}Produce 5 variants, each pulling on a different lever: 1. Outcome-led — names the user's outcome, not the action. 2. Action-led — describes the action plainly. 3. Loss-aversion-led — frames what the user avoids losing. 4. Curiosity-led — promises a specific reveal on the next screen. 5. Identity-led — speaks to who the user is or wants to be. For each variant, produce: - The CTA copy (within length budget). - The lever (label from above). - When this variant tends to win (in 1 sentence — based on the surface and audience). - A risk (1 sentence — eg "may overpromise," "may not match brand voice in a serious context"). After the variants, write: - Recommendation: which variant to test as the challenger against the control, and why. - One additional test idea: a non-copy lever (button color, placement, surrounding context, micro-interaction) that's likely as impactful as the copy itself. Rules: - Variants must be conceptually different. If two pull on the same lever, replace one. - Stay within length budget. Don't cheat with em-dashes that don't fit. - "Get started" and "Learn more" are banned as variants — they're defaults, not choices. - Match brand voice. A curiosity-led CTA on a financial product reads differently than on a consumer app — flag any variant that strains voice. Output: variants in order, then recommendation and additional test idea.
Variables
{{{{product}}}}Replace with your {{product}}{{{{surface}}}}Replace with your {{surface}}{{{{audience}}}}Replace with your {{audience}}{{{{action}}}}Replace with your {{action}}{{{{voice}}}}Replace with your {{voice}}{{{{length_budget}}}}Replace with your {{length budget}}{{{{control}}}}Replace with your {{control}}
Example
Input
product: Helix Analytics surface: Marketing homepage hero CTA — first thing visitors see; clicking goes to free trial signup audience: Marketing analysts at SMBs evaluating self-serve BI; many have used Looker, Tableau, or spreadsheets and are frustrated with both action: Opens the free trial signup form (no credit card required) voice: Calm, capable, non-finance-jargon, slightly dry length_budget: 24 characters max for button label control: "Start free trial"
Output
## 1. Outcome-led
Copy: "See your first chart"
Lever: outcome
When it tends to win: when visitors are skeptical that they'll actually get to a working state — naming the outcome makes the trial feel less like a commitment.
Risk: undersells the product as a one-chart tool when it's much more.
## 2. Action-led
Copy: "Start free trial"
Lever: action (this is the control)
When it tends to win: when visitors are warm and just need a frictionless next step. Reliable baseline.
Risk: forgettable — no differentiation from every other SaaS button.
## 3. Loss-aversion-led
Copy: "Stop fighting Looker"
Lever: loss aversion
When it tends to win: when the audience has acute pain with a competitor; landing-page traffic from comparison searches is a strong fit.
Risk: explicitly anti-competitor copy may strain the calm, dry brand voice and feel adversarial.
## 4. Curiosity-led
Copy: "Try a chart in 90 sec"
Lever: curiosity (specific time promise)
When it tends to win: when visitors are skeptical of "free trial" sprawl and want to know how much time they're committing.
Risk: if the actual time-to-first-chart slips past 90 seconds, the page over-promises and trust drops.
## 5. Identity-led
Copy: "Skip the data team"
Lever: identity
When it tends to win: when the audience self-identifies as marketers stuck waiting on a centralized data team — speaks to who they want to be (self-serve, autonomous).
Risk: may alienate visitors who do work well with their data team or whose workflow includes them.
## Recommendation
Test variant 1 ("See your first chart") as the challenger against the control ("Start free trial"). Outcome-led typically wins against generic action-led on landing pages where conversion friction is low and skepticism is the bigger barrier — and "See your first chart" stays cleanly within the brand voice. If variant 1 wins by 5%+, run a second-stage test with variant 4 ("Try a chart in 90 sec") to see if a specific time promise extends the lift.
## Additional non-copy test idea
Move the CTA above the hero image instead of below. On marketing-analyst landing pages, hero images are often skimmed and the CTA below the fold loses 10–20% of clicks compared to a CTA placed before the user has to evaluate the image. This test usually has higher impact than copy variants and should run before another copy round.
Tips for best results
- 1Five variants on five different levers beats fifty variants on one lever. Lever diversity is what makes tests informative.
- 2Always include a non-copy test idea in the recommendation. Copy is rarely the highest-impact lever — placement, weight, and surrounding context usually matter more.
- 3Loss-aversion variants win in narrow contexts (acute competitor pain) and lose in broad ones. Use surgically.
- 4Watch out for variants that strain brand voice. A 1% lift that breaks voice consistency is a long-term loss.
- 5Tag every variant with its lever in your test tool so the next person inherits the framework, not just the winner.
Related prompts
Rewrite error messages to be human and actionable
intermediateTake a list of system error messages and rewrite each one to be plainspoken, blame-free, and actionable — with the structure: what happened, why, what to do next.
Designcontent-designmicrocopyerror-messages
Generate empty-state copy for product surfaces
beginnerWrite empty-state copy that does three jobs: explains what the surface will look like when populated, gives the user a single concrete next step, and matches brand voice without being cute.
Designcontent-designmicrocopyempty-state
Write microcopy for a multi-step onboarding flow
advancedGenerate every line of copy in a multi-step onboarding — step titles, body, button labels, error states, success states — written as a coherent voice across the whole flow.
Designcontent-designmicrocopyonboarding
Need help implementing this prompt in your workflow?
Book a call