Most CTA buttons say “Buy Now,” and most of them under-convert. The button is also the single highest-leverage element on a page: in one of conversion optimization’s most-cited tests, Michael Aagaard changed Unbounce’s button from “Start your free trial” to “Start my free trial” and clicks rose up to 90% (Unbounce). A single word. These 15 prompts target that side of CTA writing: button text mapped to a specific buyer hesitation, microcopy that removes friction, a secondary CTA for not-yet buyers, and a structured A/B test matrix so you find the winner with traffic instead of guessing.
The prompts are model-agnostic. As of June 2026 we ran them on GPT-5.5, Claude Sonnet 4.6, and Gemini 3.1 Pro; all three handle CTA generation well. Sonnet 4.6 tends to produce the tightest microcopy; GPT-5.5 is strongest at the test-matrix table.
TL;DR
- First-person (“my”) beats second-person (“your”) in most consumer tests; make it your default control variant.
- Brief the model on the exact hesitation each variant should solve (price, time, fit, commitment), or it returns forgettable copy.
- Don’t call a winner until each variant hits ~100 conversions and 95% significance, and let the test run at least two full weeks to cover weekly buying cycles.
- Test copy before color: copy can move conversions double digits, color rarely does.
Who this is for
CRO leads, growth marketers running landing-page tests, Shopify operators tuning checkout, founders A/B-testing their first hero block, and agency CRO consultants.
When not to use these prompts
Skip these for purely brand-awareness pages where conversion is not the goal. Skip too if you have thin traffic: a CTA test needs roughly 100 conversions per variant before the result means anything, so on a page converting at 2% that is ~5,000 sessions per variant. Below that, you are reading noise.
Prompt anatomy / structure formula
A conversion CTA prompt should always carry six elements:
- Role: who the AI plays (luxury copywriter / Amazon listing strategist / DTC brand voice / paid-ads hook writer).
- Context: product, brand voice, target buyer, platform, price tier, season — anything that shifts copy.
- Goal: one concrete deliverable — 5 bullets, a 150-word hero, 13 tags, 10 hook lines, a refund reply.
- Constraints: must / must-not (FTC claims, banned words, character limits, tone, no emoji, no superlatives).
- Output format: numbered list, table, JSON, or labeled blocks so you can paste straight into the seller backend.
- Examples / signal: 1-2 reference lines you like, or anti-examples (“not like this competitor”).
Best for
- Landing-page primary CTA button
- Pricing page tier CTAs
- Checkout-step button copy
- Microcopy under buttons (trust + friction-removal)
- A/B test variant generation
15 copy-ready prompt templates
1. Action + outcome button copy
Default scaffold; outperforms “Buy Now” in most A/B tests.
You are a conversion copywriter for [brand]. Generate 8 primary CTA button variants for [product / page]. Each follows the action + outcome pattern (e.g. "Get My Sleep Score", "Start My Trial", "See My Match"). Keep each under 4 words. Avoid "Buy Now", "Click Here", "Submit".Variables to swap: brand, product / page, primary buyer outcome
Optimization: If the variants feel too similar, append this line: “Make each variant solve a different buyer hesitation: price doubt, time doubt, fit doubt, commitment doubt.”
2. First-person CTA variants
For [product / page], write 6 first-person CTA button variants (e.g. "Send My Box", "Reserve My Seat", "Build My Plan"). First person ("my") consistently outperforms second person ("your") in consumer conversion tests. Keep each under 4 words. No emoji.3. Friction-removal microcopy under button
For the CTA button "[paste button text]", write 5 microcopy variants that go directly under the button. Each addresses one buyer hesitation in 8 words or fewer: cancellation, payment, commitment, returns, privacy. Example: "No card required. Cancel anytime."4. Secondary CTA for “not yet” buyers
For a page where the primary CTA is "[paste primary CTA]", write 5 secondary CTA variants for buyers who are interested but not ready (e.g. "See How It Works", "Get the 60-Second Tour", "Read 3 Reviews", "Compare Plans", "Email Me the Details"). Keep each under 6 words.5. Pricing-page tier CTA
For a 3-tier pricing page (Starter, Pro, Enterprise), write a distinct CTA per tier with an action verb that matches the commitment level. Starter = low-commit, Pro = decision verb, Enterprise = scheduling verb. Keep each under 4 words. Match tone to tier.6. Checkout-step buttons
For a 3-step Shopify checkout (Info → Shipping → Payment), write button copy for each step that reduces drop-off: progress language ("Continue to Shipping"), reassurance language ("Review My Order"), final-action language. Avoid "Place Order" if "Send My Order" or "Complete Purchase" tests better.7. CTA + urgency without manipulation
Write 5 CTA button variants for [product] that signal urgency without fake scarcity. Examples: "Ship Today", "Reserve Before Tuesday", "Lock In November Price". Each must be literally true; nothing invented. Keep each under 5 words.8. CTA for high-AOV / considered purchase
For a high-AOV product (over $300), write 5 CTA variants that match the considered-purchase mindset. Examples: "Book a 15-Minute Call", "Send Me the Spec Sheet", "Reserve One". Avoid impulse-buy verbs. Keep each under 6 words.9. Sticky bar / scroll CTA
For a sticky CTA bar that appears after the buyer scrolls 50% of the page, write 5 short headlines (under 6 words) plus 5 button copies (under 3 words). The bar must add urgency or value the original hero did not deliver, not repeat it.10. Mobile-thumb-zone CTA
For a mobile-only CTA (bottom-fixed, thumb-zone), generate 5 button variants under 3 words each that fit a 90px width. Each must contain one verb and one outcome word. Examples: "Get Mine", "Start Free", "Try $1".11. CTA test matrix (6 variants)
Generates a structured A/B test set, not just clever copy.
For [product page], generate a 6-variant CTA test matrix: 2 control-style (verb + product), 2 outcome-led, 2 first-person. For each, output a table row with: button copy, the buyer hesitation it solves, and the success metric to watch.12. CTA + social proof microcopy
Below the CTA button "[paste button]", write 5 social-proof microcopy variants under 8 words (e.g. "Joined by 12,400 home cooks", "Loved by 8,300 readers", "Trusted by 600 teams"). Each number must be true and citable. No "thousands of customers".13. CTA for free-trial / freemium
For a freemium / free-trial product, write 5 CTA variants that name what the buyer gets immediately, not what they save later. Examples: "Open My Free Account", "Get My 14-Day Pro Trial", "Start Free, Upgrade Later". ≤ 6 words.14. CTA inside email body
For an email campaign for [product], write 5 in-email CTA button variants under 4 words each. Each must contain one action verb and one specific reward. Avoid "Learn More" and "Read More". For each CTA, also output the 1-line context sentence that sits above it.15. CTA recovery (post-abandonment)
For an abandoned-cart email or retargeting ad, write 5 recovery CTA variants. Each must reference the abandoned action specifically: "Finish My Order", "Save My Cart", "Pick Up Where You Left". Keep each under 5 words. Voice: helpful, not nagging.Common mistakes
- Defaulting to “Buy Now.” It tests worse than first-person or outcome-led variants in most consumer categories.
- Shipping a single CTA without testing it. The button is the highest-leverage A/B test on most pages, so guessing here is the most expensive place to guess.
- Stuffing the CTA with two verbs (“Sign Up and Get Started Today”). Pick one action.
- Microcopy that contradicts the button. “Buy Now” sitting above “Cancel anytime” reads as confusing rather than reassuring.
- Letting the model write CTAs without a stated buyer hesitation. With no hesitation to anchor on, it returns generic, forgettable copy.
- Calling a winner before ~100 conversions per variant and 95% significance. Below that, random noise hands you a fake winner that regresses in production.
- Stopping the test on day three because one variant “looks” ahead. Run at least two full weeks so weekday-versus-weekend buying patterns wash out.
- Ignoring the mobile thumb-zone. A desktop-first CTA loses on the screen where most traffic now lives.
Tactics that compound
- Always brief the model on the specific buyer hesitation each variant should solve; that single instruction is the difference between a usable batch and filler.
- Make first-person the default control. Aagaard’s “my” beat “your” by up to 90% in the canonical Unbounce test, and the direction holds across most consumer tests even when the size is smaller.
- Pair every button test with one microcopy test underneath. Stacked friction removal usually beats moving one variable at a time.
- Test the highest-traffic step first (usually the hero block); a win there compounds through every downstream step.
- Keep a CTA library split by category. What wins for SaaS rarely wins for jewelry, and reusing a cross-category “winner” is how teams ship duds.
- Log the losers and the reason they lost. The “why it failed” note is the cheapest input to your next test.
- For mobile, size and color the CTA from the thumb-zone outward and treat desktop as the secondary layout.
FAQ
- How long does a CTA A/B test need to run?: Long enough to reach roughly 100 conversions per variant at 95% significance, and at least two full calendar weeks regardless of when significance appears. The two-week floor lets weekday-versus-weekend buying patterns average out so you don’t call a Tuesday spike a winner.
- How much traffic do I need before testing CTAs?: Work backwards from your conversion rate. At a 2% page conversion rate, ~100 conversions per variant means ~5,000 sessions per variant. Below a few thousand sessions per variant, you can’t separate signal from noise, so improve the offer first and test the button later.
- Should I test CTA color or copy first?: Copy first. Copy can move conversions double digits; color rarely does. Color still matters for contrast and accessibility, just not as a conversion lever.
- How many CTAs should a page have?: One primary action, repeated 2-3 times as the buyer scrolls, plus one secondary CTA for not-yet buyers. More than that splits intent and dilutes every option.
- Is a first-person CTA always better?: Usually for DTC and consumer, where “my” creates ownership; it’s less clear in B2B, where “your” can read more professional. The 90% Unbounce result is the ceiling, not the average, so keep first-person as your control and test it in your own category.
- Can AI predict the winning CTA?: No. The model generates strong candidates; real traffic decides. As of June 2026, GPT-5.5, Claude Sonnet 4.6, and Gemini 3.1 Pro all produce good variants but none can forecast which one your buyers will click. Use them for generation, not for picking the winner.