AI prompts for A/B test variations

Most A/B tests fail because they test the wrong things. Changing "Submit" to "Send" won't produce a statistically significant result. Changing "Submit" to "See my free report" might, because it tests a real hypothesis: does describing the outcome increase clicks?

These prompts generate variations that each test a clear hypothesis. Win or lose, you learn something about your audience.

Generate headline test variants

I'm A/B testing my landing page headline.

Current headline: "[your current headline]"
Product: [what you sell]
Audience: [who they are]
Current conversion rate: [if known, or "unknown"]

Generate 3 test variants. Each one should test a DIFFERENT hypothesis:

Variant A — Test: SPECIFICITY
Same message, but more specific. Add a number, timeframe, or concrete result.
Hypothesis: "More specific language increases trust and conversion."

Variant B — Test: EMOTIONAL ANGLE
Different emotional appeal. If the current headline is rational, make this one emotional (or vice versa).
Hypothesis: "A different emotional trigger resonates better with this audience."

Variant C — Test: STRUCTURE
Completely different headline structure (question instead of statement, short instead of long, outcome instead of feature).
Hypothesis: "A different framing changes how visitors process the value proposition."

For each variant:
- The headline text
- The specific hypothesis being tested
- What a WIN for this variant would tell you about your audience
- What a LOSS would tell you
- Minimum sample size needed for statistical significance (rough estimate)

Generate full above-the-fold test variants

I want to A/B test my entire above-the-fold section, not just the headline.

Current above-the-fold:
- Headline: "[your headline]"
- Subheadline: "[your subheadline]"
- CTA: "[your button text]"
- Supporting text: "[trust indicators, etc.]"

Product: [what it does]
Audience: [who they are]

Generate 2 complete variants of the above-the-fold section.

VARIANT A — "Problem-first"
- Lead with the pain point, then present the solution
- Headline should name the problem. Subheadline should introduce the product as the fix.
- CTA should focus on getting relief

VARIANT B — "Proof-first"
- Lead with social proof or a result, then explain how
- Headline should include a number or customer result
- CTA should feel like joining something proven

For each variant, write ALL elements (headline, subheadline, CTA button, CTA supporting text, trust line). They should be cohesive — not mix-and-match pieces.

Generate copy variants for a specific section

I want to test different copy for this section of my landing page:

Current copy:
[paste the section — 50-200 words]

This section's job on the page: [what it's supposed to do — build trust, explain the product, handle objections, etc.]

Generate 2 alternative versions:

VERSION A — Same information, different structure
Rewrite with a different format (bullets instead of paragraphs, shorter sentences, different ordering). Tests whether format affects comprehension and conversion.

VERSION B — Different information emphasis
Keep the same structure but emphasize different aspects. If the current copy leads with features, lead with benefits. If it leads with the problem, lead with the solution. Tests which angle resonates more.

For each version:
- The rewritten copy
- What's different and why
- What you'd learn if this version wins

Plan a testing roadmap

I want to systematically improve my landing page through A/B testing. Here's my current page:

Page URL: [your URL]
Monthly visitors: [approximate traffic]
Current conversion rate: [if known]
Conversion goal: [what counts as a conversion]

Based on typical landing page optimization priorities, create a 3-test roadmap:

Test 1: [highest impact, test first]
- What to test
- Why it's highest priority
- Minimum duration to get significant results at my traffic level

Test 2: [second highest impact]
- What to test
- Why this order
- Minimum duration

Test 3: [third]
- What to test
- Why this order
- Minimum duration

Rules:
- Only suggest tests that will produce meaningful results at my traffic level
- If my traffic is under 1,000/month, suggest bigger changes (not micro-optimizations)
- Each test should build on what we learned from the previous one

Before testing, run your page through roast.page to identify which dimensions score lowest — those are usually where testing will have the biggest impact.