Creating Variants

[In Progress]

Generate A/B variants from your market research, not guesswork.

HolyShift generates landing page variants from the signals in your validation data. Instead of manually writing alternative headlines and hoping one is better, you let the AI create variants grounded in what your market actually responds to.

How variant generation works

When you create an A/B test, HolyShift looks at your market research data — the Pretotyping Signal Report, real conversation insights, competitive analysis, and demand signals — to identify testable hypotheses.

For example, if your validation revealed that customers care about both "saving time" and "reducing errors," the AI might generate:

• Variant A (control): your current page, leading with the time-saving angle
• Variant B: same page structure, but the headline, subheadline, and primary CTA emphasize error reduction instead

The difference isn't cosmetic. Each variant leads with a different value proposition that was validated through real market conversations. The A/B test then determines which angle converts better for actual visitors.

What the AI varies

Variant generation focuses on the elements that have the highest impact on conversion:

Headlines and subheadlines

The most-tested element. Different headline variants might emphasize different pain points, different benefits, or different emotional angles — all pulled from your market research data.

CTAs

The action you ask visitors to take and how you frame it. "Start your free trial" speaks to a different motivation than "See it in action" or "Join 500+ teams already using Deskflow."

Value propositions

Which features and benefits are highlighted most prominently. Your market research may reveal multiple strong value propositions — A/B testing determines which one resonates most with real traffic.

Section ordering

The sequence in which your story unfolds. Does leading with social proof convert better than leading with features? Does putting pricing early help or hurt? Variant generation can test different section orders to find the optimal flow.

Design presets

If you're unsure which visual style works best, variant generation can create the same content in different presets — for example, Clean SaaS vs Bold Tech — and test which aesthetic drives more conversions from your audience.

Creating a test

1. Go to A/B Testing in your project's Build tab
2. Review suggested variants — HolyShift suggests variants based on your market research signals. Each suggestion includes what's being tested and why.
3. Select or modify — use the suggested variants as-is, or edit them through the chat editor before launching
4. Set traffic split — choose how traffic is divided between variants (50/50 by default)
5. Launch — the test begins immediately. Visitors are randomly assigned to variants.

Best practices

Test one variable at a time

If you change the headline, CTA, and color scheme all at once, you won't know which change made the difference. Focus each test on one meaningful variable. Run sequential tests to optimize multiple elements.

Let your market data guide hypotheses

The strongest A/B tests come from real signals, not hunches. If your validation showed that 70% of conversations mentioned "integration fatigue," that's a headline worth testing against your current angle.

Give tests enough time

Don't call a winner after 50 visitors. Let the test run until HolyShift declares statistical significance. Premature conclusions waste the test and may lead you to the wrong variant.

Document your hypotheses

Before launching a test, write down what you expect to happen and why. This turns A/B testing from random experimentation into a structured learning process. HolyShift tracks your hypotheses alongside test results.

FAQ

Can I create variants manually instead of using AI generation?

Yes. You can duplicate your current page and edit it through the chat editor to create your own variant. AI generation is a shortcut that produces market-informed variants, but manual creation gives you full control.

How many variants can I test at once?

You can run multiple variants simultaneously. For most use cases, 2-3 variants are ideal. More variants require more traffic to reach significance for each one.

Can I test entirely different pages?

Yes. You're not limited to small tweaks. You can test your current page against a completely different version — different preset, different copy, different structure. The AI can generate a fundamentally different page from the same product description using different market signals.

What if I haven't run a validation?

You can still create A/B tests by manually writing variants or asking the AI to generate alternatives from your product description. However, variants grounded in validation data tend to produce more meaningful results because they're testing real market hypotheses.

What's next

• Statistical significance — know when results are real
• A/B testing overview — understand the full testing system
• Analytics — engagement metrics per variant