What is Variant?
A different version of the same ad concept. Variants share the same product and offer but differ in headline, visual composition, or creative angle. Testing 3-5 variants per concept is the minimum for meaningful data.
In mani: Mani generates genuinely different creative angles per variation, not just rewording.
What makes a variant genuinely different
A true variant changes the creative angle, not just the wording. Five variants of "Buy our product" with different adjectives is not testing; it is rephrasing. Genuine variants: Variant A uses a problem-solution hook, Variant B uses social proof, Variant C uses a founder story, Variant D uses humor, Variant E uses a before/after comparison. Each variant tests a fundamentally different persuasion strategy. AI tools that generate genuinely different angles outperform tools that only rephrase.
How many variants to test
Minimum: 3 variants per concept (enough for a clear winner). Optimal: 5-7 variants (captures more angles without fragmenting data). Maximum: 10 variants per ad set (beyond this, each variant gets too few impressions). At $20/day per ad set, 5 variants each get about 200-300 impressions per day. Statistical significance arrives in 3-7 days. With AI generation, producing 5-7 genuine variants takes 15-30 minutes instead of 2-3 days with traditional creative production.
Managing variant performance
After 1,000 impressions per variant, turn off underperformers (below 0.8x your average CTR). After 2,000 impressions, you have enough data to identify your top 1-2 winners. Scale winners by increasing their budget allocation. Use winners as the baseline for the next round of variant testing. This iterative process (generate, test, cull, scale, generate) is the core creative optimization loop. AI tools make the "generate" step fast enough to sustain weekly testing cycles.