Ad copy generator

The AI-copywriting hype cycle peaked around 2023-2024. By 2026 we've learned the boring truth: templates outperform LLM output for ad copy at scale because the structural patterns are what drive performance, not the linguistic novelty. A 'direct-response opener' beats a 'creative-genius novel sentence' nearly every time the funnel optimization is honestly measured.

LLM-generated copy has a place - for brand-voice tuning, for category-specific phrasing, for refining a template variant - but the *starting* copy should come from validated templates. The generator above is the floor; LLM polish is the ceiling.

The biggest mistake performance teams make: they treat LLM output as a substitute for templates. The result is copy that sounds 'creative' but underperforms. The right workflow is templates first, LLM polish second. Get the structural pattern right; then refine the language.

The 10 templates above work cross-platform but each platform has specific norms that shift which archetypes win:

Meta rewards short, direct, scroll-stopping openers. The 'Direct-response opener' and 'Specificity hook' archetypes outperform 'Founder-voice short' on Meta because Feed scroll is anti-narrative.

Google Search rewards keyword-matched copy that explicitly states the search intent. 'Question hook' and 'List opener' tend to win because they mirror the way users phrase queries.

LinkedIn rewards narrative-first, first-person openers. 'Founder-voice short' is the strongest archetype on LinkedIn; 'Direct-response opener' often reads too salesy for the platform.

TikTok rewards conversational, native-feeling openers. 'Curiosity gap' and 'Direct address' outperform 'Social proof short' because TikTok's audience reads social proof as inauthentic.

Pick the archetype that fits your platform. The generator above lets you switch platforms via the input, but the template selection logic is on you - the generator outputs all 10 regardless.

Most teams test copy by varying everything at once - new hook, new body, new CTA. That's not testing; that's wishing. The right cadence:

Test archetype, not phrasing. Pick 4 archetypes from the generator (problem-first, founder-voice, question, comparison). Run 1 variant of each. The performance difference between archetypes is 3-5x larger than the difference between phrasings within an archetype.

Hold the offer constant. Vary the hook and the body; keep the offer identical. Otherwise you're testing the offer + the copy + their interaction, which requires 4x the sample size to read.

Run for a full purchase cycle. 24-hour CTR reads are noise. For DTC, 7-14 days. For SaaS, 21-30 days. Most teams kill winners early because the day-1 numbers look bad.

Document what won. When an archetype wins, write a one-line teardown of why - 'problem-first hook with specific audience worked because endurance athletes recognize the pain language'. The teardown is the brief input for the next sprint.

Internal: creative test sample size calculator for the math layer; test significance calculator for the post-test read.