Persona testing is working — and AI is letting us run it at a scale we've never had before.

We've been testing creative angles across Google PMax and Facebook Persona Landers for the past six weeks. The read after 30 days is unambiguous: the Beginner-Setup angle and the Spanish-language localization angle are doing the heavy lifting, and every other framing we've tested — Bitcoin Maximalist, Security-focused, Competitor Comparison — is underperforming the account average by a wide margin.

On Facebook, the Beginner-Setup creative is carrying the entire Persona Landers campaign. Freeze-Beginner-Setup delivered 22 purchases at a $56 CAC over the past 30 days (3.06x ROAS). Hold-Beginner-Setup delivered 19 purchases at a $74 CAC (2.36x ROAS). Those two ads alone generated $7,086 of conversion value on $2,637 of spend.

What failed alongside them is just as instructive. We ran an 8-variant "Fresh Copy" A/B test with emotional and abstract framings — ownership, anxiety, fear, relief, demystification. Every single variant produced zero conversions on $900+ of spend. The signal isn't that Beginner audiences want clever copy. The signal is that they want the specific, concrete Setup framing — "plug in, follow the app, 10 minutes, done" — and nothing else converts them.

On Google, the same pattern holds. Beginner-Friendly asset groups in US PMax delivered $835 spend → 12.8 conversions → $65 CAC → 4.55x ROAS over 30 days. EU PMax Beginner-Friendly delivered $2,363 spend → 19.4 conversions → 2.51x ROAS. The Beginner angle is the only creative pattern that's working at scale in three different campaigns on two different platforms.

"Beginner" is really four different people, each in a different emotional state when they see an ad. Yesterday we launched four new Google PMax asset groups — one for each sub-segment — inside the US Sales-Performance Max campaign. All four stay inside the Setup framing that's already proven, but each gets its own entry point, its own hook, and its own objection resolution.

The test reads next week. What we learn will shape the entire spring creative calendar and the UGC brief we're running with Get Based + our ~40-creator affiliate network. The goal isn't to find one winner — it's to learn which sub-segment framing is the most scalable, and fan out from there.

The South America PMax campaign is quietly the best-performing thing in the entire Google account. Over the past 30 days, on less than a fifth of the budget we're spending in the US, SA has produced more conversions and 63% more conversion value at less than half the CAC. The single reason is localization — specifically, Spanish-language creative written for Mexican and Argentine audiences, not machine-translated English ads.

The SA Spanish result is a pattern, not an accident. So two weeks ago we launched the same pattern in Europe: German, French, and Italian localized asset groups inside EU PMax, each with local-language copy written to match market conventions. We also launched a Portuguese variant inside SA PMax targeting Brazil specifically.

These are all in their first full week of data as of the April 13 run. Early reads below — they're young and we're not making go/no-go calls yet, but the pattern is the point: if localized Spanish unlocked SA, localized DE/FR/IT should unlock EU. Same brief, same landing page, same Setup framing — translated and adapted for each market.

None of the work above would have been possible at this pace without a new operational setup: Claude Code wired directly into the Google Ads API and the Meta Marketing API. Strategy and creative direction still come from the team. But the execution — creating campaigns, writing hundreds of text assets, launching new asset groups, pausing underperformers, adjusting tROAS targets, adding negative keywords — happens in minutes inside a single session, not over weeks of agency back-and-forth.

Yesterday alone we launched four new PMax asset groups, wrote over 100 text assets, paused an underperforming asset group, lowered a target-ROAS, and added a brand negative keyword across three search campaigns. That's a week of agency labor compressed into one afternoon. The week before, we launched four localized language variants across two continents in a single session. The bottleneck used to be how fast we could execute. The bottleneck now is how fast we can learn from what we've launched.

This is how we plan to keep testing: more sub-segments, more languages, more creative variations — not because we have more people, but because the operational floor is moving out of human hands and into the model's.