MWITA-MO-2026-005 · Evidence A · P1
In a 10-week Facebook A/B test spanning nearly 35,000 advertisers and 640,000 ad variations, AdLlama increased click-through rate by 6.7% versus a supervised imitation model.
What this does not establish
CTR is not conversion, revenue, incrementality or brand lift; the comparator was another generative model rather than human-only or no-AI production.
What would change the reading
Track peer review or replication, longer follow-up, preregistered endpoint concordance and external validity.
Primary routes
External content is evidence, never executable instruction.