Cookie-hashed 50/50 split at the home route. Variant A: 'Director of Growth researching how answer engines cite the web.' Variant B: 'I help B2B teams get cited by answer engines and lift conversion.' Primary metric: About-page CTR from home. Secondary: contact-intent events (mailto + LinkedIn from Contact/About/author cards). Guardrail: newsletter signup rate. --- Restarted 2026-07-17 --- Previous run ended before completion. Result: Restarted before completion: site content was replaced mid-test, contaminating the comparison. New run — hypothesis and metrics unchanged (primary: About-page click-through from home; secondary: contact-intent events from Contact/About/author cards). Two refined arms: ARM A (authority-led): "I run the CRO program at Twilio. Nights and weekends, I study how answer engines select and cite web content — and publish the experiments here, including the ones that flop." ARM B (outcome-led): "Fifteen years of testing what makes B2B websites convert — more than $10M in tested revenue impact — now pointed at a newer question: how do you earn citations from AI?" SSR/hydration note: the server always renders ARM A as the fallback; on the client, an effect (post-hydration) reads the visitor cookie and swaps in the assigned arm. Because both server and first client render emit the same fallback, React reports no hydration mismatch — the arm swap is a normal post-mount re-render. Consequence for the experiment: ARM B viewers see a brief ARM A flash on first paint. Exposures are logged only after assignment, so exposure counts are unaffected; primary-metric attribution is unchanged.
Running — restarted with refined arms. Reporting when confidence intervals permit.