Description
Almost nothing on the internet was A/B tested. Not the button you clicked to start this, not the headline that made you click it. Somebody decided, somebody shipped it, and the number moved or it didn't.Most people doing that feel slightly guilty about it — like there's a proper version of this job they're not doing. This episode is about where the line actually is between fixing something and guessing, and how to find out in twenty minutes whether the test you're about to run has already been run by somebody else.Two real teardowns. A high-ticket lead form where roughly one in three people opened it and never touched a key. And a B2B contact form from a year earlier where the real problem wasn't persuasion at all — the form didn't fit a phone screen.Then I ran my own advice on my own work for the first time, and the answer came back worse than I expected.IN THIS EPISODE (00:00) Almost nothing on the internet was A/B tested (01:10) Who I am (01:25) The form, and the number that stopped me (03:25) What a test actually costs (05:05) The five things I changed (07:05) The second case: a form nobody could see (08:45) Why I thought I was allowed (10:05) The check — and the row with no evidence at all (12:15) The calculation nobody runs: ten months versus six weeks (13:55) The four things that have to be true before I ship untested (14:55) What it cost me (15:45) Your site nextTHE PROMPT The research prompt from the episode — role, rules, output format — is in the show notes on the site, ready to copy.MENTIONED Nielsen Norman Group, "Placeholders in Form Fields Are Harmful" Baymard Institute, "Mobile Form Usability: Never Use Inline Labels" GoodUI's public archive of A/B test results Evan Miller's sample size calculatorEvery number in this episode came out of the client's own GA4. I read it with ConvRadar, which I built after doing this by hand for eleven years — it takes the funnel that's already being tracked and tells you which step is losing the money. convradar.comWANT YOURS TORN DOWN Send me your URL. I'll pick one or two, contact the owner, and the next episode is that teardown, with the analytics open on the recording. Not a sales call — you watch me work with the data.