In the spring of 1954, drivers in a few towns north of Seattle started finding tiny pits in their windshields. The story hit the Seattle papers on April 14. By April 15, close to 3,000 windshields had been reported.
Nobody could explain it. Some people blamed a new navy radio transmitter nearby and others blamed sand fleas. The mayor asked the governor and President Eisenhower for help. Then a sergeant from the police crime lab summed it up as "5 per cent hoodlum-ism, and 95 per cent public hysteria." Most of those pits had been there all along. People had just never looked at their windshields that closely.
Sometimes, I think about that story when someone says "let's just watch the recordings." If you haven't used it, session replay records what people do in your app and plays it back like a video, every click and every pause. It's the closest thing to sitting behind your users, and that's exactly why it feels like the truth.
The recordings don't lie.
They don't. The person watching them does, usually without noticing.
The tempting way to watch goes like this. You already suspect the new filter is confusing. You open the newest recordings and find someone hesitating on the filter. Then you Screenshot. And now you have proof, but almost every recording has somebody hesitating somewhere. You found it because you went looking for it, same as the windshields.
Every recording has a little bit of whatever you walked in looking for.
PostHog's guide on watching recordings leans toward opening them from a chart instead of from the list, and I think that's the right call. You can open recordings straight from a funnel, which is the chart of the steps people take toward something like signing up. Pick the step where people quit, and you can watch the ones who dropped off or the ones who made it through.
There's also a built-in collection called frustration signals. It pulls sessions from the last seven days that had errors or lots of rage clicks, which is someone clicking the same spot over and over. And you can hide the recordings you or your team already watched, so you're not rewatching the ones that already agreed with you.
The part I'd add is to always watch both groups. If the people who finished signing up hesitate on that same filter just as much as the people who quit, the hesitation isn't why they quit. It's just what people look like while they read a screen.
I call this the windshield check. Before you blame a step, watch the people who made it through.
The way I do it now is boring on purpose. Open the funnel, pick the step with the biggest drop, watch some recordings from the people who quit and some from the people who finished. Then write down what both groups did on that step, before I propose a fix. Most of the time the two lists look the same, and the redesign dies in the doc, which is the cheapest place for it to die. When the lists are different, that difference is the bug, and it’s usually something I hadn’t guessed.
"So recordings are useless"
Not even close. Sometimes one recording is all you need. If a button doesn't work, you don't need to be fair about it, you need a fix, and a recording shows you the bug faster than any report.
Watching both groups also takes twice as long, and on a busy week that's a real cost. I'd still rather pay that than ship a redesign for a problem I mostly brought with me.
Seattle's windshields really were pitted. The epidemic was what people brought with them when they finally looked.
What AI did, what I did
AI checked the 1954 dates and the sergeant's quote against the newspaper archive, and pulled the PostHog docs on funnels and frustration signals.
I wrote the text, picked the story, and the windshield check is how I actually watch recordings.
Keep committing,
Eder