Five Subscribers From Nowhere

I got five new subscribers to my newsletter over three days. The default newsletter analytics attributed them to direct traffic. Apparently, five people had independently typed my URL into their browsers. Seemed unlikely.

This is the kind of thing I can now hand to Claude Code or Codex. I asked it to inspect PostHog (analytics), my main site, and the newsletter nested under it, then figure out where the attribution was getting lost. It traced the setup across both sites, changed the links, deployed the updates, and tested them. Then it caught that one of its proposed fixes didn't fit the way my newsletter was actually hosted and revised its approach. Most of that happened in a session I wasn't even watching.

Then I got another subscriber. This one arrived with the attribution data the others were missing. Claude confirmed that the fix worked.

Then it looked at the web server logs and identified the whole campaign as bots.

Sad trombone.

That investigation began as almost a throwaway task. I used to work at a company where someone spent a meaningful part of their job keeping this machinery working. Attribution bookkeeping wasn't their whole job. They monitored traffic, kept experimenting, responded when patterns changed, and developed a thesis for how to manage that traffic.

But none of the strategic work held up without the bookkeeping. Every link had to carry the right information. Every source had to arrive with the right label. If that chain wasn't simple and reliable, we couldn't tell where the traffic came from, whether we were spending in the right places, or what we should try next. The whole house of cards came down.

The bookkeeping consumed time and forced decisions because every experiment depended on being able to measure it. That strategic job still exists. Someone still has to decide what the traffic means, which experiment is worth running, and how to respond. But the prerequisite work that used to eat into those decisions can now happen in the background.

I wanted six new subscribers. I got reliable attribution and an answer I didn't like. Without these tools, I wouldn't have identified the bot traffic until the mass spam reports started arriving.

A sneaking suspicion prompted an investigation that diagnosed the problem, repaired it, deployed and tested the fix, and then challenged the conclusion the fix appeared to confirm. I was doing something else for most of it.