The "engaged" 20% that was robots

The "engaged" 20% that was robots

A media site celebrated a referral source sending 14k highly "engaged" sessions a month — long durations, deep scrolls. They pursued a partnership. The advertiser asked why those sessions never converted on anything.

The clue was in the technical detail nobody reads: those sessions came from three data centers, used identical screen resolutions, and fired events at robotic intervals. It was bot and scraper traffic that standard filters had missed.

The "don't worry about bots, GA filters them" assumption had let synthetic traffic inflate every engagement metric on the site.

We added IP and data-center exclusions and a known-bot filter. Reported sessions dropped 18%, but real engagement metrics held steady — and the genuine top referral source, previously buried, finally got the credit and budget.

—
Lesson: default bot filtering misses plenty. Inspect suspiciously perfect engagement before you build a partnership on traffic that may not be human.