The Marketing Metric That Was Lying to Me

I was assembling organic search numbers for a case study — GA4 on one side, Search Console on the other. The GA4 side showed a stretch where traffic jumped hard. On its own it looked like a win.

Then I lined it up against Search Console for the same weeks and it fell apart.

Two GA4 properties were spiking at almost exactly the same rate, in the same window. Search Console — impressions, clicks, average position — showed nothing over those weeks that would produce that kind of jump.

Real organic growth shows up in both systems, roughly in proportion. This showed up in one, in a shape that looked more like a measurement artifact than actual demand.

I could have reported it. It was in the tool, it was technically traffic, and it made the trend line steeper.

I excluded it and used the number both systems agreed on. The growth figures in that case study — 203% more organic sessions, 232% more organic pageviews — are what held up after the anomaly came out.

Cross-check every headline number against a second source. GA4 against Search Console for organic. The ad platform against GA4 for paid. Never report a number that only one system can see.

Be suspicious of clean correlation. Two properties, two channels, or two campaigns moving in near-lockstep is usually instrumentation, not the market.

Ask what mechanism would produce the change. If you cannot name a real-world cause — a campaign, a ranking gain, a season, a press hit — treat the number as suspect until you can.

Segment before you celebrate. A spike concentrated in one source, one geo, or one hour of the day is a flag, not a result.

An inflated number does not just make one report wrong. It sets a baseline you now have to keep beating. It sends budget toward something that did not actually work. And the first time someone catches one, every other number you have reported gets a second look.

Excluding noise costs you a smaller headline once. Reporting it costs you your credibility later.

If a number would change a decision, it gets verified against a second source before it goes in a deck. If it cannot be verified, it goes in with a flag or it does not go in.

That is the whole rule. It is slower. It is also why I can hand someone the underlying data and let them check it themselves.