This is the last part of the series. You have chosen metrics, installed the tag, named events by meaning, divided the path into a funnel, labelled your inbound links, decided what to collect and given the report a shape. One question remains: can these numbers be trusted?
On most sites the figures are inflated. The trouble is that the inflation is not constant. If it were, the trend would still be readable — but when it swells in one particular month, that month’s change gets mistaken for a result.
Three causes
Internal traffic is the most common. The people who open the site most often work for you, and they open it most of all just after a redesign — producing the classic illusion that traffic rose in the month you relaunched.
Bots behave differently. Some are legitimate search crawlers, others are scrapers. They rarely trigger events, so they appear as more visits with no more conversions. If a conversion rate falls for no visible reason, suspect this first.
Duplicated tags are the problem flagged in the second part: a plugin and a tag manager loading the same tool, or leftover code from a previous theme.
The order to check in
The fourth line is the quiet one. Cloning a site to staging clones the tracking code with it, so development clicks land in production data. And because that traffic only appears while the team is working, the numbers rise in the months you were busiest. Put this on the checklist you run straight after any clone.
Duplicated tags reveal themselves immediately: search the page source for the tracker’s identifier. More than one occurrence is your answer.
What to suspect, in order
When a metric jumps, having a fixed order of suspicion saves a great deal of time: ① did the measurement change → ② did the site change → ③ did the market change. Most investigations end at ①, yet conversations almost always begin at ③.
Which is why it is worth logging the days you touched the measurement. One line each for tag changes, new events and tool settings. Laying that list beside the graph explains half the kinks in it without any further work.
Make the check a habit
Quality checking belongs in the calendar, not in the aftermath of an incident. Once a quarter, walk the five lines above. It takes far less time than unwinding a decision made on the wrong numbers.
Closing the series
What these eight parts built is not a tool configuration but a measurement system: decide what to measure, name it by meaning, divide it into steps, label the routes in, collect the minimum, write reports that end in decisions, and verify that the numbers deserve belief. Change tools and the system moves across intact.
The step after measurement is improvement. For the areas that are hard to take on yourself — speed, indexing, security — our optimization program measures before and after and handles them as one job, and exactly what we measure and hand over is published on the process page. Further reading lives in the SEO archive.