Reading your own traffic data without fooling yourself

13 August 2026 · Measurement

Four numbers describe almost everything you can learn about how a page performs in search: impressions, clicks, the rate between them, and position. Each is an average of something, and averages of averages mislead in ways that are worth learning to spot.

What the four numbers mean

An impression means your listing was included on a results page that was shown. A click means someone chose it. The rate between them is the proportion of people who saw the listing and picked it. Position is where the listing sat, averaged across every time it appeared.

That averaging is the trap. A page appearing at position three for one query and position forty for another is reported at a middling position that describes neither situation. The average moves when the mix of queries changes, even if nothing about your ranking changed at all.

Segment before concluding

Always split by something before drawing a conclusion. By query, so you can see which demand is real. By page, so a sitewide dip resolves into one section. By device, since phone and desktop layouts differ. By country, if you are visible in several. A single line on a chart is the sum of a dozen unrelated stories.

The diagnostic pairs

• Impressions rising, clicks flat. You are being shown more, often for broader queries, and not chosen. Look at the listing text before touching the page.
• Impressions falling, position steady. Demand is falling, or a results-page block has taken the space above you. Not a penalty.
• Clicks steady, position improving. Movement inside a region nobody looks at. Position eighteen to eleven changes very little.
• One page collapses, the rest are steady. Page-level: a duplicate, a retired link, a change to the page itself.

Seasonality and the comparison window

Compare like with like. A month against the previous month conflates the change you care about with the calendar. A month against the same month a year earlier removes the season but adds a year of unrelated change. Do both, and treat any conclusion that depends on which comparison you chose as unproven.

Sampling, thresholds and small numbers

Reports round, threshold and omit. Queries below a certain volume may be withheld, so the sum of your queries will not match your page total, and that gap is not an error. Below roughly a hundred impressions, a rate is noise; do not rewrite a title on the strength of three clicks becoming one.

A monthly routine that takes an hour

Export the last three months by page. Sort by impressions and read the top thirty rows, asking only whether the click rate looks reasonable for the position. Note the outliers. Then sort by clicks lost against the previous period and check whether each loss has a page-level explanation. Fix the two clearest problems and stop. Most measurement time is spent producing charts nobody acts on.

Permalink: /notes/reading-your-own-traffic-data/