Anomalies Snapshot

What it is

The Anomalies Snapshot highlights up to five days inside the currently selected reporting period where one of your two top-line metrics — clicks or impressions — deviated unusually far from the period's own typical day.

It is purely a statistical alarm on your existing trend data: no machine learning, no external benchmarks. The same days that look like spikes on the trend chart will surface here, ranked by how unusual they are rather than by absolute size.

How we compute it

1. For the selected date range, we collect the daily clicks and impressions reported by Search Console for your site.
2. Across those days we calculate two summary numbers per metric: the average (μ) and the standard deviation (σ) — a measure of how much a typical day varies from the average.
3. Each day is then scored by how far it sits from the average, in standard-deviation units: z = (day's value − average) / standard deviation. This is the z-score.
4. Days where |z| ≥ 2.0 on either clicks or impressions are flagged — these are days more than two standard deviations away from the mean.
5. If a day trips on both metrics, we keep the one with the larger absolute z-score so each day appears only once.
6. Flagged days are sorted by |z| (most unusual first) and the top five are shown.

How to read each tile

• Spike — a day with an unusually high value on the louder metric.
• Dip — a day with an unusually low value.
• σ chip on the right (e.g. +2.4σ) — how many standard deviations away from the mean. Bigger absolute value = more unusual.
• Date — the calendar day the anomaly was observed.
• Detail line — the actual metric value alongside the period's baseline (e.g. 1,240 clicks vs ~480 baseline).

The header summary line shows the sample size (number of days analysed) and the baseline clicks/day used for the calculation.

What to do with it

Anomalies are the start of an investigation, not the end. A few useful next steps:

Caveats & limits

• Short ranges are noisy. With fewer than ~10 days, σ becomes very small and almost any movement looks like an anomaly. Prefer ranges of 28 days or longer.
• Trend periods bias the baseline. A range that already contains a permanent step-change (e.g. a new section launched mid-period) will pull μ toward the new normal. Slice the range around the change for a cleaner read.
• Weekly seasonality isn't decomposed. Sites with strong weekday/weekend rhythms may flag every Sunday as a dip; cross-check with the trend chart's weekly view before drawing conclusions.
• "Both metrics" days collapse to one tile on the louder metric — open the trend chart for the full picture.