How to investigate a change in search traffic
Turn a rising or falling chart into a question you can investigate. Use a clear baseline, consistent filters and a record of what changed.
On this page
A chart can show a change without explaining its cause. A more useful investigation starts with one question, a repeatable comparison and a record of what happened on the website. This is a method you can apply; it does not present a new experiment as if we have already run it.
Define the question before looking for a win
Choose a page or a small group with the same purpose. A question such as “Did the revised delivery information help this product page attract relevant visits?” is more useful than “Did SEO work?” Write down what changed, when it changed and which outcome you expect to observe.
Separate that expectation from the evidence. Search impressions, clicks and orders measure different things. A rise in one does not automatically establish a rise in another.
Preserve the baseline and filters
Record the Search Console property, search type, reporting dates and country, device, page or query filters. Save the original export privately. Use equivalent reporting periods where possible, and record why you chose them. Do not compare a short incomplete period with a full month as if they were equivalent.
Record page-level figures separately from any query-specific sample. The queries visible in a report may not account for every click in its totals. Google's performance report documentation explains its metrics, filters and data limitations.
If there is little data, make the qualitative work useful first: can the intended customer understand the page, is it discoverable through internal links, and is the next action clear? A tiny numerical difference is not a dependable verdict.
Keep a change log and a comparison
Write down copy edits, redirects, product availability, campaigns, site outages and relevant seasonal changes. If a comparable page remains unchanged, it can provide context. It is not automatically a controlled experiment: audiences, search demand and competitors may differ.
Where practical, make one coherent change at a time. If several things changed together, report the combined change instead of assigning the outcome to the tactic you prefer.
Use this research log
| Field | What to record |
|---|---|
| Question | The audience, page and specific change being investigated. |
| Baseline | Dates, report, filters, original export and relevant figures. |
| Change | What was edited, by whom and when. |
| Comparison | Equivalent dates or comparison pages, with the reason for choosing them. |
| Outcome | Observed clicks, impressions and separately measured business outcomes. |
| Other factors | Promotions, seasonality, availability, tracking changes and other work. |
| Conclusion | What the evidence supports, what it cannot establish and the next action. |
The downloadable workbook supports the page baseline and action list. Keep a separate dated research note for the fuller explanation.
A fictional example of an honest conclusion
Suppose an invented repair business adds a clearer service-area explanation and an easier quotation route. Its page receives 18 search clicks in the following comparison period, up from 12. Two enquiries are recorded instead of one. These figures are fictional and are not performance benchmarks.
An honest conclusion is that the measured counts increased after the changes, while the sample is small and more than one aspect of the page changed. It would be premature to claim that the new button caused a 100% increase in leads. Review enquiry quality and other influences, then decide what to test or improve next.
Investigating AI citations
Use a fixed set of genuine questions and record the service used, date, prompt, location/context and cited URLs. Preserve the answer where permitted so a later reviewer can understand your observation. Repeat the sample consistently, but expect generated answers to vary.
A citation in that sample is an observation, not a count of all users or a guaranteed position. Google says its AI search features rely on normal search eligibility and do not require special AI files or schema. Google's AI search guidance.
Read a real measurement example
Our Fleetalyse analysis explains a real reporting period and distinguishes measured search figures from the owner's account of sales. The content and measurement path connects this method to choosing useful topics and outcomes.
If you want help implementing changes, our sister agency SuffolkWeb offers scoped website, search and automation work. Both publications are operated by Fleeta Limited. You can use this method and the workbook without buying a service.
