The same search data can look better or worse depending on the dates and the measure you pick. Our case studies show numbers, so this note sets out the rules we follow when we publish them.
Every number carries its source and dates
Each figure names the service it came from, such as Search Console or GA4, and the dates it covers. When we compare, we compare equal windows: on September 28, 2026 we compared September 25–27 with September 22–24, three days with three days.
Recent data is marked
Google's search data for the most recent two or three days can still change. When a window includes those days, the page says so, for example "September 26–27 are preliminary". When the data is final, the page says that instead.
Counts are lower bounds, and terms are defined
Search Console leaves out searches it anonymizes for privacy, so when we write that a site held the #1 position for 40 queries, the true number may be higher. We define the terms where we use them: a query at #1 has an average position below 1.5, and the first page means an average position of 10 or better.
An average can hide where guests find a shop
An average position across every search mixes the searches guests use with rare, long searches far down the results. A site can sit near the top for the searches that matter and still show a middling average. So next to an average we also show counts, such as the number of queries at #1, and we name example queries: the most searched general ones first, never the restaurant's own name or a competitor's.
Clicks, not orders
Restaurant sites send guests to DoorDash, Grubhub, Uber Eats or ChowNow to order. We see the click that leaves the site, not the order that follows. So we write "people who clicked Order", not "orders", and we put no dollar value on a click.
No visitor counts
Raw visitor numbers include automated traffic, some of it from data centres, that no filter removes completely. On the sites we measure, we report actions instead, such as clicks to order or to call.
No before and after without a before
When a site's search data starts on the day we connect it, there is nothing earlier to compare, and we say so. When a new domain really does start from zero, as Cocobang's did on September 22, 2026, we give the registration date.
Timing is not cause
When numbers rise after a change, we write that the rise came after it. A few days of data cannot separate one change from everything else that moves search results: the season, a competitor, a new review, or Google itself.
We choose the measures, not the numbers
A case study has room for three figures, so we choose which true measures to show: a count of queries at #1 can say more than an average position, and a daily figure can show a new site's first days. Choosing the measure is our call. Every figure is the one the source reported for the dates beside it.
The rules, in short
- Every figure names its source and dates.
- Comparisons use equal windows.
- Preliminary data is marked.
- Counts that exclude anonymized searches are called lower bounds, and #1 and first page are defined.
- Clicks toward an order are not called orders and carry no dollar value.
- Visitor counts are not reported.
- No before and after without data from before.
- A rise after a change is described as timing, not cause.
These rules apply to the search and analytics figures in our case studies.
