Observability
Know what your data is doing before you act on it
Search data has become volatile. Observability measures the data about your data - error rates, sampling, and drop-and-pop episodes - so a tracker artifact never gets mistaken for a ranking loss.
Why does search data need observability?
The volatility is directly correlated with the value this data carries. It is scarce, hard to get at scale, and it shows the forces that reflect, influence and shape consumer behaviour at a scale available nowhere else. The players who claim to own it thrive on information asymmetry and are not inclined to make it easy to collect.
As that volatility has risen, it has become necessary to build data about the data. What is the error rate, the sampling rate, the number of pages fetched, on any given day? Observability answers those questions so that a collection artifact is never mistaken for a ranking loss, and a real loss is never written off as noise.
Observability in search is the practice of measuring second-order data about your first-order search data. First-order data is what you pay for: SERP results and AI search responses. Second-order data describes how that data was collected - error rate, sampling rate, pages fetched, and whether a ranking movement came from the SERP or from the collection process.
The raw record
Row-level events for every collection, the substrate every aggregate metric is computed from.
Observations about the events
Error rate, sampling rate, fetch counts and episode classification, computed daily and broken out by tier.
The path back to the source
The signals needed to trace any metric on screen back to the specific collections behind it.
What is a drop-and-pop?
A drop-and-pop is Google temporarily tanking a domain's rankings for a few days, then restoring them roughly where they were. Customers started noticing a marked increase in May 2026. These views measure the extent, the impact and the volatility, on a rolling 7-day window.
Tier by pattern, volume-weighted
Every tracked keyword lands in one pattern for the period. Stable means operationally fine after imputation. Tentative, eroded and the two loss states separate a brief gap from a real one, and the Drop & Pop column shows what share of each tier saw at least one resolved episode. Reading it by tier matters: a loss in Top 1 to 3 costs far more than the same count in Top 11 to 20.
Prevalence over time, count-weighted
Prevalence is the share of keywords sitting inside a DnP episode on a given day. Every keyword counts once, so this is the view for how widespread an event was. The amber dashed line is a Theil-Sen trend fit on the pre-boundary days, and the shaded band on the right is the last 7 days, where imputation is still settling and the numbers should be read with that in mind.
Prevalence over time, volume-weighted
The same episode qualification, weighted by monthly search volume instead of counted evenly. Comparing this against the count-weighted view is the fastest way to tell whether an event hit your head terms or your long tail. A spike here that is flat in the count view means a small number of high-volume keywords moved.
Position trends by tier
Average and median position per tier across the period. This is the noise test. Damage concentrated in the premium tier is genuine trouble; the same slope across every tier usually means the tracker, not the SERP. Slope is positions per week.
Confirmed lost keywords
Keywords that did not come back. Sorted by search volume, with the baseline median position before the drop, the date it began, and how many prior DnP episodes that keyword had already survived. A high episode count before a confirmed loss is the pattern worth escalating.
Where this sits
In short
Trust the number, or know why you cannot
Every ranking report answers a question. Observability answers the question underneath it: can this report be trusted today? When a position moves, the useful first question is whether the SERP changed or the collection did, and until recently no platform in this market gave you a way to tell.
This is a first version and it will keep changing, because the things worth measuring change every time the engines do. Spurious result rates and sampling percentage are shipping next. If there is a second-order metric your team needs, we would like to hear about it.
See Observability on your own domain
A 30-minute walkthrough on your own keywords and your own data.
See it with your own data.
30-minute demo. We'll run it on your domain - no prep required.