Skip to main content
Experiments answers the question after you publish a fix: did the change work, or did your competitors move? Every published change is measured against them, over the same questions and the same days. You do not create an experiment here. One opens when you mark a gap published on Gap detail.

​
Experiments

​
What is on it, top to bottom

  • The summary: how many changes you have published, how many are still measuring, how many are confirmed (their effect sits outside its noise zone), and the best result so far. Until any experiment has a result, that place shows What happens next instead: the next experiment waiting on an engine. A page has to be read by an engine before it can be measured, and the line says when the next check is.
  • Every published change: one row per experiment, with its Window (how far it has got, how many days it still needs, and when the next read is), its Net effect and its Verdict.
  • How an experiment is read: a short explainer of the three ideas below.

​
Verdicts

​
Experiment detail

One change, in full.
  • The header: the change, the segment it is measured on, the day you published it and the day an engine first read it. A line states the scope: which questions, over which days, and how many competitors the baseline is the median of.
  • The readout: the net effect against its noise zone, the smallest effect this segment can show at its current sample, and the two numbers it comes from.
    • Your change: How much your visibility on this segment’s questions moved, from before you published to after, in percentage points.
    • Competitor baseline: The median change across the competitors you track, over the same questions and the same days. It is roughly what would have happened to you anyway.
    • Net effect: Your change minus the competitor baseline: the part of the move that was yours. If it is larger than its ± range, the change worked or hurt; if not, we cannot tell yet.
  • Which days count: a strip of the before window, the days between publishing and the first read, and the after window. Nothing is drawn past the last completed day; the rest of the window is empty because it has not happened.
  • By engine: your visibility on each engine, before and after.
  • Drift in these windows: any day an engine changed how it answers. Those days are labelled and kept: a move every engine made appears in your change and in the baseline over the same questions and days, and cancels in the net effect.

​
Which days count

Before is the period before you published. After starts the day an engine first reads your page, not the day you published it, so the days in between count for neither.
That is why an experiment can sit at Awaiting first retrieval for a while after you publish: until an engine reads the page, there is nothing to measure. The first day each page was read is on Sources, under Your pages.

​
What you can do

Open an experiment from the list, and open Gaps from the empty state. Closing or annotating an experiment is not available yet. A claim correction is not measured here: it is followed on Claims.

​
Empty and error states

  • No experiments yet: an experiment starts when you mark a gap published. Open Gaps takes you there.
  • No such experiment: it may have been closed, or it belongs to another project.
  • Until a result can be read, the readout says when it will land rather than showing a number.

Gaps

Where an experiment starts.

Within noise

Why an effect inside its range is “we cannot tell”.

Drift labels

How an engine-wide move is handled.

Drift

The method behind the baseline.