When a number is missing
Why the platform says not measured, excluded or unmeasurable instead of showing you a zero, and what each of those states means.
You will see "not measured", "excluded" and "unmeasurable" in this product more often than in most tools. That is deliberate. A zero is a claim that something was measured and found to be nothing. Most of the time what actually happened is that nobody measured it, and saying so is the difference between a number you can take to a board and a number that quietly flatters you.
What you'll learn
- The difference between zero, null and excluded
- Why a change can be shipped, live, and still unmeasurable
- What to do about each state
An excluded engine is not a zero
An engine is excluded from your rate when nothing it returned in the window could be bound to a source. The panel reports the exclusion and how stale it is, rather than a rate.
The reason this matters: 0% and "excluded" look similar on a chart and mean opposite things.
- 0% means the engine was asked, answered, and never named you. That is a real and useful finding.
- Excluded means there is nothing to draw a conclusion from. Acting on it as though it were 0% would send you off to fix a problem nobody has established you have.
The same rule runs through movement. An engine with no rate cannot have a movement, and an engine whose previous period was excluded has nothing to have moved from. Both come back as no movement rather than as a change of zero, because zero already means something else: it means nothing moved.
An absent AI Overview is not a failure to be cited
Google does not draw an AI Overview for every query. When there is no panel, there is no answer you could have been cited in, so the check is recorded as an absent overview and left out of your citation rate entirely. It is a fact about Google's results page, not a fact about you.
A change can be live and still unmeasurable
The Changes lens climbs a ladder: proposed, attested, observed, confirmed. Confirmed means the answers moved, scored against a control captured the moment you shipped. A change that cannot be scored is marked unmeasurable with a named reason:
| Reason | What happened |
|---|---|
| No control | Nothing captured a baseline before the change shipped |
| Control metric mismatch | The baseline measured a different metric from the outcome, so the difference is not a difference |
| No outcome | Nothing has been measured since |
| Not observed | Our own fetch has not confirmed the change is live on the page |
| Sample too small | Fewer than eight checks since it shipped |
| Too few engines | The checks landed on fewer than two surfaces |
Unmeasurable is not a zero and not a failure. It is the product refusing to report a measurement nobody made. Most of those reasons are fixable by waiting, and two of them, no control and control metric mismatch, are fixed by shipping the next change through the ledger rather than outside it.
There is one state ranked deliberately above confirmed: drifted. That means we saw the change live, and the page has since stopped satisfying it. A confirmed measurement whose page has since drifted must not read as a clean win, so it does not.
Attested is not landed
If you tell the platform a change shipped, that is attested: your word, recorded as your word. Landed means our own fetch saw it live on the page. The distinction is kept everywhere it matters, including in the counts of how many changes actually turned the wheel, because a plan that counts intentions is a plan that drifts from the site.
Other honest empties you will meet
- No prompts yet. The trace says so in as many words, rather than reporting a rate of zero across a workspace nothing has been asked about.
- A plan gate. The Intelligence layer, Trace and Industry Pulse are Scale and Enterprise features. If your plan does not include one, you get told which plan does, not an empty chart.
- A thin window. A figure computed from too few checks is withheld rather than published with a confidence nobody can defend.
What to do about it
- Excluded engine? Check the prompt set is answerable on that surface, and give it another cycle. Persistent exclusion on one engine while others report normally is worth a support conversation.
- Unmeasurable change? Read the reason. Waiting fixes most of them. Ship the next one through the ledger so a control is captured for you.
- No movement? Look at whether both halves of the window had a rate at all before concluding anything is flat.
Try this in Outercite
Ask the assistant "What is my citation rate on each engine?" and read the rows that carry an exclusion rather than a percentage. Those are the surfaces where you have no measurement yet, which is different from the surfaces where you are absent.
Related
Changes
The ledger, the tiers, and how a change gets to confirmed.
Platforms
Reading per-surface rates, including the exclusions.
Data definitions
Exactly what each metric counts, and over what window.
Last updated on