Not measured is not zero
Absent, unreachable and measured-zero are three different findings. Collapsing them turns a data-collection failure into a false statement about a real business.
Three states, not two
Measured zero. We looked, and the signal is genuinely absent. A real finding about the business.
Unreachable. We tried and could not read it. A finding about the attempt, which may be transient.
Not measured. This was never checked, because the source was unavailable, out of scope, or not applicable. A finding about the evaluation.
An AI visibility evaluation that renders all three as zero has told a business that something is wrong with it when the truth may be that something was wrong with the check.
Why it always collapses toward zero: a zero fits in the same column as a real score and requires no explanation. Saying "we could not read this" requires a different display, a reason, and someone to fix it. The tidier option is chosen by default, and it is a lie by omission.
What it costs
A business fixes something that was never broken, and the score does not move, so they conclude the work does not help. A transient network fault becomes a permanent-looking gap in a trend line. And a real decline hides inside a column of zeros that were always there.
All three erode the thing that makes measurement worth paying for, which is the ability to trust that a change in the number means a change in the world.
How it appears in a report
Unmeasured signals are shown as unmeasured with the reason attached, and they are excluded from the total rather than counted as failures. A score of 60 out of a possible 80, with 20 points unmeasured and named, is more useful than 60 out of 100, because it tells a business where to look rather than implying a verdict nobody reached.
The same rule about trends
If the method changed between two readings, the comparison is withheld rather than shown. A number that improved because we changed the ruler is worse than no number, and the temptation to show it is strongest precisely when it flatters everyone. This is why every reading carries the version of the method that produced it.
What to ask any tool
Does it ever return "not measured"? A tool with a value in every cell, every time, is either checking nothing that can fail or filling gaps with zeros. Neither answer is good, and the second one is far more common.