Not Measured, No AI Answer, Refused and 0%

Not measured, no AI answer, refused and 0% are different states. Learn the denominator rule, coverage and partial SOV with a worked example.

Learning goals

Tells the four states of an engine attempt apart: answered, no AI answer (NO_AI_ANSWER), refused (REFUSED), not measured.
Explains why "no AI answer", "refused" and "not measured" are not counted as 0%.
Tells a measured 0% SOV apart from an unmeasured SOV.
Knows what coverage and partial SOV mean.

In this lesson

Ask an AI engine a question and you won't always get a usable answer. Sometimes the engine produces no AI answer at all, sometimes it declines, and sometimes a technical problem stops the measurement.

Counting these as 0% is the most common, and most costly, mistake in AI visibility measurement. This lesson explains why.

Four states

Every engine attempt gets exactly one of these states:

State	What it means	Measured?	In the denominator of answer-based metrics?
Answered	A usable AI answer came back.	Yes	Yes
No AI answer (NO_AI_ANSWER)	The request worked, but the engine produced no AI answer. Example: Google didn't show an AI Overview for that search.	Yes	No
Refused (REFUSED)	The engine was available but declined to answer.	Yes	No
Not measured	A provider error, timeout or technical failure prevented measurement.	No	No

The rule in one line: only answered responses enter the denominator of answer-based metrics. The other states are not counted as 0%.

Why they aren't zero

0% is a result. It means "the question was asked, an answer came back, and you weren't in it." For SOV, a 0% also needs at least one competitor to have been named.

In the other three states there's no answer to judge:

No AI answer: There's no AI answer for your brand to be in or out of. Google not showing an AI Overview for a search says nothing about your visibility.
Refused: The engine didn't answer that question. That tells you about the engine's behavior, not about your brand.
Not measured: The measurement never happened. You can't record something you don't know as zero.

Counting these as zero makes your visibility look lower than it is. Worse, it turns a technical failure into a story about "visibility dropping".

Worked example (representative)

Fictional brand, fictional numbers.

In one measurement window, 12 attempts are made for Acme Helpdesk:

8 attempts are answered; Acme Helpdesk appears in 2 of them.
2 attempts get no AI answer: Google showed no AI Overview for those searches.
1 attempt is refused.
1 attempt is not measured: the provider timed out.

Right: Discovery Visibility = 2 ÷ 8 = 25%

Wrong: 2 ÷ 12 ≈ 17%. That treats the 4 attempts with no answer as "Acme Helpdesk wasn't there", when there was nothing to look at.

Take it one step further. If a provider outage meant none of the 12 attempts could be measured that week, the right display would be "not measured". Showing 0% would dress up an infrastructure problem as a brand problem.

A measured 0% vs an unmeasured SOV

The same distinction applies to SOV.

SOV is a measured 0% when valid answers came back, a competitor set exists, and competitors were mentioned but you never were.

Representative example: Acme Helpdesk 0, Competitor A 9, Competitor B 4. SOV = 0 ÷ (0 + 9 + 4) = 0 ÷ 13 = 0%. That's a real result, and a real warning worth acting on.

SOV is not measured, and no number is shown, when:

no competitor set is defined,
no valid answer came back,
or neither you nor any confirmed competitor was named in the answers (0/0),
or not enough engines were measured healthily (see below).

In those cases you'll see "not measured", not 0%. Without a valid comparison set, no number is made up.

Coverage and partial SOV

Failed attempts aren't hidden; they show up in the coverage figures, so you can see any gap.

Coverage is checked per engine. An engine counts as healthy for the week when at least 95% of its scheduled questions were measured. "Measured" here means the attempt ended as answered, no AI answer or refused. Not measured lowers coverage.

How weekly SOV is shown depends on how many engines are healthy:

5 of 5 healthy: SOV is complete.
3 or 4 healthy: SOV is shown as partial, with the engine count (for example "partial · 4/5 engines").
Fewer than 3 healthy: SOV isn't shown that week.

So one engine having a bad week doesn't blank the whole measurement, as long as enough healthy engines remain. But a partial SOV shouldn't be read as if it were complete.

Gaps in charts

If a week couldn't be measured, the trend chart should show a gap, not a point dropping to zero. "We fell to 0% last week and bounced back this week" is often just an unmeasured week mistaken for zero.

Official rules: invalid answers are not zero and when SOV is measured.

Summary

Each attempt gets one state: answered, no AI answer (NO_AI_ANSWER), refused (REFUSED) or not measured.
Only answered responses enter the denominator; the others are not 0%. Representative example: 2 ÷ 8 = 25%, not 2 ÷ 12.
A measured 0% is a real result: answers came back, competitors were named, you weren't.
With no competitor set, no valid answer, no one named in the answers (0/0) or too few healthy engines, SOV shows as "not measured".
An engine is healthy at 95% coverage or more; 5/5 is complete, 3–4/5 is partial, fewer than 3 isn't shown.