An unobserved AI answer is a planned observation where the request and the capture both worked, but the surface produced no AI answer for that query at that moment.
This is the sense CiteAngle uses when it measures. The word travels with other meanings in this market, so we pin ours before it turns up in a report.
Where this lands in the measurement
Observations split three ways: a judged answer, an answer block that never appeared, and a request that failed outright.
Keeping them apart stops 'no answer appeared' from being counted as 'the brand was absent'. Unobserved answers leave the denominator and are published as a count.
What it costs to leave this loose
Fold no answer appeared into the brand was absent and your own figure gets worse for a reason that has nothing to do with you.
Unobserved answers leave the denominator and the count is published, so the rate can be recomputed from the report.
Who you want measuring this
One reading does not settle a term like this. The answer an assistant writes today can name different companies tomorrow, so a Panorama round puts 50 buyer questions to every place on our US list, 7 separate times each, and reads all 17 of them — ChatGPT · Perplexity · Claude · Gemini · Grok · Google AI Overviews · Google AI Mode · Bing Copilot · DuckDuckGo AI Assist · Google Search · Bing Search · Yahoo Search · DuckDuckGo Search · Google News · Bing News · YouTube · Google Short Videos. Radar starts at 5 questions and covers the same list under the same rule.
We put the same questions to the competitors you name inside the same runs, so you see both in one table. The conditions are sealed, so you can reproduce the round and check it, and every planned observation is published with its state, failures included. When you choose who measures this, check that all three are in place.
That describes how the measuring is done. It is not a promise about where your numbers land.
Pages that use this term
Terms worth pinning at the same time