Answer engine measurement
You are being cited. Nobody is counting.
Ausnov tracks who answer engines name for the questions your buyers ask. Weekly. Across six engines. Reported as numbers, not as a narrative.
Engines
6
Tracked as standard. Each queried separately, because they disagree.
Prompts
300+
Per client. Built from your sales calls, not from a keyword tool.
Refresh
Weekly
Answers drift. A quarterly snapshot hides most of what happens.
01 / Output
Share of citation, tracked over time
The headline number is simple. Of the prompts we track, what share returns an answer that names you, and how does that compare to whoever is winning.
Citation share, weeks 1 to 12
Share of 300 tracked prompts returning an answer that names the brand.
Illustrative figures. This is the report format, not a result we are promising. Your first reading is a baseline with no trend in it, and it usually takes six weeks before movement means anything.
View as table
| Week | You | Leader |
|---|---|---|
| 1 | 8 | 34 |
| 2 | 9 | 33 |
| 3 | 11 | 35 |
| 4 | 12 | 34 |
| 5 | 15 | 33 |
| 6 | 17 | 31 |
| 7 | 18 | 32 |
| 8 | 22 | 30 |
| 9 | 24 | 29 |
| 10 | 27 | 30 |
| 11 | 29 | 28 |
| 12 | 32 | 27 |
02 / Metrics
Six things we report
Each one is measured. None of them is modelled unless the report says so.
Citation share
Measured
Share of tracked prompts returning an answer that names you. The headline number, split by engine and by topic cluster.
Answer position
Measured
Whether you are the first source named or the fourth. Position inside an answer behaves differently from position in a results list, and it moves faster.
Sentiment of mention
Measured
Being named is not the same as being recommended. We classify each mention as recommended, listed, or cautioned against, and we show you the raw text.
Source attribution
Measured
Which of your URLs the engine actually cited. Frequently it is not the page you would expect, and that finding alone changes what you publish next.
Competitive set
Measured
Who else appears in your answers. This is often the most useful page in the report, because the set an engine considers your peers rarely matches the set you track.
Referred sessions
Partial
Traffic that identifies an answer engine as its source. Labelled partial because referrer data is stripped often enough that this number is a floor, never a total.
03 / Method
How the number is produced
-
Build the prompt set
From your sales call recordings, support tickets, and win-loss notes. Three hundred prompts in the language buyers actually use. You approve the set before anything runs.
-
Query cold
Fresh sessions, no history, no personalisation, from neutral IPs. A logged-in account returns a different answer, which is how most internal spot checks end up wrong.
-
Classify
Every response is parsed for named brands, cited URLs, and sentiment. A sample is checked by hand each week to keep the classifier honest.
-
Report the delta
Weekly. What moved, what did not, and which changes are inside normal variance. Answers are noisy. We show the noise band rather than pretending a two point move is a trend.
04 / Limits
What we will not tell you
Every measurement vendor in this category should publish this list. Most do not.
Attributed revenue
Answer engines strip referrer data. Anyone reporting AI-sourced revenue to the dollar is modelling it and calling it measurement.
Why you moved
We can show that citation share changed. Causation needs a controlled change on your side, and we will help you design one.
What a model knows
We observe outputs. We do not have visibility into training data, retrieval indexes, or ranking internals, and neither does anybody selling you a dashboard.
A guaranteed direction
Reporting is reporting. If your number falls, the report says it fell and shows you where.
Get a baseline before you spend anything on this channel.
One prompt set, one reading, fixed fee. If it shows you are already winning your question set, that is a useful answer and a cheap one.
shaban@ausnov.com