Answer engine visibility

Being on page one matters less when nobody reaches page one

Buyers increasingly ask an assistant and act on the answer. Noverva runs your prompts against the answer engines on a schedule and records what actually came back — who was named, who was cited, and which sources the answer trusted.

Run your first check See pricing

How a run works

A measurement, not a vibe

Each run is a recorded experiment: a fixed prompt set, a fixed set of engines, a timestamp, and the raw answers kept exactly as they came back.

Define prompts

The questions your buyers actually ask, in the locales they ask them. Start from suggestions generated off your own site, then edit them.

Run on a schedule

Daily, weekly or on demand, against the engines you enable. Every run records which engines were asked and which answered.

Store the evidence

Raw answers are written once and never modified. Metrics are recomputed from them, so an extraction improvement re-scores history instead of losing it.

Extract and compare

Mentions, citations, competitors and source domains are extracted, versioned by extractor, and compared against previous runs.

What is tracked

Eight signals, each with a defined meaning

Every one of these is defined precisely in the product documentation, because a metric nobody can define is a metric nobody should act on.

Mention

Were you named at all in the answer, in any form — brand, product or domain?

Citation

Did the answer link to you, and to which URL? Linked and merely named are different outcomes.

Position

Where you appeared among the vendors named — recorded only where the format makes it measurable, and left null where it does not.

Competitors

Who else was named, and how often. Share of voice is only meaningful against the field.

Source domains

Which sites the answer leaned on. This is usually the most actionable output: it tells you where to earn a presence.

Sentiment

How you were characterised, where the answer says enough to judge. Reported with low confidence when it does not.

Answer drift

How the response to the same prompt changed between runs, so a sudden drop has a diff behind it.

Prompt coverage

How much of your prompt set actually ran. A partial run is reported as partial.

The uncomfortable part

Answer engines are not a clean data source

Answers are non-deterministic, personalised, regional, and change without notice. Any tool claiming a precise, stable AI ranking is overstating what the medium can support. Here is exactly what Noverva does about that.

Runs are samples, and are labelled as samples

One answer is an anecdote. Noverva reports trends across repeated runs and shows the sample size next to every rate.

A failed engine is not a zero

If an engine rate-limits, errors, or is not enabled, it is recorded as not-answered and excluded from the denominator. It never quietly counts as "did not mention you".

Grounded and ungrounded models are not comparable

A model without retrieval cannot cite. Noverva excludes it from citation metrics rather than scoring it zero, and tells you why it was excluded.

Terms are respected

Each engine is queried through a supported interface within its terms. An engine that forbids automated querying is listed as unsupported — not quietly scraped.

Reading the numbers

Every figure carries its provenance

How AI visibility figures are sourced
Figure	Provenance	Because
Mention rate	Measured	Counted from stored answers.
Citation rate	Measured	Counted from citations the engine returned.
Share of voice	Measured	Counted across answers that named any vendor.
Sentiment	Modelled	Inferred from the answer text, with a confidence value.
Estimated exposure	Modelled	A projection. Nobody publishes assistant query volume.