Citations and sentiment

How a citation is detected, how answer position is measured, and how sentiment is classified.

Three measurements are extracted from every stored answer: whether you were cited, where in the answer you appeared, and how you were characterized. This page explains how each is determined and where each can be wrong.

What counts as a citation

A citation is recorded when the answer references your tracked domain. That happens in more than one form, and all of them count:

A link to your domain, including a link to any page on it.
A named source in an engine's reference list, which retrieval-heavy engines such as Perplexity and the Google surfaces attach to answers.
A brand mention in the prose, where the engine names your product or company while answering.

Matching is done on the registrable domain, so rivarise.com, www.rivarise.com and blog.rivarise.com all resolve to the same tracked domain.

A mention without a link still counts. In generated answers most brand references never carry a link, so requiring one would understate visibility badly.

What does not count

A mention of a competitor that happens to include your category name.
A generic reference to your industry with no brand or domain named.
Content behind a login that the engine could not read, which simply never appears.

Citation rate

Citation rate is the share of checks in the selected range where a citation was recorded:

citation rate = checks with a citation / checks completed

Two details matter for reading it correctly:

The denominator is completed checks, not attempted ones. If an engine was unavailable, those checks are excluded rather than counted as misses, so an outage does not deflate your rate.
It is computed across every prompt and engine in scope. Filtering the dashboard by engine or range recomputes it for that slice, which is the fastest way to find your weakest engine.

The Visibility Score shown on the Overview is this citation rate expressed as a percentage.

Answer position

When you are cited, Rivarise records where you appear relative to the other brands named in that same answer. First mention scores best, and later mentions score progressively lower.

The dashboard reports this as average rank in answer, for example #2.8, meaning that across the answers where you were cited you appeared on average slightly below third position.

Position is measured within the answer only. It has nothing to do with your Google ranking, and the two frequently disagree, which is the entire reason to measure it separately.

Sentiment

Each citation is classified into one of three classes based on how the answer characterizes you:

Class	What it looks like
Positive	Recommended with supporting reasons, described as a good fit, or named as the top choice.
Neutral	Listed among options without advocacy, or described factually with no judgement.
Negative	Caveated, warned against, described as expensive, limited, hard to use, or unsuitable for the asker.

Sentiment is computed from the answer text by a language model, and it is available from the Starter plan upward.

Reading sentiment honestly

A single classification can be wrong. Treat the distribution over dozens of answers as the signal, and open the stored text before acting on any individual one.
Negative is usually specific and fixable. In practice most negative mentions trace back to one identifiable source: an outdated price, a removed feature, or a single old review the engine treats as representative.
Neutral is not failure. For a new entrant, moving from absent to neutral is real progress and often the realistic near-term goal.

Competitor detection

Rivarise also records which other domains were cited in the same answer. Competitors you have explicitly added are matched by domain and tracked over time as share of answer. Brands the engine names that you have not added still appear in the stored answer text, which is how most people discover the rivals they were not tracking.

When the analysis is wrong

Automated extraction has failure modes, and knowing them saves you from arguing with a number:

Ambiguous brand names. A product named after a common word can be matched where the engine meant the ordinary word. Domain matching is more reliable than name matching for this reason.
Hedged recommendations. "It is powerful but has a steep learning curve" is genuinely mixed, and any three-way classification will flatten it.
Implicit mentions. An engine describing your product without naming it will not register as a citation, and should not.
Truncated answers. If an engine returns a shortened response, later positions in that answer do not exist to be measured.

The stored answer text is the ground truth. Every metric on this page is derived from it, and it is kept for your plan's full history window precisely so you can check the derivation rather than trust it.

Move on to the app reference, starting with the Overview dashboard.