Positive, neutral, negative per mention
Score the rung that decides whether engines recommend you. Every extracted mention is classified and aggregated over a rolling 14-day window, and the positive share is what scores Be Trusted on the ladder.
Be Trusted, the fourth rung
Check the sentiment behind every mention
Mention rate counts how often you appear. It says nothing about how. Being named constantly as the expensive option is not a visibility win.
Key takeaway
Mentions are counted linked and unlinked, separately. A brand named often but linked rarely is being discussed rather than recommended, and the two call for different work.
What sentiment tracking actually measures
Classification runs per mention, not per page or per brand, so the unit of analysis is the sentence an engine actually produced.
Split by engine and question type
Find the question type where a competitor's framing is beating yours. A brand can read positively on brand-direct questions and negatively on comparison ones, and aggregating the two hides exactly the problem worth fixing.
Linked against unlinked mentions
Tell goodwill apart from retrieval. An unlinked positive mention is goodwill in the model and a linked one is goodwill plus retrieval, so they move for different reasons and are counted separately.
The evidence behind every count
Read the answer rather than trusting a bar chart. Each classification traces to the probe that produced it, because a sentiment number nobody can audit is a number nobody should act on.
What sits underneath
The aggregate is daily, the classification is per mention, and both are inspectable.
Rolling 14-day aggregation
Long enough to smooth a single unusual answer, short enough that a genuine change in framing shows up within a fortnight rather than a quarter.
Mention rate alongside sentiment
Total probes, probes where the brand was mentioned, and the sentiment split within those. Rate and framing on the same row, because either alone is misleading.
Position within the answer
Where your brand appeared in the response. First recommendation and grudging final mention are both mentions, and averaging them describes neither.
Competitor sentiment
The same classification applied to named competitors, so your framing is read against theirs on the same questions rather than against an absolute scale.
Feeds the Be Trusted rung
The positive share of brand mentions scores rung 4.It sits above Be Retrieved because a page has to be extractable before its framing matters, and below Be Chosen because good framing does not guarantee selection.
Insufficient data reads as insufficient
Below the minimum probe count no sentiment score is published. A brand with four mentions does not have a sentiment trend, and reporting one would be noise wearing a percentage sign.
Why this is measured separately from mention rate
The two metrics fail in opposite directions, and a combined score hides both.
A high mention rate can be bad news
Check the framing before you celebrate the volume. Being named often as the wrong choice is worse than not being named, because the engine is steering buyers away with a specific reason you can read.
Framing is a content problem with a content fix
Turn a vague reputation worry into a specific target. Negative framing usually traces to a comparison page, a review site or a forum thread the engine keeps reaching for, and the co-cited list names it.
Leading indicator, reported as one
Read sentiment as the leading indicator it is. It is published with its noise level and never converted into revenue, alongside citation rate and share of voice.
Against a mention counter
Most AI visibility tools report presence and stop.
Sentiment classified per mention
TrustData
GA4
Other tools
SometimesSplit by engine and question type
TrustData
GA4
Other tools
Linked against unlinked mentions
TrustData
GA4
Other tools
Competitor sentiment on the same questions
TrustData
GA4
Other tools
RarelyEvery classification traceable to a probe
TrustData
GA4
Other tools
| Feature | TrustData | Google Analytics 4 | Other tools |
|---|---|---|---|
| Sentiment classified per mention | Sometimes | ||
| Split by engine and question type | |||
| Linked against unlinked mentions | |||
| Competitor sentiment on the same questions | Rarely | ||
| Every classification traceable to a probe |
FAQ
Direct answers.
Each extracted brand mention is classified positive, neutral or negative and aggregated over a rolling 14-day window. Classification happens per mention rather than per response, so an answer that recommends you for one use case and warns against you for another produces two classifications rather than one averaged verdict. Every classification links back to the probe and the response text behind it.
Because the engine is not merely failing to recommend you, it is actively steering buyers away and giving them a reason. That reason is usually specific, repeatable and readable in the responses: price, support, a missing integration, a comparison that lands badly. Absence is a discovery problem you fix with content and authority. Negative framing at volume is a positioning problem, and it will not improve by publishing more.
Because the pattern is where the information is. Positive on brand-direct questions and negative on comparison questions is the signature of a competitor whose framing is winning the head-to-head, which is a different fix from being framed badly everywhere. Aggregating the two produces a middling number that describes neither and points at nothing.
A linked mention comes with a citation to a URL; an unlinked one is your brand named in the text with no source attached. Unlinked mentions indicate the model holds an opinion about you from its training data. Linked mentions indicate it went and read something. They move for different reasons: unlinked shifts with model refreshes you do not control, linked shifts with content and authority work you do.
Not causally, and TrustData will not present it that way. Sentiment sits in the leading-indicator regime with citation rate and share of voice, reported with an explicit noise level. The sixth rung reports AI Search sessions and their conversion rate as proxy signals, labelled emerging. Causal claims are reserved for paid media, audiences and smart links, where geo, time and platform holdout tests can carry them with confidence intervals.
14-day free trial
See how you are being described
14-day free trial. Sentiment tracking is part of AI Visibility, included in every plan.