AI NativeWhen an AI engine mentions your brand, the framing around that mention is not neutral by default. The answer might praise you, qualify every statement with a warning, position you against a competitor unfavourably, or describe complaints without balance. Sentiment is the metric that captures that framing as a number, and it is distinct from whether you appeared at all.
Two layers of sentiment
Sentiment in AI Native is measured in two separate layers, because the problem has two distinct causes.
The first layer is in-AI sentiment: how the answer itself frames your brand. The classification reads each answer where your brand is present and assigns a label and a numeric score. The label is one of four states: recommended, neutral, hedged, or negative. Recommended is the strongest positive framing; negative is an answer that treats your brand as a warning or a problem. Hedged covers answers that are technically present but surrounded by caveats, complaints, or qualifications that a buyer would read as discouraging. The numeric score runs from -1 to 1, where positive values are net-positive framing and negative values are net-negative.
The second layer is web-source sentiment: the polarity of the web content the AI engines draw on when they form answers about your brand. This matters because AI answers are not generated in a vacuum. The engines read and cite web sources, and the cumulative tone of those sources shapes what the engine says. A cluster of negative reviews on third-party sites, complaint forums, or critical press articles pulls the in-AI sentiment down, often before you have noticed the web narrative shifted. Web-source sentiment shows you the upstream cause, not just the downstream effect.
How the score is computed
Each answer that contains your brand presence produces a single sentiment score between -1 and 1. The overall sentiment score for a product is the mean of those individual answer scores, covering only answers where your brand was actually present. Absent answers are excluded: you cannot have a framing where you were not in the frame.
The score is then decomposed. You see sentiment broken down by engine, by funnel stage, and by prompt class (branded versus unbranded). That decomposition matters because sentiment is rarely uniform. An engine that is consistently neutral about your brand in awareness-stage answers might be hedged in comparison and decision-stage answers, which is where it does the most damage to conversion intent.
What the four sentiment labels mean
Recommended means the answer actively put your brand forward with positive framing. It is not just present; it is presented as a good choice or the right answer.
Neutral means the answer named your brand without endorsing or criticising it. You were included but the engine did not take a position.
Hedged means the answer included your brand alongside caveats, complaints, or qualifications that a buyer would read as discouraging. The brand is present, but so is the reason to hesitate.
Negative means the answer framed your brand as a problem, a warning, or an inferior choice. This is the framing you most want to understand and address.
The highest-priority signal: negative and inaccurate
When an answer is both negatively framed and factually inaccurate, that is flagged separately from either signal alone. A hedged answer with correct facts is a reputational issue. A negative answer with wrong facts is a reputational issue and a factual error, and the two compound. The platform surfaces these combinations at the top of the reputation actions list because correcting a factual error on a negatively-framed branded answer removes both a perception problem and a wrong claim in the same action.
Negative drivers: sources behind the framing
For answers classified as negative or hedged, the platform identifies the web domains that appeared as citations in those answers. These are the sources the engine was drawing on when it framed your brand negatively. They are ranked by how often they appear in negatively-framed answers, and they are the starting point for any source-level response: whether that is outreach, a response, new positive content that earns citations from those domains, or a review-management effort.
Competitor framing in negative answers
When a competitor is present in an answer where your brand is framed negatively, that is recorded separately. This is the pattern where the AI frames you as the weaker option: your brand appears, is hedged or criticised, and a competitor is named in the same breath. The frequency of this pattern by competitor tells you which rivals are most consistently benefiting from your negative framing.
Questions
What is the difference between a low recommendation rate and negative sentiment?
A low recommendation rate means you are not being chosen as the top option. Negative sentiment means the framing around your brand is discouraging. Both can be true simultaneously and they have different causes. A brand can appear often and be placed in lists (high mention, reasonable ladder position) but described with caveats every time (negative sentiment). The fix for low recommendation is different from the fix for negative framing.
Can web-source sentiment be positive while in-AI sentiment is negative?
Yes, and it is a diagnostic signal when it happens. If the web-source layer is predominantly positive but the in-AI layer is negative, the engine is either drawing on a specific subset of negative sources that outweighs the positive, or the model's parametric training contains negative associations that are not grounded in the current web. Each case has a different response path.
Does sentiment apply to unbranded prompts?
Sentiment is measured on answers where your brand is present. If your brand does not appear in a response to an unbranded prompt, there is no framing to score. When your brand does appear in an unbranded answer, the sentiment of that framing is recorded. This matters most at the consideration stage, where an unbranded category question can produce an answer that names you but hedges immediately.
How do I fix hedged sentiment?
Hedged framing usually traces back to specific sources in the citation set. The platform shows you which domains are cited in hedged answers. The most direct response is usually to reduce the weight of those sources in the answer ecosystem: earn more citations from positive primary-source content, respond to reviews on the domains driving the hedging, and build out content that gives the engine a clearer, more favourable framing to draw on. Schema markup and knowledge-panel accuracy also matter for branded answers.
Is sentiment the same across all engines?
No. The sentiment breakdown by engine shows you whether the framing varies across ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews. Engines draw on different source sets and have different training bases, so it is common for sentiment to be neutral on one engine and hedged on another for the same product. That difference points to where the upstream source problem is concentrated.
Where can I read about what the engines are measuring?
See Engine coverage FAQ for which surfaces AI Native covers and Accuracy and fact checks for how factual errors are caught separately from sentiment framing.