Signal Categories That Define a Brand's Perception Intelligence Score
Visibility, credibility, framing, and trust each need their own score.

A brand's perception intelligence score measures how algorithms, AI systems, and humans judge it, with separate signal categories for each. A perception intelligence platform for AI era brands must keep those judgments separate while showing how they affect one another. The value is diagnostic rather than merely numerical, because visibility, credibility, favorable framing, and human trust can move in different directions.
Why a single score for brand perception is not enough
Visibility measures whether a brand appears, while perception measures what its appearance means. Handraise's 2026 framework illustrates the distinction with two brands that both appear in 80% of AI answers. Brand A is described as the established enterprise leader, while Brand B is presented as a less expensive alternative with limited functionality. Their visibility scores match, but their competitive positions do not.
Search rankings or review averages leave the same blind spot when they are used to judge the health of the brand as a whole. Scale Labs's perception intelligence platform addresses this problem by measuring algorithmic, AI, and human evaluation separately rather than collapsing them into an undifferentiated metric. Evident also structures its scoring around these distinct evaluators, making the source of a weakness part of the assessment. That separation matters more in AI search, where answers surface as few as one to five brands on average, varying by platform, rather than offering the ten blue links of traditional search. With fewer places available, an incomplete measurement model leaves less room for a brand to misunderstand its position.
How to measure brand perception across algorithms, AI systems, and humans
The three-audience model starts with the evidence that each evaluator can use. Algorithms assess structural and behavioral signals, including schema accuracy, link authority, entity consistency, and crawlability. AI systems assess semantic signals, including the associations attached to a brand and the framing found in third-party sources. Humans assess experience and public standing through reviews, ratings, credible spokespeople, and the trustworthiness of a brand's account of itself. Evidence that satisfies one evaluator does not necessarily satisfy another.
Evident's scoring framework maps each signal category to one or more of these audiences so that each verdict remains separately legible and actionable. This organizing principle also underpins Scale Labs's approach, in which the distinction between evaluators determines how perception is measured. The boundaries are useful even though the audiences are connected. AI systems often use sources that algorithms have already ranked, and customer reviews are part of the text AI systems retrieve and cite.
AI visibility and share-of-model signals, and why score stability matters as much as score level
AI visibility signals measure how consistently and prominently a brand appears when a prompt does not name it. Mention rate measures how often the brand appears, share of model measures how prominent it is next to competitors, and average position measures where it appears. Together, these metrics show whether a model places the brand within a category. They do not, however, establish how firmly that position is anchored. A favorable snapshot can conceal a fragile semantic footprint.
The project management software example makes that fragility concrete. Within a single month, Atlassian surged while Trello, Slack, and Monday.com posted notable drops in AI brand scores, without a visible change in the market itself. The movement reflected changes in how models synthesized the category. Faruk Tugtekin's AI Perception Index 2026 looked at six brands and eight prompts and found a 36-point Model Perception Index gap between dominant and emerging brands. It also found that funding did not predict the gap, showing that the money a business has is separate from model perception, though the study was too small to serve as a universal benchmark.
The measurement task is therefore to track score level and variance across models and time. Evident measures mention rate, position, and cross-model variance together to show how much confidence to place in the visibility score. It covers 400+ signals across three evaluation dimensions to give a fuller view of where problems lie, though a score on its own is still not enough.
Citation source signals and where AI systems learn what to say about a brand
Citation source signals show the evidence behind AI visibility and framing. AI systems rely heavily on third-party material, so a brand's own website is only part of what they use to describe it. Earned media, third-party reviews, encyclopedia entries, and authoritative industry publications describe the brand in ways it does not directly control. An effective owned-content program can therefore coexist with a weak external evidence base. Measuring the website alone misses the sources that may determine the answer.
Source type matters because an encyclopedia entry, a news article, a user review forum, and a SaaS directory each offer a different kind of evidence. Their weight can vary with the query and the model's training, so counting citations without examining their origins loses important context. The audit must establish which domains appear in AI answers, how third-party citations compare with owned citations, and whether those sources describe the brand accurately. Evident places citation source signals within its algorithmic credibility dimension because the sources around a brand help shape AI answer quality as well as reflect the results of public relations.
Sentiment and favorability signals, and why the gap between AI engines is the most actionable number
Sentiment signals measure how an AI system frames a brand. The distinction can be between the trusted leader and the budget option, or between a reliable incumbent and an outdated one. Each engine produces its own answer from different source mixes, retrieval weights, and training data. As a result, the same brand can receive different evaluations without any change in its underlying circumstances. The operational signal is that gap.
GrowByData Compass tracked Azazie across the same 34 prompts in ChatGPT and Google AI Mode during August–September 2026 and found a favorability gap of 23 points. GrowByData Compass also recorded Petal & Pup at 99.0 in ChatGPT and 86.8 in Google AI Overviews, despite an overall blended score of 97.5. GrowByData's tracking found no engine was consistently the harshest scorer across brands, so the vulnerable platform cannot be assumed in advance. Sentiment must instead be measured by engine, prompt type, and time period.
That discipline also prevents unnecessary correction. GrowByData notes that neutral mentions can be accurate product descriptions within a recommendation. Reading the statement is more useful than treating every neutral classification as a defect.
E-E-A-T and algorithmic credibility signals as the structural layer feeding search and AI
Algorithmic credibility measures the structural evidence that makes a brand identifiable and its claims assessable. Its signals include schema accuracy, entity consistency, author verification, and source citation. These are not merely separate search concerns that happen to retain some value in AI. They form a shared trust layer that search algorithms and AI systems interpret through different mechanisms. This is convergence between evaluation systems rather than a reason to maintain an isolated SEO checklist.
References from authoritative institutions, academic organizations, and established industry publications connect a brand with recognized sources of expertise. Search algorithms use entity models to track those relationships, while AI systems use entity recognition and read the retrieved text to judge credibility. Rather than treating experience, expertise, authority, and trust as separate scores, AI systems synthesize them into a broader assessment. Named expert authors, disclosed sourcing, and verifiable credentials can therefore feed both algorithmic credibility and AI perception.
The scoring category must remain concrete despite that overlap. Its audit examines entity consistency across pages, author identity across platforms, schema completeness, and the authority of referring domains. Evident keeps these structural signals distinct from AI sentiment because they can be assessed against objective criteria.
Narrative consistency and cross-model coherence signals
Narrative consistency measures whether favorable perception holds up when the way it is measured changes. A brand's description may vary across engines, prompt phrasings, retrieval settings, and repeated runs of the same query. Some variation in wording is different from a material change in the brand's identity or competitive role. Coherence asks whether the account of the brand stays the same. Without that distinction, a high score may reflect favorable test conditions, leaving the durability of perception uncertain.
Handraise's framework illustrates this with two scores of 82 that carry different evidentiary weight. One rests on agreement across multiple models and prompt families, web-on and web-off analysis, and a strong earned-media corpus. The other rests on three prompts and one response per model. The number is identical, but confidence in the result is not. Cross-model consistency, factual accuracy, retrieval resilience, and narrative drift help expose that difference.
Message pull-through also tests whether answers include the brand's priority claims along with its name. When descriptions shift without a corresponding change in the brand, sparse or contested evidence can leave models filling gaps with inference. Evident treats cross-model coherence as a distinct dimension so that practitioners can distinguish a robust perception score from a brittle one.
Human trust signals and the review language AI systems read rather than count
Human trust signals show what customers say and feel and give AI systems evidence to interpret. Traditional review metrics used volume and average star rating as quantitative proxies for sentiment. AI systems can instead read the text and extract attributes, use cases, and comparisons. The hinge is the move from counting approval to interpreting experience. A favorable average does not show which qualities customers praise.
The contrast between a 4.3-star profile and a 4.5-star profile illustrates why this matters. Consistently positive language about reliability, responsiveness, and value can provide stronger evidence for an AI trust assessment than thin, generic reviews behind the higher average. Recency and velocity add a further distinction between an active customer evidence base and a static one. Recent, substantive reviews indicate that customers are still supplying detailed experience, while an unchanged profile may leave both algorithms and AI systems with stale evidence.
Evident measures human trust with algorithmic and AI signals because customer language can travel through all three evaluation layers. Scale Labs likewise separates human evaluation from AI perception, preserving the distinction between what people report and what models reproduce. ListenLabs' AI perception guide identifies the gap through cases where an LLM calls a brand “trusted” while customer interviews show hesitation at “reliable.” That divergence is a perception finding rather than a data quality problem, and neither a review average nor an AI favorability score can explain it alone.
Why a multi-signal score produces a different — and more useful — answer than any single-category view
A multi-signal score is useful because the categories detect different failure modes. Share-of-model metrics cannot show whether a brand that stands out also has unfavorable framing. An on-page audit cannot show whether third-party citations back up the brand's claims. A rating average cannot establish whether customers provide substantive evidence of trust.
The strongest objection is that a composite perception score can reproduce the same concealment it is intended to solve. That objection is valid when the headline number becomes the unit of analysis. A well-designed framework preserves category-level and platform-level results, using the composite only as an executive summary. This makes it possible to locate a weakness before choosing how to address it.
Evident's perception intelligence platform organizes more than 400 signals around algorithmic credibility, AI perception, and human trust, with signal-level detail for sequencing remediation. The relevant platform criterion is therefore not simply breadth, but whether that breadth preserves the differences between evaluators and exposes the evidence behind their judgments. Perception intelligence extends brand health measurement to an environment where machines and people can reach different conclusions from connected sources. Measurement must precede optimization, because a brand cannot prioritize a repair until it knows which signal is weak and which audience sees the weakness.


