Methodology

The Sentiment Triangulation Index is the headline reputation signal. It merges Trustpilot, Google Maps reviews, Reddit, Hacker News, and what AI engines now repeat about the entity into a single confidence-scored sentiment value, while preserving each source’s individual contribution.

The index does not average tone — it triangulates it. Where sources agree, confidence rises. Where they diverge, the index flags the disagreement explicitly: e.g. "Trustpilot positive, Reddit critical, AI engines neutral" with a low consensus weight.

This is what makes it different from a star rating: it tells you whether your customer reviews tell the same story as your community discussion, and whether the AI engines are repeating the consensus or one specific source.

Narrative Drift measures the rate of change in dominant themes, sentiment polarity, and citation framing across all observed sources, against the entity’s rolling baseline.

A high drift score means perception is moving — fast — and surfaces which sources are leading the change (e.g. "AI engines drifted negative seven days before SERP did"). A near-zero drift score means perception is stable.

Drift is computed per source category and globally, so you can see whether the change is concentrated (one platform) or systemic (everywhere).

Cross-Source Consensus measures how aligned the 8 signal categories are at this moment — AI engines, identity, reputation, community, news, search, geography, and NLP.

High consensus means the picture is consistent: every source category is telling roughly the same story. Low consensus means the entity is fragmented across surfaces and a single point-in-time score is misleading on its own — the gaps between sources are the signal.

Consensus is a meta-signal: it is what makes scores trustworthy. We surface low-consensus states explicitly rather than averaging them away.

The AI Visibility Score aggregates four weighted signal dimensions: narrative presence (how often an entity appears in observed AI outputs), thematic consistency (whether the same themes persist across prompts), sentiment alignment (positive/neutral/negative framing balance), and recommendation presence (direct recommendation or comparison inclusion frequency).

It is computed across ChatGPT, Gemini, Claude, Perplexity, Copilot, Google AI Overviews, and SERP. Scores are normalised to 0–100 and tracked over time so trend matters as much as the level.

Sentiment is derived from observed framing in AI outputs and from NLP analysis of community / review / news content. The five-band scale is the human-readable surface; the underlying value feeds the Sentiment Triangulation Index.

Positive: consistent favourable framing. Neutral: descriptive, non-evaluative outputs. Mixed: balance of positive and critical framing. Critical: predominantly negative framing. Absent: the entity was not meaningfully surfaced.

Recommendation Presence measures how often an entity is directly recommended, included in "best of" lists, or named as an example in AI response and AI Overview contexts. Prominent: regularly foregrounded. Frequent: consistent inclusion in multi-entity contexts. Occasional: appears but not prioritised. Rare: near-absence from recommendation outputs.

Narrative themes are clustered tags derived from recurring language patterns in observed AI outputs and community / news content for an entity. Themes are not user-generated or manually applied — they surface from consistent topic associations.

Each theme represents a meaningful and repeatable association pattern found across multiple prompt variations and source categories. Themes evolve as the underlying signals evolve.

Entidex runs continuous sweeps across live web surfaces — organic search results, AI Overview citations, knowledge panels, Google Places, news indexes, academic indexes, and entity-specific signals — to observe how an entity appears across the full public web index at a given moment. These sweeps produce provenance-rich observations: each result is time-stamped, source-tagged, and routed to the appropriate signal surface.

Drift detection compares observation windows across surfaces and computes z-score deviations for sentiment, source composition, velocity, and AI engine behaviour. When multiple detectors fire simultaneously, an inference pass maps them to a named meta-pattern: Breakout (rapid cross-surface expansion), Erosion (declining presence with negative shift), Authority Lag (AI engines trailing search index changes), Pivot (topic frame rotation), Crisis (sudden sentiment collapse), or Drift to Irrelevance (gradual signal attenuation).

Drift events are surfaced on entity profiles with severity level, pattern name, and a plain-English explanation. Detector internals are available to authenticated researchers via the API but are never shown by default on public pages — reducing noise, not hiding evidence.

Every signal Entidex surfaces is tied back to the collector run that produced it, the source it came from, and the timestamp of capture. The evidence graph is append-only — historical observations are never overwritten — so a score in March can always be explained by the evidence that existed in March.

Freshness is reported per source: when each collector last ran, when the underlying source was last updated, and how stale a given component is. We never present aggregate scores without exposing the freshness behind them.

Discovery observations carry additional provenance fields: the surface category, result rank, and whether the observation was the product of a depth-1 sweep or a recursively expanded query. This metadata is preserved in the evidence graph and available via the v1 API.