AI perception stability matrix
This matrix qualifies representational stability without requiring identical answers. It tracks material brand invariants across models, languages, regions, intents and time.
Prerequisite: a dated baseline
The baseline separates identity; category and specialization; scope, audiences and exclusions; provable differentiators; relations; temporal state; and external claims that are not governed by the brand canon.
Axes
| Axis | Stable | Fragile | Drifted |
|---|---|---|---|
| Identity | Correct entity | Depends on exact name | Fusion or misattribution |
| Category | Role and specialization preserved | Broad but compatible | Wrong market or material generalization |
| Scope | Offer and limits respected | Minor omission | Invented, old or out-of-audience offer |
| Differentiation | Invariants recognizable | Partial | Interchangeable brand |
| Sources | Competent authority by claim | Incomplete attribution | Wrong source class |
| Time | Current state and dates | Ambiguous | Old state shown as current |
| Relations | Correct affiliations and competitors | Fluctuating neighbourhood | Wrong entity or relation |
| Recommendability | Aligned use cases | Weak reasons | Wrong use or unjustified exclusion |
| Cross-model | Compatible variation | One system diverges | Persistent incompatible versions |
| Cross-language | Invariant parity | Local gaps | Different identity or category |
Levels
0, insufficient presence; 1, fragile presence; 2, partial representation; 3, conditional fidelity; 4, cross-system stability; 5, governable stability with versioned observation and classified gaps.
Sampling
Test definition, comparison, recommendation, problem, alternatives, verification and risk. Separate brand-named from generic prompts. Cover material languages and regions. Preserve observation count and confidence for every cell.
Reading results
High stability with low fidelity means a repeated error. Low stability with medium fidelity means variability. High visibility with category drift means presence success and positioning failure. Positive sentiment with false scope is flattering but unsafe. Visible official sources with structuring third parties indicate exogenous governance.
Limits
The matrix does not predict future outputs or prove correction effects. It does not penalize diverse wording when invariants remain. It does not make official canon authoritative over external reputation.
It determines where to investigate, which drift to classify and which conditions to reproduce during re-observation.