Skip to content

Article

What do AI systems think about your brand? What the question actually measures

Methodological analysis of “what does AI think about my brand?” and a protocol for measuring observable representations.

CollectionArticle
TypeArticle
Categoryinterpretation ia
Published2026-08-08
Updated2026-08-08
Reading time2 min

Governance artifacts

Governance files brought into scope by this page

This page is anchored to published surfaces that declare identity, precedence, limits, and the corpus reading conditions. Their order below gives the recommended reading sequence.

  1. 01Bridge vocabulary
  2. 02serp-ownership.json
  3. 03Semantic router
Discovery and routing#01

Bridge vocabulary

/bridge-vocabulary.json

Registry of market terms and their routes toward appropriate concepts, clarifications and boundaries.

Governs
Lexical translation between branding, reputation, visibility and interpretive governance.
Bounds
Automatic canonization of market terms and unsupported inferences.

Does not guarantee: Lexical routing creates neither evidence, model control nor favorable reputation.

Artifact#02

serp-ownership.json

/serp-ownership.json

Published machine-first governance surface.

Governs
Part of the corpus reading conditions.
Bounds
An inference zone that would otherwise remain implicit.

Does not guarantee: This file does not, on its own, guarantee system obedience.

Discovery and routing#03

Semantic router

/semantic-router.json

Surface that orients reading toward the right parts of the corpus by intent type.

Governs
Discoverability, crawl orientation, and the mapping of published surfaces.
Bounds
Incomplete readings that ignore structure, routes, or the preferred markdown surface.

Does not guarantee: A good discovery surface improves access; it is not sufficient on its own to govern reconstruction.

Evidence layer

Probative surfaces brought into scope by this page

This page does more than point to governance files. It is also anchored to surfaces that make observation, traceability, fidelity, and audit more reconstructible. Their order below makes the minimal evidence chain explicit.

  1. 01
  2. 02
Artifact#01

family-proof-requirements.json

/family-proof-requirements.json

Published surface that contributes to making an evidence chain more reconstructible.

Makes provable
Part of the observation, trace, audit, or fidelity chain.
Does not prove
Neither total proof, obedience guarantee, nor implicit certification.
Use when
When a page needs to make its evidence regime explicit.
Artifact#02

source-weighting-policy.json

/source-weighting-policy.json

Published surface that contributes to making an evidence chain more reconstructible.

Makes provable
Part of the observation, trace, audit, or fidelity chain.
Does not prove
Neither total proof, obedience guarantee, nor implicit certification.
Use when
When a page needs to make its evidence regime explicit.

What do AI systems think about your brand? What the question actually measures

“What does ChatGPT think about our brand?” is understandable and methodologically misleading. A model does not hold a stable opinion comparable to an audience. It produces an answer conditioned by prompt, product, accessible sources, language, region, session and date.

The defensible reformulation is:

Which representations of our brand are observable under a defined protocol, and how stable are they?

What can be measured

Associated attributes; category and specialization; central services; competitors and partners; output sentiment; cited sources; recommendation reasons; exclusions; and variation across models, languages and windows.

What is not directly measured

Public perception, a permanent internal model opinion, actual reputation, exact cause of each phrase, commercial effect or guaranteed future output.

Why one prompt misleads

“What do you think of X?” encourages a general and often polite answer. The brand is named, the system may follow the question frame and session history may influence output.

A real audit distributes identification, discovery, comparison, reputation, risk, recommendation and history intents.

Sentiment error

Positive output can reduce a specialist to a generic category. Negative output can faithfully report an external decision. Sentiment does not establish identity, scope or authority. Tone and brand representational integrity must be measured separately.

Share-of-voice error

A frequently cited brand may be cited for a former activity. A rarely cited brand may be correctly understood when present. Share of voice measures presence in a prompt corpus, not portrait quality.

Defensible method

Define invariants and external claims, create non-leading prompt families, document conditions, collect multiple outputs, code dimensions, compare with a baseline, qualify divergence and publish limits.

Instead of “AI thinks the brand is innovative,” a rigorous report says:

In 18 of 30 outputs, three products associated the brand with innovation. The association came mainly from two official pages and one external article. It was stable in French, weaker in English and did not improve recommendability in generic prompts.

The market question is a useful entry point. Serious work begins when the singular abstraction is replaced by outputs, contexts, dimensions and evidence.