Brand Representation in AI Answers: An Open Specification for Measuring Visibility, Alignment and Sentiment

Summary

When someone asks ChatGPT, Perplexity, or any generative AI answer engine which brands to consider, the brands that appear in the response gain an immediate credibility advantage. Those that do not appear are, for that query, invisible. For marketing teams and brand managers, this creates an obvious measurement problem. How do you track brand visibility in AI-generated responses across thousands of potential prompts, engines, and time periods, when no open, auditable specification exists? The Generative Brand Mention Framework (GBMF) addresses that gap with three separable measures. MV (brand-mention visibility) tracks how often the brand appears in AI responses. When the brand does appear, MA (brand-mention alignment) measures how accurately AI describes it against the brand's own declared facts, and MS (brand-mention sentiment) measures whether AI speaks about it positively or negatively. Together they give commercial teams an open, reproducible basis for monitoring and improving how their brand is represented in AI-generated answers, and for comparing that representation against competitors.

This page summarises the paper — the full document is the PDF

Formal definitions, evaluator architecture, falsifiability criteria, appendices, and references are in the PDF only.

Download the full paper (PDF) →

Abstract

This paper specifies the Generative Brand Mention Framework (GBMF), a proposed open measurement specification for brand representation in AI-generated responses. GBMF comprises three separable measures. Brand-mention visibility (MV) estimates the probability that the brand appears in an AI response to a category-relevant prompt. Brand-mention alignment (MA) measures, given a mention, the consistency of the description against the brand’s own declared facts, not external truth. The “alignment” construct here concerns factual correspondence, distinct from the AI-alignment construct that concerns model behaviour. Brand-mention sentiment (MS) measures, given a mention, whether the response speaks positively or negatively about the brand. MA and MS each condition on mention.

The three measures are necessary because the underlying mechanisms are separable. A brand can be mentioned yet described inaccurately, or described accurately yet positioned unfavourably, and these are different remediation problems.

MV and MA are reported on a 0–100 scale; MS as three percentage shares plus a net figure on −100 to +100, over a governed prompt set and a declared engine set.

GBMF specifies formal definitions, error quantification, companion statistics, evaluator architecture with a determinism-tested control arm, governance requirements, and pre-stated falsifiability criteria. An end-to-end worked example accompanies the specification, with per-prompt per-engine mention rates published as supplementary data. The framework is published in full to enable open, governed, falsifiable measurement.

Paper details

Author: Khan, B. (2026) Affiliation: Aiviara Research Keywords: Generative Brand Mention Framework, GBMF, MV, MA, MS, AI visibility, AI search, answer engine optimisation, AEO, GEO, brand monitoring, brand visibility, brand sentiment, generative AI, large language models, measurement methodology Version: Working paper

Validation status. This paper specifies MV, MA, and MS as proposed open measurement instruments. It does not report empirical validation results. Prospective validation targets are stated in the paper. Calibration conventions are stated as provisional, pending empirical calibration against observed variance and score dispersion.

The three measures

MV (brand-mention visibility) is a detection measure. It answers how consistently AI systems mention a brand across a standardised set of prompts. A declared prompt set of 30 to 50 queries covers the range of ways buyers approach a category: categorical, comparative, use-case, and problem-solution queries. Prompts are run across a declared engine set, with each prompt-engine pair queried multiple times to account for the natural variation in AI responses. Per-engine MV is the mean mention rate across the full prompt set. Headline MV is the equal-weighted mean of those per-engine scores, on a 0 to 100 scale. An MV of 52 means the brand appears in an estimated 52% of AI responses to category-relevant prompts, averaged across the declared engines.

MA (brand-mention alignment) is a conditional measure. It answers how accurately AI describes the brand when it does mention it. MA is measured against the Brand Profile, a structured, version-controlled document the brand provides, specifying what is factually accurate: product category, pricing entry point, key features, use cases, named integrations, and disambiguation statements. MA is computed only for responses where the brand is mentioned and where the response contains evaluable descriptive content. An MA of 67 means AI descriptions align with the brand’s declared facts 67% of the time on average.

MS (brand-mention sentiment) is the second conditional measure. It answers whether AI speaks positively or negatively about the brand when it does mention it. MS measures sentiment directed toward the brand, not the overall tone of the response. It is reported as three percentage shares (positive, neutral, negative) plus a net figure. The net is positive share minus negative share on a -100 to +100 scale. The shares-plus-net structure distinguishes uniform neutral coverage from polarised coverage in which positive and negative characterisations both occur substantially: both may produce a net near zero, but they represent different brand situations.

The Visibility-Alignment map

The featured diagnostic is the Visibility-Alignment map with MS encoded as point colour. Every brand in a category is scored from the same collected responses, so the map is a competitive comparison by default.

The four quadrants are: Visible & Aligned (high MV, high MA), Visible & Misaligned (high MV, low MA), Aligned but Unseen (low MV, high MA), and Unseen & Misaligned (low MV, low MA). Each quadrant combines with a sentiment band to produce a named brand-representation state. The reception root (AI Champion, AI Contender, AI Wildcard, AI Pariah) describes how the brand is characterised; a position prefix (none, Undiscovered, Misrepresented, or Misrepresented and Undiscovered) describes how the brand’s standing departs from the visible-and-aligned ideal. A brand with MV = 0 produces no mentioning cells and is reported as AI Absent — it carries no reception root, no position prefix, and sits outside the map.

For the worked example, Acme Analytics scores MV 52, MA 67, and sentiment shares of 51% positive / 23% neutral / 26% negative (net +25, Mixed band). The descriptor is AI Wildcard: visible and accurately described, but received with divided opinion. The negative characterisations concentrate on pricing, driven by AI responses stating the wrong price. Because the negative sentiment tracks the pricing misalignment in MA, fixing the cited sources that carry the incorrect price is the likely path to improvement on both measures simultaneously.

How to cite

Khan, B. (2026). Brand Representation in AI Answers: An Open Specification for Measuring Visibility, Alignment and Sentiment. Aiviara Research. https://aiviara.com/research/generative-brand-mention-framework/