Why AI recommends some brands and not others

murmur’s own results cited in this guide come from campaigns in Brazil using Portuguese questions. This English edition translates the analysis and question wording; measurement figures, dates and limitations remain unchanged.

A brand needs to be identified correctly and fit the buyer’s need to enter an AI recommendation. Accessible content, coverage of the question and verifiable evidence help establish that context. Answers also depend on the engine, wording and session; one result cannot reveal the cause of an absence.

murmur organizes diagnosis around entity, retrieval and answer presentation. This is our operational approach to choosing interventions, not an exact universal description of every engine’s architecture.

1. Does the AI understand which company it is?

A brand may be confused with a namesake or placed in the wrong category. Check whether the answer recognizes the offer, audience, relevant location and actual differentiators.

Descriptions should be clear across the website, official profiles and public sources identifying the company. Structured data should reflect visible content. Identity links must point to real profiles; a profile list cannot compensate for contradictory descriptions.

Asking by name tests recognition. To assess discovery, also use questions that omit the brand. Keep these objectives separate in reporting.

2. Is there useful content for that need?

A published page may still be undiscovered, unindexed or unselected for an answer. Check access, server-delivered content, internal links, sitemaps and indexing signals. Vercel’s AI crawler study documents why rendering deserves attention.

Then examine intent. A question about reducing missed appointments may need diagnosis and solution-selection criteria. A supplier comparison needs verifiable differences. Institutional descriptions alone leave buying questions unanswered.

Group equivalent intentions and write pages that add information: criteria, examples, implementation, boundaries, evidence and next steps.

3. Does the information support a choice?

An engine can use your website to explain a topic without suggesting your company. That is useful presence, but different from recommendation.

For Fly Vet, across 100 Portuguese-language questions measured in June 2026 in Brazil, classification recorded recommendations in 30/100 on ChatGPT and 41/100 on Claude. Google AI Overview recorded 81/100 as a source and zero recommendations. This is a historical snapshot for that question set, not a permanent brand condition.

The practical question is whether the content presents the company as a suitable solution with evidence and criteria, or merely explains the topic. Improving that connection is more specific than indiscriminately increasing article volume.

Compare engines without inventing a cause

In murmur’s August 6, 2026 Brazilian campaign, one Portuguese question repeated five times per engine produced a listed brand in 5/5 on ChatGPT, 4/5 on Perplexity, 0/5 on Claude and 0/5 on AI Overview: 20 captures. This demonstrates differences in observed output, not which ranking factor caused them.

The campaign preceded this article collection and illustrates measurement design here. Google means AI Overview; Gemini was not measured.

Turning diagnosis into a plan

Observed signal	Next investigation	Possible intervention
Namesake or wrong category	Descriptions and identity sources	Correct presentation and profiles
Absence for an intent	Access, competing pages and coverage	Fix access or create specific content
Source without supplier endorsement	Offer, criteria and available evidence	Explain fit and document outcomes
Variation across rounds	Comparability and repetition	Observe further before attributing effects

No row is an automatic diagnosis. A missing visible link does not reveal every consulted document. Combine signals and remeasure the same set after changes.

Frequently asked questions

Why does ChatGPT cite me when another engine does not?

Products, sources, sessions and criteria can yield different answers. Compare equivalent questions and repetitions before choosing an intervention.

Does resolving identity guarantee discovery?

No. It improves identity clarity, but content and evidence must still answer the need. Named recognition and spontaneous recommendation are different tests.

Do more mentions mean more recommendations?

Not necessarily. Classify source, mention, recommendation and absence separately, and track the SOV matching the project objective.

How murmur works

murmur works through a SWAS model, combining proprietary measurement software with specialists who diagnose, plan and execute GEO to grow Share of Voice (SOV). Eligible contract models may include an SOV-growth guarantee, with the metric, baseline, period, scope and conditions agreed in writing. This does not guarantee an individual answer, a fixed position or sales.

Author: Mateus Gomes, murmur operator. This article explains our approach and discloses our commercial interest.