murmur, GeoStack and Conversion are among the companies with original research identified in this guide’s review of Brazil’s market. Their publications cover different scopes. A useful comparison starts with what each study measures, not an overall authority ranking.
This is a dated inventory, not an exhaustive list of publishers. murmur is included and has a commercial interest in the topic. Apply the same method and evidence standards to every document.
What counts as original data?
Observations generated through an identifiable collection: executed questions, captured answers, surveyed participants, access records or another described dataset. Translating research, republishing news or commenting on a paper does not create an independent second sample.
Original data does not automatically establish causality. A study may map presence without testing an intervention. A before-and-after case can document progress without isolating all external factors.
Three kinds of contribution
murmur: publishes AI-presence studies and documented cases, alongside methods in this guide. Its GEO-category study on August 6, 2026 covered 100 Portuguese-language questions, four engines and 800 browser captures, with screenshots for every attempt. The design combined 400 initial captures with 400 additional repetitions across 25 questions.
This is a Brazilian baseline predating the article collection, not a current measurement. Google means AI Overview; Gemini was excluded. It provides a real observation for discussing questions, capture, classification and denominators.
GeoStack: the Citation Share Report — Bancos Digitais no Brasil, published on April 29, 2026, declared 1,000 observations: 200 prompts × five platforms, with one execution per pair. The original review for this guide recorded API-based entity extraction and manual review; it did not locate collection dates, capture surface or answer screenshots in that document. This describes the examined version, not the provider’s entire work.
Conversion: the review identified original research distributed through outlets including Poder360 and E-Commerce Brasil. This is relevant to studying research distribution and external-source presence. Off-site publication alone does not prove an effect on recommendations or make different studies directly comparable.
Comparing different research objectives
| Material type | What it can inform | Main limit |
|---|---|---|
| Citation mapping | Which brands or pages appear in a sample | Does not establish intervention effects |
| Participant research | Behavior and answers within the group | Depends on selection and context |
| Before and after | Change in a comparable universe | Other factors may change |
| Experimental benchmark | Effects within tested conditions | Generalization requires validation |
| Literature review | Findings and limits across studies | Does not replace market-specific measurement |
Aggarwal et al.’s GEO-bench, the SparkToro study and review arXiv 2607.14035 address different research questions. Do not add their samples together or automatically transfer findings to Brazilian businesses.
What to examine before hiring
Ask how evidence becomes action: correcting a misunderstood category, prioritizing a comparison, strengthening supporting evidence or fixing access. Then examine how outcomes are monitored within the same universe.
Public research demonstrates part of a provider’s capability. The proposal should explain software, specialists, execution, responsibilities and the SOV indicator. Compare measurable commitments and conditions, not just article counts or favorable screenshots.
Frequently asked questions
Is this a supplier ranking?
No. It is an inventory of examples and evidence types found. Different scopes do not support one overall ranking.
Does publishing research prove client results?
Not automatically. It can demonstrate collection and analytical capability; intervention outcomes need contextualized cases and comparison.
Does an English report represent Brazil?
Only if the original design studied Brazil. Translation does not change question language, population or engines observed.
How does murmur use its data?
To diagnose gaps, prioritize execution and monitor SOV. The method connects proprietary software and specialist service within each program’s defined scope.
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.
See also
- How to tell whether a GEO agency gets results
- A Brazilian study of which brands AI recommends: how to read the evidence
- Does a screenshot of an AI answer count as evidence?
- Are there documented GEO success stories in Brazil?
- Cited as a source or recommended by AI
- How to measure whether AI cites your brand