GEO (Generative Engine Optimization) is the work of growing a brand’s presence in AI-generated answers. It includes making information accessible, establishing who the company is, answering buyers’ questions and providing evidence that supports a choice. Outcomes can include a cited page, an accurate description or a brand recommendation.
The term gained an academic reference through Generative Engine Optimization by Aggarwal et al., presented at KDD 2024. In commercial practice, the objective is to turn that knowledge into interventions and track their contribution to the brand’s presence in relevant answers.
How it relates to SEO
SEO and GEO share foundations: accessible pages, useful content, site organization and consistent identity. Their measurement also differs. Traditional search uses impressions, rankings and clicks. Generative answers require examining how the brand appears and which sources support the response.
An AI system can rank suppliers in a list; it can also cite a page without recommending its publisher. List position, citation and recommendation are therefore different indicators. An answer can influence a choice without generating a website visit, but presence alone does not prove a sale.
The work involved in GEO
| Area | Practical work | Supporting evidence |
|---|---|---|
| Access and discovery | Deliver readable content, fix blocking issues and connect important pages | HTML, indexing and access records |
| Entity | Describe the category, offer, location and differentiators consistently | Correct identification of the company |
| Content | Answer questions and comparisons with criteria and examples | Retrieval and citation of pages |
| Evidence | Publish verifiable information and earn legitimate third-party recognition | Sources used and claims reproduced |
| Monitoring | Repeat relevant questions and record interventions | Comparable changes in presence and SOV |
These areas interact rather than forming a rigid sequence. If an engine confuses the company with a namesake, its identity needs attention. If it recognizes the brand but excludes it from a specific buying need, examine content and evidence for that situation.
Content that helps a buyer decide
Start with the question customers ask when making a decision. “Which system suits a retail chain?” needs different criteria from “what is management software?”. A useful page answers directly, explains which option suits which situation, provides verifiable facts and connects the problem to the offer without automatically declaring its own brand the winner.
Vercel’s study of AI crawlers also supports checking the content served by the server. It documents rendering limitations during the observed period. The practical task is to verify what each mechanism can read rather than assuming they all work alike.
Measuring progress
Define the market, language, questions, engines and repetitions before comparing results. Separate questions containing the brand name from those testing unprompted discovery. Apply consistent classifications for recommendation, mention, source and absence.
In murmur’s August 6, 2026 Brazilian study, 800 browser captures covered 100 Portuguese-language questions across four engines, with additional repetitions for 25 questions. That campaign preceded this article collection and records a historical baseline, not the operation’s current performance. Google in that dataset means AI Overview; the Gemini app was not measured.
A project needs an up-to-date, comparable baseline. One capture documents an occurrence; tracking SOV requires a defined universe, repeated rounds and analysis by buying intent.
What research supports
The founding paper introduced GEO-bench with 10,000 queries and evaluated content interventions. Benchmark findings help formulate hypotheses but do not provide a universal growth percentage. The literature review arXiv 2607.14035 discusses limits to generalization. Operationally, interventions should be tested and monitored in the market being served.
Frequently asked questions
Does GEO replace SEO?
No. Content discovery and access remain important. GEO adds analysis of generated answers, brand identity and participation in recommendations.
Does being cited mean being recommended?
No. A page may explain a topic without its publisher being suggested as a supplier. Measurement should preserve that distinction.
What should a GEO engagement include?
A scope connecting diagnosis, strategy, execution and monitoring. Ask which SOV metric will be used, which actions are included and how progress will be compared with the baseline.
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.