Generative Engine Optimization is often discussed as a way to improve the odds of appearing in AI-generated answers. A recent editorial by Pedro Dias adds a more specific idea to that discussion: AI discovery may be shaped not only by what a system retrieves in a given moment, but also by the associations and knowledge it has formed around a brand, page or topic.
In “Google Gets The Thank-You”, Dias describes GEO as a memory and association-driven layer alongside conventional SEO. His shorthand is straightforward: AI visibility = SEO + GEO.
That is a useful framing, but it should not be mistaken for a settled ranking formula. The framework is still emerging, and the research cited around it is exploratory. For site owners, its immediate value is practical: it expands the question from “Did an AI answer link to us?” to “Can AI systems accurately retrieve, understand and associate our information with the questions people ask?”
What the GEO framework adds to SEO
Traditional SEO focuses on making pages crawlable, indexable, relevant and useful for search queries. Those fundamentals remain central in AI-powered Google experiences. Google’s guidance for AI Overviews and AI Mode says there are no extra technical requirements or special optimizations to appear in those features. Pages still need to be indexed and eligible to appear in Search, and established SEO best practices still apply.
Dias’s argument is not that GEO replaces this work. Instead, it addresses the stage that can happen before a referral appears in analytics. A person may ask an assistant for a recommendation, comparison or explanation, then receive an answer that frames the available choices. That answer can influence a later visit, branded search or conversion even when the earlier AI interaction is not captured as a conventional site session.
This is where the term “memory” needs care. It does not necessarily mean an AI system permanently stores a particular page or brand in a simple, human-like way. In this context, it describes the potential importance of model knowledge and associations, alongside live retrieval, in shaping what an AI system mentions or cites.
Why citations are only part of the picture
The framework draws on early research such as the October 2026 arXiv paper What Drives AI Citations? Content, Retrieval and Model Knowledge in Generative Engine Optimization. The work uses two longitudinal datasets containing hundreds of thousands of citations in AI-generated answers and separates citation selection from what it calls absorption.
Its findings point to several distinct influences on whether content is cited:
- The content of a page and its alignment with the user’s question can affect citation likelihood.
- Page-level properties explain only limited variation in citations across a site.
- Retrieval behavior observed through API replay can align with an engine’s ranking, while differing from public search results.
- Model knowledge and association signals may affect later queries and mentions.
The key implication is that a high-ranking page is not automatically the page an AI system will cite. Nor is an AI citation proof that the citation caused a click or sale. Retrieval, answer composition, model knowledge and the query itself can all affect the output.
A related October 2026 arXiv study, Location, Location, Generation, examines how the location and structure of questions influence retrieval and generated answers. Together, these studies support the idea that AI visibility is more complex than a single position in a results page. They do not establish universal rules for every platform, language, topic or business category.
A practical GEO workflow for smaller teams
Small businesses and lean marketing teams do not need to create a separate, speculative content program for every AI assistant. A more useful approach is to extend existing SEO operations with a focused visibility and evidence workflow.
1. Start with the questions that affect decisions
Identify questions customers ask before they choose a product, service or provider. These may include comparisons, compatibility questions, local service needs, use cases, costs, limitations and implementation concerns.
Then review whether the site answers those questions directly and consistently. Useful pages should make the core answer clear, support it with relevant detail and connect it to the right product, service or expertise. This is conventional intent-led SEO, but it also gives retrieval systems more coherent material to work with.
2. Make important information accessible and unambiguous
AI retrieval cannot compensate for inaccessible or contradictory source material. Maintain indexable pages, descriptive titles and headings, logical internal links, accurate structured data where it applies, and clear statements about products, services, authorship and policies.
For organizations with data that changes frequently, Dias also points to a possible longer-term shift toward API endpoints and service integrations. The opportunity is not a mandate or a shortcut to visibility. It is the possibility of making useful data available in formats that systems can access and use over time, in addition to publishing it on web pages.
3. Observe AI answers alongside search performance
Use ordinary search analytics to understand impressions, clicks, queries, landing pages and conversions. Add an AI-search observation layer by recording whether relevant AI responses cite, mention or recommend the organization and whether the description is accurate.
Track the query, platform, date, answer format, cited sources and wording of the brand reference. This creates a repeatable record of visibility rather than relying on occasional screenshots. Google’s AI features can also be assessed through the same foundational Search performance reporting used for other Search appearances, although that data will not resolve every AI attribution question.
4. Treat attribution as an investigation, not a verdict
A rise in direct traffic, branded searches or conversions after increased AI exposure may be meaningful, but it does not prove AI caused the result. Campaign activity, seasonality, rankings, media coverage and other channels can produce similar patterns.
The responsible approach is to compare timing and trends, annotate marketing activity, use referral data where available, and preserve the distinction between correlation and causation. This matters particularly for publishers and ecommerce businesses that may be tempted to assign revenue to an AI mention that was never directly measured.
What not to conclude from the research
The early evidence does not show that there is one GEO score, one content template or one durable technique that guarantees citations. It also does not show that every assistant uses the same combination of retrieval and model knowledge.
The cited arXiv studies are early-stage research, while Dias’s article is an editorial synthesis rather than a comprehensive empirical study. Their value is in offering a measurement vocabulary for a real visibility problem, not in providing a universal ROI model.
The most defensible near-term strategy remains to strengthen the information a business publishes, check how AI systems represent it, and avoid overstating what incomplete attribution data can prove.
If AI answers are starting to shape how customers discover your category, GEO Search Leads can help turn ad hoc checks into a clearer assessment of website GEO readiness and potential optimization opportunities. Create a GEO Search Leads account to begin organizing the next steps around the information your site makes available to AI-driven discovery.
Frequently Asked Questions
Does GEO replace SEO?
No. Google says established SEO best practices remain relevant for AI features such as AI Overviews and AI Mode. The GEO framework adds attention to AI retrieval, citations and possible model associations, but it does not remove the need for crawlable, indexed and useful pages.
What are memory signals in GEO?
Memory signals describe the possible role of a model’s stored knowledge and associations in later AI answers, alongside live retrieval. They are not a confirmed standalone ranking factor or a guarantee that a brand will be mentioned.
Does an AI citation prove that AI drove a conversion?
No. A citation shows that a source appeared in an observed AI answer. It does not, by itself, establish that a user clicked, converted or made a decision because of that answer.
What should a business measure first?
Start with decision-relevant questions, the accuracy of AI mentions and citations for those questions, and normal search performance such as impressions, clicks, landing pages and conversions. Keep records of dates, platforms and answer wording so patterns can be evaluated over time.
Conclusion
GEO’s memory and association model is a useful way to think about AI discovery beyond rankings and referral traffic. The evidence is still developing, but the practical direction is clear: maintain strong SEO foundations, publish accessible and intent-aligned information, and measure AI visibility without claiming more certainty than the data supports.
Top comments (0)