What Is an Enhanced Entity Page? The GEO Page Format

Why Page Architecture Matters for AI Retrieval

Most content that fails to appear in AI-generated answers does not fail because it lacks information. It fails because of how that information is organized and presented.

AI retrieval systems do not read a page the way a human reader does, moving linearly from introduction to conclusion and synthesizing meaning along the way. They scan for specific things: named entities, direct claims, defined relationships between concepts, explicit signals of authorship and credibility, and structural patterns that indicate what type of content a section contains. When these things are present and clearly arranged, AI systems can extract and use the content efficiently. When they are embedded in narrative prose or hidden in code that the retrieval pipeline treats as undifferentiated text, even excellent content may not be effectively used.

This is not a new problem. Web content has always been written primarily for human readers and only secondarily for machines. Traditional search optimization addressed part of this gap, helping machines find and classify pages, but did not need to address the deeper question of whether specific claims within a page could be reliably extracted and attributed. That question was less consequential when the output of a search engine was a ranked list of links and the user did the synthesis themselves.

AI answer systems change this. When the output is a generated response that synthesizes content from multiple sources, the question of whether a specific claim can be extracted cleanly from a specific source becomes central to whether that source receives credit. A page that is intelligently written but poorly structured for extraction may contribute to an AI answer without receiving attribution. A page that is explicitly structured for extraction may be cited accurately and linked to. The architecture of the page is part of what determines which outcome occurs. For more on the broader shift in how AI systems process content, how AI is changing search covers the retrieval and synthesis mechanics in more depth.

What an Enhanced Entity Page Is

An Enhanced Entity Page is a page format in which the structural information that helps AI systems understand, navigate, and attribute content is made visible on the page itself rather than hidden in embedded code.

The distinction from a standard page with schema markup is important. Schema markup, typically written as JSON-LD embedded in the page code, provides machine-readable labels for the entities and relationships on a page. When implemented correctly, it is valuable. But it is only visible to systems that specifically parse schema blocks. Most AI retrieval systems in widespread use today process pages as flat text: they receive the page content as a single text stream, embedded code and all. Schema markup embedded in a code block at the bottom of a page may be processed differently or partially in these contexts.

An Enhanced Entity Page takes a different approach. It makes the key structural information visible in the readable content of the page itself. The entity type the page is about is stated as readable text near the top. The relationships between this entity and connected concepts are shown as labeled, navigable links within the page. A summary that can stand alone without the rest of the page is placed near the top so AI systems can extract it cleanly. Instructions for AI agents about how to navigate the site are written as readable text rather than relying on programmatic signals alone.

In this format, the page is doing two things simultaneously: serving human readers with clear, useful information, and serving AI retrieval systems with the structural signals they need to extract, attribute, and navigate accurately. These two goals are compatible. A page that is well-organized for human readers tends to be better structured for AI extraction than one that is poorly organized. The Enhanced Entity Page format makes this dual purpose deliberate and explicit rather than accidental.

This format is connected to but distinct from the broader practice of entity-based SEO. Entity-based SEO is the practice of defining, naming, and connecting entities consistently across a site and external presence. An Enhanced Entity Page is the specific page-level architecture that implements those principles in a format optimized for AI retrieval systems.

The Elements of an Enhanced Entity Page

The format has seven elements that work together. Not every page requires all seven, and the specific implementation of each element should reflect the nature of the page and the entities it covers. But these are the components that define the format.

A self-contained executive summary

The page opens with a paragraph or short block that describes the page's subject precisely enough to be used as a standalone extract. It answers the question the page is about without requiring the reader to continue. This is different from an introduction that sets context for what follows. It is a complete, standalone description of the entity or concept, placed where AI systems encounter it first.

The principle behind this element is simple: AI retrieval systems do not always process a full page. They may extract from the sections most semantically aligned with the query. A self-contained summary near the top ensures that even a partial extraction produces an accurate, attributable result.

A visible type classification

A short line near the top of the page shows what category of thing the page is about, expressed as readable text. Not a navigation breadcrumb, which indicates where the page lives in a site hierarchy. A type classification that indicates what the subject of the page is in a broader conceptual structure.

For a service page about GEO audits, this might read: "Type: Professional Service, GEO Service, AI Visibility Audit." For a concept page about answer units, it might read: "Type: Concept, GEO, Content Structure, Answer Unit." The purpose is to give AI systems an explicit category signal that helps them classify the page correctly, independent of whether they successfully parse the schema markup.

This element should be styled distinctly from the navigation breadcrumb so that readers understand it is a category label, not a navigation path.

A Related Entities section with labeled relationships

A section that lists the concepts, services, people, and pages most directly connected to this page's entity, with the relationship type explicitly labeled. Not a "Related Articles" widget based on recency or tags. A structured set of named links where each relationship is identified.

For a GEO service page, this might include: "Uses methodology: Enhanced Entity Pages," "Preceded by: GEO Audit," "Measures: Answer Presence, Attribution Quality, Faithfulness, Hand-off Success." Each link is navigable and each relationship label tells both the human reader and the AI system what the connection is.

This element transforms a page from an isolated document into a node in a linked entity network. AI systems that can follow these relationships can build a more complete understanding of the subject and its context than they could from a single page alone.

An Agent Instructions block

A visible section, typically near the top of the page and clearly labeled, that provides explicit guidance for AI systems navigating the page and site. This is written as readable text, not hidden in code. It tells AI agents what the page contains, where to find related information, how to navigate to adjacent content, and how the brand prefers to be cited.

This element is analogous in concept to the llms.txt proposal, which suggests a standardized file at the site root that gives AI systems navigational guidance for the whole site. The Agent Instructions block does this at the page level, making the guidance visible in the page content rather than in a separate file. Treating it as visible page content rather than hidden metadata means AI systems that process pages as flat text can still use it.

Answer units throughout the body content

The body content is organized as answer units: self-contained blocks where a direct claim is followed immediately by supporting context and evidence, and closed with a clear takeaway. This is not the same as well-organized prose, though good answer units read naturally. It is a specific structural approach to content that prioritizes extractability: each unit can be lifted from the page and placed in an AI answer without losing coherence or accuracy.

The relationship between Enhanced Entity Pages and answer units is one of container and content. The Enhanced Entity Page provides the page-level architecture. Answer units provide the content-level structure within that architecture.

Schema markup that mirrors visible content

JSON-LD schema is still present and valuable, but it reflects and reinforces what is already visible on the page rather than introducing entity information that does not appear in the readable content. The author named in the schema is the same author visible in the author byline. The entity described in the "about" field is the same entity defined in the executive summary. The publication date in the schema matches the visible date on the page.

This alignment matters because the alternative, schema that claims properties not present in the visible content, is a reliability risk. If the schema says the author is Dr. Smith but no author appears on the page, AI systems cross-referencing the schema against the visible content encounter a discrepancy. Aligned schema reinforces entity certainty rather than introducing it separately.

Trust signals: authorship, dates, and credibility indicators

A named author with specific credentials visible near the content. A publication or last-updated date prominently placed. Citations or source references adjacent to the claims they support. These are elements that appear in good editorial content generally, but in an Enhanced Entity Page they are treated as structural requirements rather than stylistic choices.

For AI systems evaluating whether a source is credible enough to attribute, these signals are explicit evidence rather than inferred from domain authority. A page that shows a named, credentialed author makes it possible for AI to identify and attribute the source at the person level, not just the domain level.

The Difference Between a Standard Page and an Enhanced Entity Page

When an AI retrieval system processes the standard page, it receives a text stream. It identifies that this is a service page for a company. It may extract some information about what the service does. It may include the page in a generated answer. But because the entity relationships are not explicitly visible, the type classification is not stated, and the body content is organized for narrative flow rather than extraction, the quality of what it extracts is contingent on the AI's ability to infer structure from prose. The result may be a generic, partially accurate summary with weak attribution.

When an AI retrieval system processes the Enhanced Entity Page version of the same content, it encounters a self-contained summary first that defines the service precisely. It sees the type classification: this is a Professional Service, GEO Service, AI Visibility Audit. It finds a Related Entities section that tells it this service is preceded by the GEO Strategy engagement and measured through Answer Presence and Attribution Quality metrics. It encounters answer units where each major claim is followed immediately by context and evidence. It reads the Agent Instructions block that tells it how to navigate to related services and where to find case studies. The schema confirms all of this, mirroring what the readable content already established.

The extraction from the Enhanced Entity Page is more accurate, more attributable, and more navigable. The content is not different. The architecture is.

The Research Behind the Format

The Enhanced Entity Page format is grounded in findings from a research preprint published on arXiv in March 2026 by Volpini, Raad, Gamba, and Riccitelli. The study is titled "Structured Linked Data as a Memory Layer for Agent-Orchestrated Retrieval."

Because this work is a preprint and not yet verified through a formal peer-review process, the findings should be treated as emerging research rather than established consensus. The experiment's setup and results are described here accurately, with those qualifications in place.

What the study tested

The researchers tested seven different conditions for content representation across four industry domains: editorial, legal, travel, and e-commerce. Three document formats were compared: plain HTML, HTML with embedded JSON-LD schema, and an enhanced entity page format. These were tested in two retrieval conditions: standard RAG and agentic RAG with link traversal. A total of 2,439 individual answer evaluations were conducted using Vertex AI Vector Search and the Google Agent Development Kit.

What it found

Adding JSON-LD schema to standard HTML produced only a small improvement in retrieval accuracy: approximately 0.17 points on a 5-point scale. The effect was statistically significant but very small in practical terms.

The enhanced entity page format, which incorporated natural language summaries, visible linked entity navigation, agent instructions in llms.txt style, and schema breadcrumbs, produced an accuracy improvement of approximately 1.04 points over plain HTML. That is roughly a 29 percent improvement in retrieval accuracy as measured in this experimental setting.

The researchers also found that in agentic retrieval conditions, enhanced entity pages allowed AI agents to follow fewer links while maintaining the same accuracy, suggesting that the visible structure reduced the interpretive work required from the retrieval system.

What this result does and does not mean

The result is specific to the experimental conditions used: particular retrieval infrastructure, particular evaluation prompts, particular industry domains. It measures retrieval accuracy in a controlled setting, which is informative but not identical to real-world performance across all AI systems. The finding that visible structured information improves retrieval accuracy more than hidden schema alone is consistent with the structural logic of how most AI retrieval pipelines process content, but the specific magnitude of improvement will vary across contexts.

The finding should not be read as "your pages will improve by 29 percent if you use this format." It should be read as: in a carefully constructed experimental setting, making structured information visible on the page produced substantially better retrieval outcomes than embedding the same information in schema code. This is directionally consistent with Growthino's implementation rationale for the format.

Why Growthino uses this format

The experimental results from the Volpini et al. preprint are one input into Growthino's implementation methodology. The other input is the structural logic that the results confirm: AI retrieval systems that process content as flat text benefit from visible structure in the same way human readers do, and a page designed to serve both audiences simultaneously performs better in AI answer environments than a page designed for only one. The format is a practical implementation of this principle, not a proprietary system.

How Enhanced Entity Pages Connect to Other GEO Practices

The Enhanced Entity Page is not a standalone improvement. It is a page-level architecture that integrates several other GEO practices and makes their effects cumulative rather than isolated.

Answer units are the content layer within this architecture. Every body section of an Enhanced Entity Page should be built as an answer unit: a self-contained claim with context, evidence, and a conclusion. The Enhanced Entity Page provides the structural container. Answer units provide the extractable content inside it. Neither works as well without the other: a page of well-written answer units without the entity architecture is harder to attribute; a page with full entity architecture but poorly structured body content still produces weak extraction.

Entity-based SEO is the foundation this format builds on. The visible type classification, related entities section, and schema elements of an Enhanced Entity Page only produce reliable results if the entities they reference are defined consistently across the site and in external profiles. If a company is named differently on the page than on its Clutch profile, the entity relationship signals on the Enhanced Entity Page are in conflict with the external validation layer. The format amplifies entity clarity when it exists; it does not substitute for it when it does not.

Schema implementation is still required but plays a supporting role. JSON-LD schema markup remains valuable for search engines that extract it separately and for AI systems that are designed to parse it. Within the Enhanced Entity Page format, schema reinforces rather than replaces the visible structured information. The schema author field should match the visible author. The schema "about" field should match the visible type classification. The relationship between schema and visible content should be one of confirmation, not contradiction.

External validation is what AI systems use to confirm the claims. When an AI system retrieves an Enhanced Entity Page and finds a precise entity definition, a clear type classification, and labeled entity relationships, it still cross-references these claims against external sources to assess credibility. Strong external profiles that corroborate the on-page entity definitions increase the confidence with which AI systems attribute content from this page.

The format directly affects Attribution Quality and Faithfulness. Attribution Quality improves because the explicit entity identification and agent instructions give AI systems the confidence to name and link your brand rather than using generic attribution. Faithfulness improves because the self-contained summary, the answer unit structure, and the schema alignment reduce the interpretive work required from the AI system, which reduces the opportunities for misrepresentation. These are two of the four GEO metrics most directly affected by page architecture.

AI agent readiness is the frontier application of this format. AI agents that can follow links, traverse pages, and build answers from multiple sources benefit most from Enhanced Entity Pages. The labeled related entities section provides explicit traversal paths. The dereferenceable linked entity concept means that each important entity on the page has its own stable URL that an agent can visit to get more detailed information. The agent instructions block tells the agent where to go next. This level of navigability is what the researchers in the Volpini et al. preprint called "agent-orchestrated retrieval" and what they found produced the best outcomes in their experimental conditions.

The broader GEO framework situates all of this: Enhanced Entity Pages are the page-level implementation of the principles that GEO as a discipline describes.

Which Pages to Build in This Format First

Not every page on a site needs to be rebuilt as an Enhanced Entity Page simultaneously. A sequenced approach produces better results and uses resources more efficiently.

The hub page or primary educational page should come first. This is the page most likely to be the subject of AI retrieval for your core category queries. For Growthino, this is /what-is-geo. For a talent-matching startup, this might be a page like “How to hire remote developers.” For a SaaS product, it might be the main feature or use case page. This page handles the highest volume of relevant queries and has the greatest impact on Answer Presence when it is well-structured.

Service or product pages that represent commercial intent queries come second. These are the pages users would encounter when asking AI systems for recommendations or comparisons. When a user asks "What is the best platform to hire remote developers?” or “How can I find vetted tech talent quickly?”, they are asking questions that should return service pages with clear entity architecture.

The about page and founder page deserve early attention because they affect Attribution Quality directly. When AI systems attribute content to a brand, they draw on person-level entity information as well as organizational information. An Enhanced Entity Page about page with a named, credentialed founder, a clear organizational description, and labeled relationships to the startup's services and methodology gives AI systems the person-level entity information needed for strong named attribution.

Concept and methodology pages are the third tier. These are pages that define the key ideas the startup wants to own. For a talent-matching startup, this could include pages on vetted hiring, remote team building, skills-based matching, and startup recruitment workflows. These pages build topical authority and contribute to the entity network that makes commercial page citations more credible.

Blog posts and individual pieces of content are the last priority for this format, because they produce lower attribution impact per page than the structural and commercial pages above. However, the highest-performing blog posts, those that rank well or generate significant traffic, are worth restructuring once the core pages are done.

The practical first step is to identify the five pages most likely to be cited by AI systems for your most important queries. These are the Enhanced Entity Page builds that will produce the greatest improvement in measurable GEO performance.

Common Mistakes

Treating JSON-LD as the full implementation. Adding a schema block to an existing page and calling it an Enhanced Entity Page is the most common misapplication of this format. Schema is one element. It is valuable. But a page with good schema and a wall of undifferentiated prose does not produce the extraction quality that the format is designed for. The visible structure is what does most of the work for AI systems that process pages as flat text.

Writing the type classification as a navigation element. The visible type classification that identifies what category of thing the page covers should be distinct from the site navigation breadcrumb. If both elements look the same and appear in the same location, users are confused about which one is for navigation and which is a content classification. The type line should be visually different from the navigation breadcrumb and clearly labeled or styled as a category indicator.

Adding labeled relationships without real entity definitions. A Related Entities section with labeled links is only valuable if the entities those links point to are actually defined. Linking to "GEO" with the label "foundational concept" is useful when /what-is-geo has a clear, precise definition of GEO. It is not useful when the destination page is a thin content page with a vague description. The entity network only functions as a credibility amplifier if the entities in it are well-defined.

Exposing framework labels in the body content. Marking paragraphs with visible labels like CLAIM, CONTEXT, EVIDENCE, or TAKEAWAY is a presentation of the writing framework, not a content element. These labels are tools for the writer. They are not design features for the reader. When these labels appear on a published page, they interrupt reading, signal that the page was built to a template rather than written with intent, and do not improve AI extraction quality. The structure is in the writing. The labels are not needed.

Creating the format without author and trust signals. An Enhanced Entity Page that has all the structural elements but no visible named author, no publication date, and no credentials or source citations is missing the trust layer that makes the entity claims credible. The format is designed to be credible to both human readers and AI systems simultaneously. Credibility for AI systems requires explicit trust signals. A page without them has the appearance of the format without the substance.

Building the format before fixing entity definitions. An Enhanced Entity Page built on inconsistent entity foundations produces inconsistent results. If the company is named differently in the type classification than it is on external profiles, the related entities section links to a company page with a different description than the one in the summary, and the schema author does not match the visible byline, the format is creating more entity confusion rather than less. The entity-based SEO foundations need to be in place, or at least consistent on the specific page being built, before the Enhanced Entity Page format produces its intended effect.

Frequently Asked Questions