Future-Proof Data: Lessons from Information Science for Your SEO

Esse artigo pode ser lido em:

I have been working with digital content creation and planning and search engine optimization for quite some time. But after this experience in the academic halls of Library Science, I began to perceive more clearly something that bothered me but I did not know: the SEO market frequently tries to reinvent the wheel.

We become fascinated by novelties such as RAGs, language models, and generative tools, and many specialists forget that the organization, treatment, and information retrieval are disciplines with decades of accumulated scientific rigor, and that they are available to be learned and applied in your work. And I make my mea culpa—even having had this information for decades, only recently did I resolve to take courage and pursue a new degree to study this subject.

And during these studies, I read a report that addresses the development and curation of corpora between 2021 and 2026, which was created using the Semantic+Agent. My objective with this research was to find and analyze what we have most recently in research on the massive use of texts for creating a “database,” the corpora, with the aim of feeding LLMs and generating term lists, controlled vocabularies, taxonomies, glossaries, and other artifacts. My intention is to further qualify the initial step of the Semantic Workflow, the SWF.

Reading the text brought me interesting reflections from Information Science on data governance, preservation, and algorithmic ethics—something I did not expect to see, at first, in this type of research. I compiled and expanded these reflections for our practical context here at the Semantic Blog, connecting Library Science theory with the day-to-day of technical optimization, and I chose to bring direct examples from e-commerce, as they are something that everyone, specialists or not, has experience dealing with.

If you work with SEO and want to understand why the deep structure of your data determines whether your project will be found by your users wherever they search, I invite you to reflect with me on the lessons I learned in this research.

The Myth of the Eternal Document and the Illusion of Digital Permanence

We tend to believe in the idea that the digital environment is synonymous with eternity. We think that by uploading a database to the cloud or publishing thousands of pages in an online catalog, that information will be preserved forever. After all, the data are published and there is already a mature process to keep this online. A grave mistake.

For those who study Information Science and Archival Science, this illusion quickly dissolves. Think, for example, of a legal or institutional document validated by a digital cryptographic signature: the keys of that signature frequently expire long before the end of the legal retention and custody periods for that record. If the technology that authenticates the file dies, how do we prove that that datum is intact?

This technical fragility shows us that the concept of a static, untouchable “original document” is outdated. In the digital universe, authenticity is not a fixed artifact locked in a vault, but a dynamic quality that must be actively managed and cultivated through rich metadata. This is so important and current that there are professionals dedicated to creating management standards for these digital collections, and an entire group of specialists applying this in databases of all kinds.

In the environment of a large e-commerce operation, we experience this same volatility every day. Think of an online store with more than 50,000 products: descriptions change, SKUs (Stock Keeping Units) are discontinued, URLs are redirected, and technical product attributes are overwritten in ERP integrations. If your site does not maintain a consistent and structured history of each entity, search engines lose track of the meaning. Therefore, the integrity of your catalog does not lie in the rigidity of a relational database, but in the density and consistency of the connections in your data ecosystem.

It is no wonder that at this point solutions based on artifacts such as knowledge graphs and ontologies emerge. The combination of rigid control over terms (or products) and their relationships and the malleability of their structures is the perfect combination for maintaining catalogs that are alive and consistent.

The Corpus as Living Infrastructure and Hybrid Search

Between 2021 and 2026, the concept of corpus underwent a true conceptual revolution. But what is a corpus (whose plural in Latin is corpora)?

In Linguistics and Information Science, a corpus is a selected, finite, and structured set of texts assembled with the specific purpose of studying and analyzing language. Traditionally, it was a passive research material, a snapshot in time. However, with the rise of modern AI, the corpus began to be seen as an excellent tool for training, fine-tuning (fine-tuning), and validating large-scale language models and semantic retrieval systems.

To understand the evolution of content treatment in e-commerce, we can look at the way we structured information to compare the scenarios.

Before, we had only a static repository of texts, composed basically of loose descriptions stored in HTML and traditional databases such as MySQL. Today, by applying harvesting processes and structuring with ontologies and graphs, we have migrated to a new semantic approach. We transformed that static text into an active corpus that, connected to a knowledge graph, becomes the necessary foundation for feeding RAG systems, vector search, and artificial intelligence.

Making an analogy with Library Science, we had the classic figure of the librarian associated with the guardian of shelves. Today, this professional acts as an architect of information infrastructures. In Information Science, building a corpus requires systematic methods of automated collection (harvesting), rigorous definition of inclusion/exclusion criteria, and treatment of edge cases (edge cases) to prevent documentary noise from contaminating the created base.

Looking at the vanguard of information retrieval, we see that the great frontier is the fusion between dense vector search (based on mathematical embeddings that capture semantic proximity) and traditional lexical retrieval (based on exact terms and frequencies, such as the established BM25 algorithm).

Here at the Semantic Blog you will find posts discussing vector search and hybrid information retrieval systems. And if you follow me on LinkedIn, you can also find what I say about this subject.

The Practical Application of the Corpus in E-commerce

Imagine a large marketplace for sporting goods. Your product catalog, frequently asked questions (FAQs), consumer reviews, and technical specifications constitute the corpus of this business.

If this corpus is treated only as a pile of raw texts in relational tables, the internal search of your site will continue to commit elementary errors. As I always comment, if a user enters the store and types “running shoes”, a semantic system supported by a controlled vocabulary and a Knowledge Graph knows immediately that it should retrieve the entity “running shoes”, eliminating the frustration of the empty results page.

By integrating knowledge graphs with hybrid search, e-commerce responds both to searches for specific technical terms and to natural, conversational queries made to voice assistants and, likely, in generative AI engines.

Authenticity Resides in Provenance and Metadata, Not in the Isolated File

In the classical theory of Archival Science, there are two foundational pillars:

1. Principle of provenance: establishes that documents originating from the same institution, person, or activity should not be mixed with those from other sources.
2. Original order: dictates that records must maintain the arrangement and contextual flow in which they were created.

However, when we submit documents to AI systems that use Retrieval-Augmented Generation (RAG, or Retrieval-Augmented Generation), the original text is sliced into small blocks, the chunks, and converted into mathematical vectors within vector databases. In this process, the linear “original order” of the nineteenth century dissolves completely.

To understand the dynamics of the RAG process, imagine the path that information travels.

It all begins with a complete document, which must be sliced into small pieces, known as chunks, to then undergo vector indexing.

Vector indexing is the process of converting information, such as texts, documents, or images, into mathematical coordinates (known as embeddings) and organizing them in a database optimized for this format. Unlike traditional indexing, which stores exact words, vector indexing maps the semantic meaning of the content. It is this process that allows modern search systems and LLMs to compare the intent of a question with the context of text fragments, retrieving information by similarity of meaning, and not merely by exact keyword matching.

With this in mind, we see that the critical moment occurs in the retrieval phase of this information. If the system retrieves these loose fragments without adequate contextual metadata, the large-scale language model loses its main reference, which inevitably leads to AI hallucination. On the other hand, when this retrieval comes accompanied by a solid chain of provenance, we maintain the integrity of the origin and guarantee the delivery of a precise answer.

If the information fragment does not carry with it detailed contextual metadata about its origin, authorship, temporal validity, and conceptual relationships, it becomes an orphaned datum. And data without context are the fertile ground for the so-called “hallucinations” of large-scale language models, as we have already said. Again, the authenticity of a digital datum does not rest in the rigidity of the static file, but in the abundant and continuous capture of metadata throughout its entire life cycle.

The Practical Application of Provenance in E-commerce

Think of the exchange policy of an online store or the shipping conditions of a product. If your company changes the return policy from 30 to 7 days during holidays, and this information is fragmented in a RAG system without precise metadata on date, version, and scope, the customer service chatbot or search agent may retrieve obsolete excerpts and instruct the consumer incorrectly.

How do you solve this in your company? Tell me in the comments.

In semantic optimization, we solve this by adopting open metadata standards such as JSON-LD, the Schema.org vocabulary, and other solutions. Each product entity must be anchored with explicit properties of validity, brand, availability, and provenance, allowing search engines and indexing robots to trace exactly the source of truth for that statement.

Algorithmic Auditing and Curation Against Information Asymmetries

The construction and selection of databases are never neutral or purely technical acts; they are processes laden with sociocultural choices and perspectives. In Information Science, we discuss extensively how massive digitization and unregulated content aggregation can act as mechanisms of “data extractivism,” reinforcing hegemonic views and erasing minorities or regional knowledge.

To prevent intelligent systems from becoming echo chambers of historical distortions, librarians, archivists, and data curators use specific quality control and diversity metrics:

• Cultural Coverage Index (Cultural Coverage Index): a metric that assesses whether the linguistic, geographic, and demographic representativeness of a corpus reflects the diversity of the real domain or merely replicates the preferences of the institution that collected the data.
• Transparency Compliance Index (Transparency Compliance Index): an indicator that audits the clarity of data provenance, the methodological filtering criteria adopted, and the licenses applied.

To ensure quality and governance in the semantic domain, we typically rely on three fundamental indicators.

The first is the cultural coverage index, which measures diversity and helps mitigate representativeness biases in the collection. The second is the transparency compliance index, responsible for auditing the provenance of sources, the methodological criteria for collection, and usage licenses, and finally, we have the energy intensity score, which assesses environmental impact by measuring the energy consumption and carbon footprint required in the computational processing of these data.

The Practical Application of Algorithmic Auditing and Curation in E-commerce

In e-commerce, algorithmic bias can directly harm revenue and customer experience. Suppose a cosmetics and perfumery store. If the taxonomy and recommendation engine were trained with an unbalanced corpus, products specific to certain hair types or skin tones may never appear in automated showcases or in the main search filters.

Conducting a taxonomic audit in your store means assessing whether the navigation facets and vocabularies truly serve the entire consumer base. Strategic human intervention is irreplaceable for calibrating these nuances that no pure probabilistic model is capable of deducing with ethical sensitivity.

Data Sovereignty and the Balance Between FAIR and CARE Principles

If you have never heard of the FAIR and CARE principles, I like to think of them as two sides of the same coin when it comes to publishing and organizing any type of information on the internet.

The FAIR principles represent the technical side, focused on making technology work efficiently, and the acronym establishes that data should be:

• Findable: easy to find, like a book that has the exact label on the correct shelf in a library.
• Accessible: available to those who need to access them, with clear rules on how to obtain that information.
• Interoperable: capable of “talking” and exchanging information with other systems in a standardized way.
• Reusable: very well documented so that other people (or even a large-scale language model) can use that information correctly in the future.

Basically, the goal of FAIR is to ensure that machines and algorithms can read and process data freely, but it is precisely here that a problem arises: if everything is totally open and easy to use, how do we protect sensitive knowledge, cultural data, or people’s privacy?

It is to close this gap that the CARE principles exist, which bring the human, ethical, and protective side to the balance, as they argue that the use of data must always guarantee:

• Collective benefit: the information should generate value for the community from which it was extracted, and not only for those who collected it.
• Authority to control: the people or communities that generated the information have the right to decide who can access it and in what way it will be used.
• Responsibility: those who store and manage these data must be accountable and actively protect those who were the source of the information.
• Ethics: respect for human rights, culture, and people’s well-being must come before any technological or commercial interest.

In my day-to-day work, I see that the great secret is precisely to balance these two forces. Our challenge is to build data structures that search engine robots can crawl perfectly (FAIR), but that, at the same time, shield privacy and respect the origin of the information (CARE).

Historically, it was in the enthusiasm for the Open Science and Open Data (Open Data) movement that the paradigm of the FAIR principles was consolidated. And as we have already stated, they aim to make data broadly understandable and reusable by machines. However, Information Science warns us: treating any and all data as a resource for free exploration can violate individual privacy and disrespect the sovereignty of communities and indigenous peoples, for example.

It is for this reason, among others, that the FAIR principles must be balanced with the CARE principles.

A notable example of applying this balance is the Mukurtu CMS system, a collection management platform designed to preserve indigenous cultural heritage in conjunction with traditional knowledge labels (TK Labels—Traditional Knowledge Labels). These labels assign specific access protocols to records, determining who can access certain sacred knowledge, in which period, or under which community circumstances.

To visualize how this dynamic works in practice, we can understand the balance in data governance as a point of convergence between two complementary forces: on one side, we apply the technical focus oriented toward the machine through the FAIR principles, ensuring that our information is findable, accessible, interoperable, and reusable; on the other side, we apply the ethical and human focus of the CARE principles, ensuring collective benefit, authority to control, responsibility, and ethics over these same data.

When we unite these two approaches in our projects, the direct result of this convergence is structured and responsible access to information.

The Practical Application of the FAIR and CARE Principles in E-commerce

How does this concept connect to the world of e-commerce? Think about the governance of sensitive data from your customers and partners. Not all data generated by your operation should be exposed indiscriminately[cite: 1, 2].

Consider a marketplace that sells products from family agricultural cooperatives or indigenous handicrafts. Applying governance inspired by the FAIR/CARE binomial means granting structured interoperability so that the search engine understands the origin and sustainability of that product (via Schema.org), but respecting the licenses, pricing authority, and disclosure limits established by the producing community. Semantic technology must support the valorization of producers’ identity without transforming them into mere anonymous records.

Digital Sustainability and the Invisible Cost of Processing

The creation, constant retraining, and processing of large-scale computational models demand a massive amount of electrical and water resources. G1 explained how the use of artificial intelligence demands colossal generation of electrical energy. After everything you have read, you should already know that in evaluating the quality of a corpus, it is important to assess this impact.

The academic literature has incorporated the Energy Intensity Score (Energy Intensity Score), which monitors the ecological impact generated by massive data processing, as a response to this problem.

Building data sustainably requires combating waste and the proliferation of “digital waste,” which is why a sustainable data repository must be treated as a living organism that depends on:

1. Transparent versioning: meticulous recording of data evolution through collaborative platforms such as GitHub, ensuring auditability and reproducibility.
2. Statistical Disclosure Control (Statistical Disclosure Control): mathematical and anonymization methods that ensure the analytical value of data without exposing the individual privacy of the subjects involved.
3. Systematic format migration: active policies for updating structures and syntaxes, preventing knowledge from being held hostage by obsolete or proprietary formats (vendor lock-in).

The Practical Application of Digital Sustainability in E-commerce

In SEO for e-commerce, data sustainability translates into indexing efficiency. Platforms that generate thousands of duplicate pages through filter parameterization in URLs (without well-planned faceted navigation) not only waste the crawl budget (crawl budget) of search engines, but force continuous and unnecessary requests to servers, which are major consumers of all types of resources.

To understand the impact of structuring on data portability, we need only look at two contrasting approaches.

When we base our infrastructure on rigid relational models, the natural scenario that forms is difficulty in migration and the creation of information silos, where knowledge is isolated and rigid. In contrast, when we evolve to semantic formats, using structured languages such as RDF and OWL, we completely change this dynamic. This technical choice is what delivers true portability, facilitating the efficient reuse of information by different systems and, most importantly, ensuring the long-term longevity of our catalog.

We can go further and organize the catalog from clear ontologies and taxonomies in RDF (Resource Description Framework), avoiding redundancy, optimizing storage, and allowing requests made by intelligent agents to consume exactly the necessary semantic nodes, which reduces latency and global computational cost.

The Future Is Human and Conceptually Structured

At the end of all these lessons, I drew a conclusion and I think you, SEO specialists, can arrive at the same place I did. The answer to the challenges of artificial intelligence does not lie in producing more disordered volume of text to try to deceive robots; it lies in deep structure.

Artificial intelligence operates by calculating statistical probabilities of subsequent tokens. On the other hand, human intelligence, supported by the methodologies of Library Science and Information Science, can construct meaning through well-defined ontologies, controlled vocabularies, and open metadata standards.

An example is annotations in TEI (Text Encoding Initiative) for textual documents and CEDAR models for ontological representation in JSON-LD.

Create the table below, and it functions as an equivalence framework that connects academic theory to market practice. I needed to establish a direct parallel between structural instruments and frameworks originating from information science and archival science (such as TEI, CEDAR templates, and Mukurtu) and their corresponding applications in the day-to-day of semantic SEO and e-commerce.

You can quickly synthesize the central idea of this text, understanding that advanced optimization practices, such as the use of structured markup, governance of sensitive data, and the creation of audit trails in knowledge graphs, are not recent inventions of the digital market; they are, in fact, the technological implementation of structuring and information organization methodologies already validated and scientifically consolidated.

Comparative Framework: Academic Frameworks and Market Practices

When we apply these principles to our Semantic SEO work, we transform our sites and online stores into knowledge ecosystems that are accessible, trustworthy, and enduring.

The decisive question we must ask ourselves today is not about the quantity of pages our servers can store, but rather an ethical and strategic provocation: are we building databases that the future, mediated by humans and machines, will be able to understand and validate?

References Consulted and Recommended for Further Study

• ALMEIDA, Maurício Barcellos. Ontology in Information Science: Theory and Method. Curitiba: CRV, 2020.
• BRÄSCHER, Marisa; CAFÉ, Ligia Maria Arruda. Information Organization or Knowledge Organization?. In: National Meeting of Research in Information Science (ENANCIB), 2008.
• CORREA, R. et al. A metadata model for authenticity in digital archival descriptions. Data & Knowledge Engineering, 2023.
• LANCASTER, F. W. Indexing and Abstracting: Theory and Practice. Brasília: Briquet de Lemos, 2004.
• MUSEN, Mark A. et al. The Center for Expanded Data Annotation and Retrieval (CEDAR). Stanford Profiles, 2024.
• SILVA, Alexander Rodrigues. Semantic SEO: Semantic Workflow. Porto Alegre: Semântico, 2022.
• WILKINSON, Mark D. et al. The FAIR Guiding Principles for scientific data management and stewardship. Scientific Data, 2016.