}

Accessible PDFs and AI: Challenges for Investor Relations

Companies dedicate significant resources to preparing earnings reports and presentations, equity stories, corporate documents, and more. But today, a question is becoming increasingly relevant: are these documents prepared to be correctly interpreted by artificial intelligence models?

For years, the PDF has established itself as a benchmark format for corporate and financial communication. Its ability to preserve design, typography, and layout has made it a particularly suitable solution for distributing information consistently.

But that same strength can become a limitation, as a document may offer an excellent visual experience while simultaneously presenting significant difficulties for systems that need to automatically extract, interpret, and process its content.

Artificial intelligence is transforming how we access information, giving document accessibility a new strategic dimension.

A visually perfect PDF is not always an accessible document.

When a company publishes an earnings report, an equity story, an annual report, or a corporate presentation, it usually pays special attention to the clarity of the messaging, the quality of the design, and the visual representation of data. However, there is a significant difference between how a person views a document and how an AI model interprets it.

For a reader, identifying a headline, following the order of paragraphs, or relating data in a table is usually simple. For certain computer systems, these tasks can be considerably more complex.

The problem arises when the document lacks an internal structure that allows for the recognition of which elements are titles, which are paragraphs, how table cells relate to one another, or in what order the information should be read. In other words, accessibility does not depend exclusively on what we see, but also on how the document is built.

WCAG and PDF/UA: two complementary standards.

Document accessibility relies on international standards that establish the necessary conditions to facilitate access to information.

On one hand, the Web Content Accessibility Guidelines (WCAG) define the general principles that digital content (websites or applications) must meet to be perceivable, understandable, and robust. On the other hand, the PDF/UA (PDF Universal Accessibility) standard establishes specific technical requirements for creating accessible PDF documents.

Both frameworks are complementary: while WCAG sets general accessibility criteria, PDF/UA focuses on aspects such as content tagging, semantic structure, and logical reading order. Key features of an accessible PDF include:

· A correctly identified hierarchy of titles and subtitles.

· A reading order consistent with the organization of the content.

· Tables that allow for the recognition of rows, columns, and headers.

· Alternative text for images that conveys relevant information.

· Selectable and editable text, along with adequate contrast.

· Elements that facilitate navigation through the document.

The new challenge: when an AI model reads the report

The expansion of language models and document analysis tools are creating a new landscape.

It is becoming increasingly common to use AI systems to summarize documents, identify financial indicators, compare business results, or locate relevant information in extensive reports. However, the fact that an AI model can process a PDF does not necessarily mean it interprets all the information correctly.

Many systems subject documents to a process of text extraction, element recognition, and structural reconstruction. If this conversion causes information loss or alters the relationships between content elements, the quality of the responses may be compromised.

There are several particularly important issues:

1. Reading order: documents containing multiple columns, highlighted tables, marginal notes, or complex visual layouts can cause some systems to extract text in an order other than intended. As a result, paragraphs belonging to different sections may become mixed up and lose their original context.

2. Financial tables: tables are among the most important elements of any financial report. When a tool extracts their content without properly preserving the relationship between rows, columns, and headers, it can lead to misinterpretations. For example, an EBITDA figure for 2026 could be mistakenly associated with 2025, or a percentage change could be attributed to a different figure. In the financial sector, these types of errors are by no means insignificant.

3. Charts and infographics: a significant portion of corporate information is conveyed through charts, diagrams, and other visual resources. Although AI models can analyze images, their ability to correctly interpret complex graphics depends on factors such as resolution, label legibility, available context, and the tools used. When relevant information appears exclusively within an image, its retrieval can be less reliable.

4. Loss of context: a figure taken in isolation does not always provide sufficient information. To correctly interpret a financial indicator, it is necessary to know its reporting period, unit of measurement, consolidation perimeter, and accounting treatment. When the data extraction process separates a data point from its headers, explanatory notes, or definitions, the risk of misinterpretation increases.

Accessibility is also a matter of information quality

Properly structured documents provide additional information about the function of each element and the relationships between them. This structure can make it easier for certain tools to identify headers, extract relevant snippets, and maintain data relationships.

However, compliance with accessibility standards does not, by itself, guarantee that any artificial intelligence system will correctly interpret a document. Results also depend on the extraction software, the model's capabilities, and the process used to analyze the information.

Therefore, accessibility is a factor that facilitates content reuse, but it does not replace the need to verify that the information has been interpreted correctly.

What are the implications of this for Investor Relations?

Investor Relations departments are responsible for ensuring that the market has access to accurate, consistent, and sufficient information to understand the company's performance, evaluate its results, and form an idea about its future.

Traditionally, much of this effort has focused on content quality, the clarity of financial messaging, and consistency with the company's investment thesis. However, the increasing use of artificial intelligence tools introduces a new variable: the ease with which automated systems can retrieve and interpret published information.

A financial report may contain all the necessary data and yet hinder analysis if it presents structural or data extraction issues. This can affect tasks such as identifying operating indicators, comparing periods, analyzing business segments, or retrieving historical information. Therefore, the preparation of corporate documents must take into account not only their visual presentation but also their accessibility and suitability for digital processing. In other words, it must be ensured that both people and artificial intelligence models can access the same information consistently.

From visual design to structural design

The solution is to incorporate accessibility criteria from the earliest stages of conceptualization, drafting, and layout, while ensuring the document remains attractive, with elaborate graphics and striking visual compositions. To achieve this, one must work with well-defined content structures, use semantic tags correctly, preserve data relationships, and ensure that the document maintains its coherence when text is extracted.

It also requires combining automated verification tools with manual checks. Automated tagging solutions can handle part of the workload, but they do not always correctly identify the structure of complex documents, especially when they include financial tables, charts, or multi-column layouts.

To be certain, the ideal approach is to provide financial data in complementary formats, such as Excel or HTML, as this will facilitate its analysis and reuse.

A new dimension in financial communication

The digital transformation of Investor Relations is not limited to incorporating new communication channels or analytical tools; it also requires rethinking how corporate information is prepared, organized, and distributed.

The quality of financial communication must be addressed holistically: from defining messages and crafting the equity story to the design, production, and distribution of documents. PDF accessibility is no longer just a technical issue; it has become a factor that contributes to improving the quality of communication.

For publicly traded companies, having accessible, well-structured, and digitally ready documents can make life easier for investors, analysts, and other stakeholders. They must ensure that this information can be found, understood, and interpreted correctly, regardless of who, or what, is accessing it.

-

Daniel Campoy Silva

Founding Partner of Sigma Rocket