Attending this year's AI for Good Global Summit was a reminder that artificial intelligence is no longer a future technology—it is rapidly becoming part of almost every aspect of society.

Much of the discussion focused on the role that international standards will play in enabling trustworthy AI. Topics such as governance, ethics, transparency, interoperability and safety featured prominently throughout the conference, reinforcing the message that standards will be essential to the responsible development and deployment of AI.

But one presentation in particular made me think about structured publishing in a completely different way.

Moving Beyond the "Black Box"

Large Language Models (LLMs) have become remarkably capable, yet one challenge remains: understanding how they arrive at a particular answer.

Researchers often describe this as the "black box" problem.

If an answer is incorrect—or simply ambiguous—it can be difficult to identify where the reasoning broke down.

Several researchers presented approaches that seek to address this challenge by introducing structured reasoning workflows.

Rather than asking one enormous language model to solve a complex problem in a single step, the task is broken down into smaller, specialised reasoning stages.

One AI agent tackles a specific aspect of the problem. Its output is passed to another specialised agent, which performs the next stage of reasoning. This process continues until a final answer is assembled.

Each stage becomes transparent and can be independently inspected, validated or improved.

Instead of one large "thinking process", the reasoning becomes structured, modular and traceable.

A Familiar Pattern

As I listened to these presentations, I couldn't help thinking that this sounded remarkably familiar.

For years, the publishing community has been making exactly the same argument about information.

A Word document or PDF is essentially a collection of formatted text. Humans understand the structure almost instinctively, but software has far less context.

Structured XML changes that.

Instead of presenting information as an undifferentiated block of text, XML identifies the role of every component: requirements, clauses, definitions, tables, figures, notes, normative references and metadata all become explicit pieces of information with clearly defined relationships.

In publishing, we have long understood that structure improves consistency, validation, interoperability and reuse.

AI research now appears to be reaching a remarkably similar conclusion.

More Structure, Better AI

There is sometimes a perception that technologies such as XML are becoming less relevant in the age of AI.

My impression after the conference was quite the opposite.

If anything, structured information is becoming more valuable.

Modern AI systems increasingly perform best when they are supported by well-organised, authoritative data rather than being expected to infer everything from enormous collections of unstructured text.

Techniques such as Retrieval-Augmented Generation (RAG), knowledge graphs and semantic indexing all reflect the same principle: give AI high-quality, structured information and it produces more reliable, more explainable results.

Rather than asking an AI model to search through an ocean of text, organisations can provide carefully structured content where relationships, metadata and meaning are already explicit.

That reduces ambiguity, improves traceability and often lowers the computational effort required to obtain accurate answers.

Standards Are Already Well Positioned

This is particularly significant for standards organisations.

Standards are among the most highly structured documents produced anywhere. Their carefully defined clauses, terminology, normative references and metadata make them ideal candidates for semantic representation.

Representing standards using Standards Tag Suite (STS) provides exactly the kind of structured foundation that modern AI systems increasingly benefit from.

When combined with complementary technologies such as knowledge graphs and semantic metadata, STS XML becomes far more than a publishing format—it becomes a trusted knowledge source that AI systems can navigate with far greater precision.

Investing in Structure Is Investing in AI

For many years, XML has sometimes been viewed simply as a publishing technology.

In reality, its value extends much further.

Structured XML enables validation, interoperability, accessibility and multi-channel publishing today.

Increasingly, it also provides the high-quality data foundation upon which trustworthy AI systems can be built.

Far from being made obsolete by artificial intelligence, structured content may become one of AI's most important enablers.

As the AI for Good Summit demonstrated, the future of AI is not simply about building larger models.

It is also about introducing more structure, more transparency and more reliable information into the way those models reason.

The publishing community has been working with those principles for decades.

Perhaps XML has been preparing us for the AI era all along.