Optimising reporting for AI.
The rise of AI is transforming access to corporate information, revealing fundamental problems with traditional reporting design processes. As a result, business-critical annual and sustainability reporting information is not reaching readers as accurately as it should. A digital-first approach is the way to optimise AI searches.
Author: Jon Rowden
When AI or Google searches deliver information from third-party sources, companies risk losing control over the facts, due to AI tools’ confident but frequently error-prone answers, known as ‘hallucinations’.
This white paper explains what causes these errors, and why ‘digital-first’ reporting processes (using ‘native HTML’ as the source for document production) are essential to ensure content is accessible, accurate, and trustworthy for AI and human readers alike.
What is digital reporting, and how is it optimised for AI?
Digital-first reporting is a very simple change in a corporate report's design process. Instead of a print-first process using software like InDesign, digital-first software uses HTML code as the source to create all formats (PDF, web, filings, print). This is often referred to as ‘native HTML’.
Crucially, the digital outputs from this process include the meaningful structure, tags and data which AI needs. This creates a digital route map that AI can find, read and understand. This dramatically improves AI accuracy.
Academic testing involving over 24,000 results has proven that annual reports published online in HTML were cited as sources by ChatGPT three times more frequently than PDF annual reports. If relevant HTML content is available online, AI tools will usually ignore PDF sources by default.
AI has changed how all information is accessed
Investment professionals increasingly use Agentic AI, in the form of bots, to follow instructions, filter information, and populate financial investment models with a high degree of autonomy. One study identified 175 separate AI bots interacting with digital-first annual report data. These bots seek HTML content, not PDFs.
To find out more, simply ask any AI tool why AI finds tagged HTML easier to read than PDF for annual reports.
The importance of high quality ‘tagging’
Although HTML is materially more AI-readable than PDF, HTML code alone is not enough. AI also relies on meaningful structure provided by additional data ‘tags’ within the code. Again, ask your favourite AI tool and it will explain why data tagging is crucial for investment-grade analysis. Data tags are placed in the HTML code. Invisible to humans, these tags turn words and numbers into the essential digital structure for AI machines to automatically search, read, and compare, doing so efficiently, reliably and comparably. Tagging a full digital report with XBRL and JSON-LD transforms AI’s accuracy when it seeks, extracts and analyses content.
Just like the optimisation of websites, this is now required in reporting in order for AI accuracy to improve and for AI’s role as a high value, trusted assistant to be realised.
How digital-first processes aid compliance controls
The digital-first process integrates many previously separate tasks, such as XBRL tagging, which enables the processes to be completed earlier, and at calmer moments. This means that checks and controls can be more effective which is vital for good governance and supports compliance with regulatory requirements, for example the new Provision 29 controls declaration being introduced in the UK.
How AI is changing the landscape
AI is becoming a foundational technology, requiring organisations to develop long-term strategies for AI-related services and processes. As AI use cases and risks are evolving, it seems clear that humans will remain responsible for the decisions and judgements, using AI as an efficient “servant” for a wide range of tasks – where it can be proven to be reliable.
There is much focus on how AI tools can be used to assist in many different areas, but perhaps less focus on the biggest barrier to using it professionally: AI errors. Understanding the cause of these errors is key – they are frequently not caused by AI, but by lack of suitable digital sources. Too much vital information is locked in inaccessible sources like PDFs.
How will this affect reporting?
Reporting has remained comfortably in a largely PDF-based process, until now. But in an AI world, the format and data quality of all published content, including reporting, will become the key differentiator in building trust in AI analysis.
Reporting is among the most trusted and complete content most companies publish. This can be formatted for AI and search tools to access and process reliably by using a digital-first process. PDF is not a suitable source for this.
So when reporting teams are assessing their options, AI-readability naturally sits amongst the established key criteria of process efficiency, control, design, tagging and accessibility when considering value for money and the best way forward.
Conclusion
Over the coming years, companies will focus on the quality and accessibility of their digital reporting format and data to optimise the AI readability of their reports. As has been explained by regulators, data scientists and AI tools themselves, reporting which is created directly from ‘native HTML’ sources will be the best solution.
