Structured power: AI backs XBRL for energy data
Following their recent experiment with Johnson & Johnson’s financials, XBRL US has taken their AI testing into the energy sector, asking whether structured data improves how large language models handle public utility filings.
Using Anthropic’s Claude AI tool, they ran a series of queries comparing structured XBRL data from FERC filings with non-XBRL sources for Puget Sound Energy’s electricity sales. The results were definitive: XBRL data was not only accurate, but also richer, breaking out figures by customer type and quarter. The unstructured sources? Inconsistent, sometimes wrong, and often missing entire datasets.
This experiment echoes what we saw with J&J: structured data empowers AI to deliver clarity and depth. Without it, models rely on a patchwork of third-party sites and PDFs, producing results that are incomplete or misleading.
AI isn’t a substitute for data standards, it thrives because of them. And for regulators and market participants, the message is that structured data isn’t optional if you want trustworthy machine analysis.
Read the full energy data experiment here.
