Your report has a new stakeholder, and it doesn’t blink

Sustainability reporting has never been a single-audience exercise. The same report lands on the desk of an ESG rater building a scorecard, an investor weighing risk, a regulator checking for compliance gaps, a procurement officer running a supplier assessment and a prospective employee deciding whether your values match theirs. The report, the hub, the data appendix, the framework disclosures — the whole suite exists because different audiences need different entry points into the same underlying story.

But how these assets are read HAS changed.

According to PwC’s Global Investor Survey 2025 published in June 2026, 62% of investors now use AI to analyze company filings and earnings call transcripts. Regulators, ratings agencies and procurement teams are deploying AI as well.  Brunswick Group’s 2026 US investor survey puts a finer point on the trend: 68% of investors say AI has already changed how they approach earnings calls, and 54% now call generative AI at least moderately important to their research process. Regulators, ratings agencies and procurement teams are deploying AI as well. IOSCO, which sets standards for securities regulators across more than 130 jurisdictions, has found AI now embedded throughout capital markets, from investment research to regulatory surveillance itself.

That means the same report read by multiple audiences in the past is now being processed by multiple AI systems, each configured to look for something different, each capable of cross-referencing your disclosures against third-party sources and prior-year filings in seconds. This can have far-reaching implications for your reputation, rankings on lists, risk exposure and beyond.

Consistency is key

As an example, previously, an error or unclear point in your report might be caught by an attentive reader who could put it in context or seek clarification. Today, large language models are synthesizing information across reports, websites, data tables and disclosures. An inconsistency that once lived in a single document can now influence how multiple stakeholders understand your company. The reporting suite that teams have built over years is no longer being read one document at a time, it is being interpreted as a single interconnected signal.

It’s also worth noting that AI doesn’t read the way a careful human does. A person reading “we’re targeting significant emissions reductions in the coming years” understands it as a directional, good-faith commitment. A language model scanning the same sentence for a signal — a number, a date, a defined scope — may simply find none and score the disclosure as vague or unsubstantiated rather than extending the benefit of the doubt. The fix isn’t necessarily softer language; it’s pairing every directional claim with the specific figure, timeframe or boundary that backs it up, wherever that claim appears.

This puts pressure on the controls that sit upstream of publication:

• Data governance that ensures metrics are defined and applied consistently across every document.
• Review processes that trace narrative claims back to substantiated data.
• Sign-off protocols that treat the full suite as a single disclosure, not a collection of separate outputs.

These are existing requirements that now carry significantly higher stakes with AI as the reader on behalf of your existing stakeholders.

What this means for your next reporting cycle

To gauge the strength of your reporting suite controls? and to best serve your newest non-human audience, we recommend asking several questions before the next cycle begins:

• Is there a defined owner of not only each individual data point, but also an owner of data consistency across every document and platform?
• Are metrics and their boundaries documented in one place and referenced wherever else they’re used to reduce the risk associated with each manually transcribed number?
• Does your review process explicitly check for inconsistencies between the narrative layer and the data layer across all publications, whether this content is updated annually or not?
• Is structured, tagged data (such as XBRL/iXBRL, or at minimum consistent metadata and defined units) attached wherever a figure appears so that AI systems can extract it directly rather than re-deriving it from prose?

Companies that have invested in robust reporting controls (such as the topic blocks we introduced in a recent webinar or on a platform like Workiva) will be better positioned for this new era of AI-mediated scrutiny. When disclosures are consistent across reports, web content and data sources, AI is more likely to surface an accurate, coherent picture of the company. In other words, there’s a new stakeholder in town, one that rewards organizations that can tell the same story, no matter where it looks.