AI usage data: Businesses face a black box problem

Picture a Fortune 500 manager sifting through glossy reports on ChatGPT or Claude, hoping for clarity on how their teams—or their competitors—might actually be using these tools. The reports, full of carefully selected statistics and glowing anecdotes, come straight from the AI companies themselves. But every graph and user story is handpicked. The raw data, the messy details, the real breadth of use cases—those stay behind closed doors.

Anthropic and OpenAI publish regular updates claiming millions of users and highlighting creative deployments. What they don’t share: breakdowns of failures, how often prompts lead to unsatisfactory answers, or how many users quietly churn out of frustration. Researchers and business analysts alike are left piecing together a puzzle with most of the pieces missing.

Unlike financial reporting or energy consumption, there’s no external audit of how people actually use commercial AI systems. Stanford researchers point out: if the company controlling the product also controls the narrative, how can anyone else verify what’s really happening?

For businesses investing in AI, this is more than an academic point. Without independent usage data, companies can’t benchmark their AI adoption or measure whether vendors’ claims align with practical outcomes. When procurement teams or risk managers ask for proof, they get curated marketing material instead of third-party validation. The result: decisions made on faith, or worse, on hype.

If a chatbot’s published numbers say 80% of users are satisfied, is that across industries? Regions? Use cases? Or is it a cherry-picked slice? Without independent scrutiny, it’s impossible to tell where AI deployments thrive—and where they fall short.

This lack of transparency creates a risk for businesses pursuing automation or customer service improvements. They may overlook hidden patterns like bias, hallucinations, or low engagement with key demographics. Legal and compliance teams remain in the dark, unable to anticipate regulatory scrutiny or public backlash if internal usage diverges from the shiny numbers in vendor slide decks.

The uncertainty trickles down to budget planning, workforce strategy, and customer engagement. Should a retailer invest in AI-powered support based on a vendor’s glowing stats, or wait until independent usage benchmarks appear? Is the productivity boost real, or a mirage sustained by selective reporting?

In the projects we run, we've seen clients ask for hard evidence that AI deployments work for peers in their industry—not just in a vague, global sense, but with specifics relevant to their use case. Without transparent data, those questions rarely get satisfying answers. This leaves organizations with a stark choice: experiment blindly, or risk missing out entirely.

Some in the research community are calling for external mechanisms—think of the software analytics and independent certification that mature industries rely on. Such monitoring could offer businesses a credible map of AI’s strengths and hazards, not just a manufacturer’s brochure.

For now, enterprises must read between the lines and demand more than self-interested reporting. Until independent usage audits are standard, every AI adoption decision carries an extra layer of uncertainty—one that even the most sophisticated LLM can’t remove.