AI companies publish usage reports regularly, yet these disclosures reveal limited insight into how people actually interact with their tools. Anthropic and OpenAI share data on user behavior, but the information remains heavily curated and often excludes the messy reality of day-to-day AI use.
The gap exists because companies control the narrative around their products. They highlight successful use cases and aggregate statistics while withholding granular data on failed queries, abandoned sessions, or unpopular features. This creates a skewed picture where marketing priorities outweigh transparency.
Understanding real AI usage matters. It determines which features get developed, how safety teams prioritize risks, and where researchers invest effort. If companies only see polished usage patterns, they miss critical feedback about what users actually want versus what they've been told to want.
Third-party researchers face obstacles accessing this data. OpenAI and Anthropic control the spigot. Academic studies rely on whatever companies choose to share, limiting independent verification of usage claims. This creates an asymmetry where the builders understand their products better than the broader research community.
Flock, a design-focused AI startup, takes a different approach. Rather than hiding behind aggregate metrics, the company has made specific product choices visible. This includes transparent documentation of how users interact with its interface and what workflows succeed or fail. The strategy reflects a belief that openness about real usage builds trust.
The contrast highlights a tension in the AI industry. Larger companies prioritize protecting proprietary insights and controlling their public image. Smaller players can differentiate by being forthcoming. Neither approach is inevitable, but the market currently rewards secrecy over clarity.
Users deserve better. Knowing how people actually use AI would help customers make informed decisions about which tools fit their needs. It would also pressure companies to build features users genuinely need rather than features that look impressive in marketing decks.
The data exists