How Organizations Use AI: Evidence from ChatGPT — Key Findings from OpenAI's Study

OpenAI released a study analyzing how organizations actually use ChatGPT, based on 1.8 million messages from 8,198 organizations. The data offers a ground-truth look at where AI fits in real workflows — and it's not where the hype is.

Key Findings

• Creative vs analytical tasks: 71% of messages were creative in nature, 29% analytical. Creative includes writing, brainstorming, and problem solving; analytical covers data analysis, coding, and querying.
• Most frequent tasks: Writing (28%), data analysis (20%), troubleshooting (7%), information retrieval (6%).
• Business context matters: ChatGPT use skews toward the 'innovating' domain (60%) — creating new products or services — vs 'production' (15%) and 'coordinating' (13%).
• Topic clusters: Common topics include marketing, product design, business strategy, software development, and research.

What This Means for AI Agents

For developers building or using AI coding agents, the takeaways are practical: writing support and data analysis are the highest-value use cases. Organizations aren't just asking for code — they're using AI to draft proposals, analyze spreadsheets, and debug infrastructure. An effective agent needs to handle multi-step reasoning across those domains, not just autocomplete functions.

The study also highlights that 'troubleshooting' — a core task for AI agents — represents only 7% of usage, suggesting agents should prioritize assistance with writing and data workflows to match real demand.

Methodology and Caveats

The study draws on a random sample of messages from organizations that subscribed to ChatGPT, filtered to exclude individual users. It's a snapshot of one platform's ecosystem — not a census of all AI use.

For a deeper dive, including charts on task distribution and industry patterns, read the full PDF.