OpenAI just published a landmark research paper — "How Organizations Use AI: Evidence from ChatGPT" — that provides the first large-scale, data-driven look at how enterprises are actually using generative AI. The study, which hit 116 points on Hacker News, analyzed over 1,500 organizations and 17 million messages from ChatGPT Enterprise accounts through March 2026. The findings paint a picture of AI adoption that is both more widespread and more uneven than previously thought.
The Data: Unprecedented Access
This isn't a survey. It's actual usage data. The researchers — a team from OpenAI, Columbia Business School, and Wharton — linked ChatGPT Enterprise account records to:
- Usage data: Message-level records of how employees interact with ChatGPT
- Worker roles: Job title information and seniority classification
- Task classifications: What kind of work employees are actually doing with AI
- Public company financials: Matched to U.S. public company data for adoption analysis
This is a privacy-preserving analysis at a scale that surveys could never achieve. The worker-level sample at the six-month adoption horizon includes over 1,500 organizations and over 17 million messages.
Finding 1: Explosive Growth — 7x in 9 Months
Aggregate output tokens produced by ChatGPT Enterprise customers grew roughly sevenfold between June 2025 and March 2026. This is a staggering growth rate — roughly 25% month-over-month compounded.
What's remarkable is that this growth isn't just from new customers. Among firms that had already adopted by June 2025, output tokens increased roughly fourfold. This means about half of the total growth came from existing adopters using AI more intensively, not just from new firms signing up.
The implication: AI adoption isn't a one-time event. It's a process where organizations gradually find more use cases and increase usage intensity over time. The "ramp" from initial adoption to full utilization takes months, not weeks.
Finding 2: Adoption Correlates with Scale and R&D Investment
Among U.S.-based public companies, ChatGPT Enterprise adopters are:
- Larger (more employees, more revenue)
- More valuable (higher market capitalization)
- More R&D-intensive (higher R&D spending relative to revenue)
- More SG&A-intensive (higher selling, general & administrative spending)
This pattern suggests that early enterprise AI adoption is associated with greater prior investment in intangible and organizational capabilities. Companies that already invest heavily in knowledge work infrastructure are the ones adopting AI first.
The implication: There may be a "digital divide" forming in AI adoption. Smaller companies and those with less existing knowledge work infrastructure risk falling further behind. The benefits of AI may accrue disproportionately to organizations already at the frontier.
Finding 3: Early-Career Workers Use AI the Most
Usage within adopting firms is broadly distributed across job title classes and seniority levels — but with significant heterogeneity in intensity:
- Marketing and communications workers send more messages than executives
- Early-career workers send many more messages than senior employees
- Trainees account for a meaningful share of total usage
This is counterintuitive. You might expect senior employees — who have more complex tasks and more strategic responsibilities — to be the heaviest AI users. Instead, it's the junior employees who are lean into AI most aggressively.
Why? The paper doesn't speculate, but several explanations are plausible:
- Lower switching costs: Early-career workers have fewer established habits and workflows to disrupt.
- More task-based work: Junior employees often handle more repetitive, well-defined tasks that AI handles well.
- Less risk aversion: Senior employees may be more cautious about AI reliability and data security.
- Organizational permission: Companies may explicitly encourage AI use among junior staff while senior leaders wait for clearer ROI evidence.
The implication: AI's impact on the workforce may be bottom-up rather than top-down. Junior employees who master AI tools may develop a productivity advantage that accelerates their career progression — or they may find their tasks increasingly automated.
Finding 4: AI Is a General Purpose Technology for Knowledge Work
ChatGPT Enterprise usage encompasses a broad range of knowledge work tasks:
- Writing (the most common use case)
- Communication (drafting emails, messages, reports)
- Information synthesis (summarizing, extracting key points)
- Technical work (coding, data analysis, documentation)
- Research (gathering information, exploring topics)
- Planning (project planning, meeting preparation)
- Legal and regulatory work (document review, compliance)
- Finance (analysis, reporting)
This breadth is consistent with generative AI functioning as a general purpose technology for knowledge work — a tool that isn't specialized for one task but can be applied across many different domains.
The implication: AI's economic impact won't be concentrated in one industry or one job function. It will be distributed across the entire knowledge economy, affecting virtually every white-collar role.
The Big Picture: Broad but Uneven Adoption
The paper's conclusion is perhaps its most important finding: enterprise AI adoption is a broad but uneven organizational phenomenon.
- Broad: Adoption spans 1,500+ organizations, many job functions, and all seniority levels
- Uneven: Adoption is concentrated among larger, more R&D-intensive firms; usage intensity varies dramatically across worker groups
The long-run economic value of enterprise AI adoption will depend on whether dispersed individual use develops into complementary organizational capabilities. In other words, individuals using ChatGPT to write emails faster is nice, but the real productivity gains will come when organizations redesign their workflows around AI capabilities.
What This Means for Developers and Businesses
For Developers
- AI tooling is being adopted across the enterprise, not just in tech companies
- The most common use cases (writing, communication, synthesis) suggest opportunities for tools that make these workflows even better
- Early-career adoption patterns suggest a generational shift in how AI is integrated into work
For Businesses
- If you're not adopting AI, your competitors who are larger and more R&D-intensive probably are
- The "ramp" from adoption to full utilization takes months — start now
- The biggest gains will come from organizational redesign, not just individual tool use
- Early-career employees may be your best internal advocates for AI adoption
For AI Providers
- Enterprise growth is driven by both new customers and deepening usage among existing ones
- The breadth of use cases suggests that general-purpose AI tools (like ChatGPT) have more enterprise potential than specialized tools
- The correlation with R&D intensity suggests that marketing to more innovative companies may be more effective
Methodology Note
This paper is a working paper, and the authors note that "results are subject to change." The data comes from ChatGPT Enterprise accounts, which represent a specific segment of the market (organizations that pay for enterprise AI). Consumer AI usage and small business usage may follow different patterns. The findings should be interpreted as evidence about enterprise AI adoption, not AI adoption in general.
Based on "How Organizations Use AI: Evidence from ChatGPT" by Chatterji, Holtz, Tambe, Rakholia, and Weeratunga. Last updated August 11, 2026.
Top comments (0)