OpenAI publishes a vague report on the state of AI in companies

Contents

In the inaugural edition of an annual report on the state of AI in companies, OpenAI aims to provide a comprehensive overview of how companies are using AI, the benefits employees see in it, and how executives are translating experiments into measurable productivity and new capabilities. To this end, the AI provider refers to users of ChatGPT Business and Enterprise offerings, as well as a survey on AI adoption among 9,000 employees from nearly 100 companies. The provider does not disclose the size or origin of the participating companies – nor whether only users of OpenAI's products were surveyed. ChatGPT Business and Enterprise are available exclusively as annual subscriptions. Business costs $29 per user per month, and Enterprise pricing is only available through direct contact with OpenAI sales. Access to OpenAI's best models is flexibly designed, with the note "additional tokens can be purchased at any time."

The Numbers

The report opens with the statement that OpenAI is now renting out seven million ChatGPT Business seats and that demand for ChatGPT Enterprise has increased sevenfold. Compared to the previous year, the number of messages employees send to the chatbot on average has increased by 30 percent. Many use GPTs or Projects in their work environment. Projects are grouped chats, possibly supplemented with context. GPTs can best be understood as chats or agents for specific tasks. The most widespread GPTs either turn company knowledge into reusable assistants or automate workflows by integrating with internal systems.

The company also sees strong growth in API usage. Reasoning models and the programming assistant Codex, in particular, are seeing 320 times more messages than last year. As a metric, OpenAI states that more than 9,000 companies have processed over ten billion tokens through the LLM systems, while nearly 200 companies have surpassed the 1 trillion token mark in consumed tokens. According to OpenAI, this indicates the systematic integration of better models into products and services.

Measurability Promised, Survey Provided

The report further speaks of measurable results from AI use and cites an employee survey in this regard, where 75 percent of respondents report being able to work faster or achieve better results. On average, AI tools save 40 to 60 minutes of working time on days they are used. Data scientists, engineers, and employees in corporate communications can even save 60 to 80 minutes daily. Chatbots are most efficient for accounting and finance.

According to OpenAI, those who truly save hours use eight times more credits than the average user of the technology. Power users of Deep Research, GPT-5 Pro, and image generators – i.e., the most resource-intensive techniques – report saving up to 10 hours per week. This reinforces the "more is more" narrative of AI scaling, which has recently come under criticism. OpenAI observes that power users in the top 5 percent of users employ the available tools more than 10 times more frequently than the median users of the offerings. The report lists writing and communication, programming, generating instructions, information retrieval, analysis and calculations, and creative media use as usage examples, in descending order of usage gap.

Contextualizing the Report

While the report is brimming with large numbers and high factors, in most cases, the starting sums are missing to represent actual values. A factor of 10+ is not very informative if readers do not know the initial values. The same applies to time savings. Here, OpenAI mentions credits, which are only a vague approximation of usage and do not appear in this form in the provider's price tables. Advanced features like Codex or Deep Research consume significantly more. OpenAI also does not want to talk about ratios: "Credits map to usage, with more advanced features like Codex and Deep Research consuming an higher number of credits." Even where OpenAI specifies tokens, exact costs cannot be determined, as token consumption varies greatly depending on the task and model. Thus, the report remains little more than advertising with its figures and is of little help for strategic decisions.

Here you can find OpenAI's press release and the actual report.

(pst)