Why OpenAI’s Claim About EU Business AI Needs A Numbers Check

🔍 Read the full analysis: Why OpenAI’s Claim About EU Business AI Needs A Numbers Check on ThorstenMeyerAI.com

OpenAI said its Codex and ChatGPT Work agents generated 55% of output tokens in the company’s ChatGPT Business and Enterprise accounts across the 27 EU member states by Sept. 30, 2026. The figure is a measure of generated text, not the share of business tasks or users relying on agents, a distinction that matters as the company presents the data as evidence of Europe’s growing use of workplace AI.

OpenAI’s Economic Research team published Europe’s Agentic Opportunity: Moving from Breadth to Depth with AI Agents on Oct. 14, alongside a post from its Global Affairs team. The report defines “agentic” use as activity involving Codex, OpenAI’s coding agent, or ChatGPT Work, an agent mode introduced in July that can work with files, browsers and company tools. The remaining 45% of output tokens came from standard ChatGPT use, according to the report.

For businesses with 200 or fewer users, OpenAI reports that weekly active users of Codex or ChatGPT Work grew 8.1 times from the first week of May to the week ending Oct. 5. Every EU member state recorded at least fivefold growth in this measure. Belgium, Italy and Luxembourg had the largest reported increases, at 12.1, 10.8 and 10 times, respectively. OpenAI does not provide absolute user counts alongside those growth rates.

Reported agentic output shares varied sharply by country: Lithuania was at 85% and Cyprus at 81%, while Slovenia was at 22% and Austria at 26%. Germany recorded 51%, ranking 17th among the 27 states. The report also says EU frontier firms—defined as the top 10% by output tokens per active user—generated about four times as many tokens per user as typical EU firms. Frontier firms outside the EU produced more than twice as many tokens per user as the EU frontier group.

What the 55% Figure Measures

The report offers a detailed view of activity within OpenAI’s own business accounts, but its headline metric is output tokens: the text generated by a model. It does not establish that agents perform most workplace AI tasks, reach most employees or deliver most economic value. OpenAI itself describes tokens as an imperfect measure of business value and says agentic tasks can require substantially more computation.

An agent may read files, plan, write, check its work and repeat steps, producing more output than a short chat response. The source article illustrates the possible scale using an assumption—not a figure from OpenAI’s report—that an agent run produces 10 to 30 times the output of a chat reply. Under that assumption, a 55% token share could correspond to roughly 4% to 11% of interactions. That is an illustrative calculation, not a measured estimate, and the report does not give the data needed to determine the actual interaction share.

The distinction matters for companies and policymakers weighing whether European businesses are adopting AI agents broadly or whether a smaller number of lengthy, compute-intensive workflows account for much of the measured activity. The figures show where OpenAI’s account output is going; they do not, on their own, show productivity gains, savings or business outcomes.

A Launch Shapes the Growth Curve

The reported increase spans a period that includes the launch of ChatGPT Work. OpenAI’s chart, as described in the source material, shows Codex accounting for roughly one-fifth of EU output tokens in May and 34% by Sept. 30. ChatGPT Work was launched in July and represented 21% of output tokens by the September cutoff. The source analysis calculates that Work contributed more than half of the roughly 35-percentage-point rise in agentic share over that period.

The small-business growth comparison also starts before ChatGPT Work existed. OpenAI’s report says a major part of the increase followed the product’s introduction. That makes the 8.1-times result a record of use across a baseline and follow-up period that includes a new product—not a clean measure of organic growth independent of product changes. OpenAI’s policy conclusion is that Europe should move from broader AI adoption to deeper use, including worker skills and access to advanced tools.

The country figures also need a customer-base caveat. OpenAI reports no country-by-country account or user sample sizes. It notes that relatively high agentic use in smaller countries such as Cyprus, Estonia and Lithuania may reflect usage at larger firms. The source analysis finds little relationship between countries’ agentic-output rankings and small-business growth rankings; that comparison does not establish why the measures differ.

What the Data Leaves Out

The report does not disclose absolute user counts behind the 8.1-times small-business growth figure, or account and user sample sizes for each country. Without those details, readers cannot tell how much the rates reflect a small starting base, a few large customers or broader adoption. The company’s customer data also cannot by itself represent all European businesses.

OpenAI’s published figures do not establish how many tasks agents completed, how many employees used them, or whether use improved productivity, quality or costs. The report’s token measure is affected by the computational demands of agent workflows, and the period includes the launch of ChatGPT Work. The source material also offers an illustrative interaction estimate based on an assumed token ratio; that is not an OpenAI finding. Further detail on methods, customer composition and outcomes would be needed to answer those questions.

The Next Test: Business Outcomes

The immediate next step is for OpenAI to provide more detail on the report’s methodology, including country-level sample sizes, absolute active-user counts and how it distinguishes agentic output from other account activity. Later reporting could show whether the growth continues beyond the initial months after ChatGPT Work’s launch.

For businesses and policymakers, the more consequential evidence will be measures tied to work completed and results: which tasks agents handle, how often people rely on them, and whether those workflows improve speed, quality or cost. Until such measures are available, the report is best read as a snapshot of OpenAI customer activity, not a comprehensive measure of European business adoption or economic impact.

Key Questions

What does OpenAI mean by a 55% agentic share?

It means Codex and ChatGPT Work generated 55% of output tokens in OpenAI’s EU ChatGPT Business and Enterprise accounts by Sept. 30, 2026. It does not mean agents performed 55% of tasks or were used by 55% of employees.

How quickly did small-business agent use grow?

OpenAI reported an 8.1-times increase in weekly active users of Codex or ChatGPT Work among businesses with 200 or fewer users, comparing the first week of May with the week ending Oct. 5. It did not provide the absolute user count behind the rate.

Did ChatGPT Work affect the reported growth?

Yes. ChatGPT Work launched in July 2026, during the period covered by the report. OpenAI says a major part of small-business growth followed its introduction, and the product accounted for 21% of EU output tokens by Sept. 30.

Do the country figures show which EU nations are most advanced in AI?

No. They describe agentic activity in OpenAI’s customer accounts, and OpenAI did not publish country-level sample sizes. Customer mix and the presence of a few large users could affect national figures, so they should not be treated as a ranking of overall national AI readiness.

Does the report show that AI agents improved business results?

No. It reports usage and output-token measures, not demonstrated changes in productivity, costs, revenue or work quality. Those outcomes remain unestablished by the figures presented.

Source: ThorstenMeyerAI.com