What The Denominator Reveals About OpenAI’s EU Business AI Claim

🔍 Read the full analysis: What The Denominator Reveals About OpenAI’s EU Business AI Claim on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

OpenAI said agents produced 55% of output tokens in its ChatGPT Enterprise and ChatGPT Business accounts across the European Union by September 30, according to a report published October 14. The number measures text generated by the company’s AI systems—not the share of business tasks, users or economic value involving agents—so it is evidence of growing use inside OpenAI’s customer base, not proof that most European workplace AI use is agentic.

OpenAI Economic Research’s report, Europe’s Agentic Opportunity: Moving from Breadth to Depth with AI Agents, covers business accounts in the EU’s 27 member states. It classifies Codex, OpenAI’s coding agent, and ChatGPT Work, an agent mode launched in July 2026, as agentic tools. By September 30, Codex accounted for 34% of output tokens and ChatGPT Work for 21%; conventional ChatGPT accounted for the remaining 45%.

The report also says weekly active users of Codex or ChatGPT Work at businesses with 200 or fewer users increased 8.1 times from the first week of May to the week ending October 5. OpenAI says every EU member state grew at least fivefold. Belgium, Italy and Luxembourg had the largest reported multipliers, at 12.1, 10.8 and 10.0, respectively. The company did not provide absolute user counts in the supplied report material.

Agentic output shares varied substantially by country: Lithuania was reported at 85% and Cyprus at 81%, while Slovenia was at 22% and Austria at 26%. Germany’s share was 51%, placing it 17th among the 27 member states. OpenAI also reports a gap between its highest-use customers and the global frontier: 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 generated more than twice the output per user of the EU frontier.

What the Token Share Shows

The headline figure points to a real change in how much computation OpenAI’s business customers direct to agent products. Agents can handle multistep work—such as reading files, using tools, drafting or coding, and checking results—and may generate much more text during a run than a standard chat response. That makes output tokens useful for describing system activity, but a limited proxy for workplace adoption or results.

Tokens are not tasks or value. OpenAI’s Enterprise Signals page describes tokens as an imperfect measure of business value and says agentic tasks can require substantially more computation. A 55% token share does not establish that agents perform 55% of work or account for 55% of AI interactions. The source provides no conversion from output tokens to tasks, hours saved, quality or financial return.

The timing also matters. ChatGPT Work launched in July and made up 21 percentage points of the September token total. The report says a major part of small-business user growth followed its introduction. The figures therefore capture both interest in agents and the rollout of a new OpenAI product; they cannot, on their own, establish a broad economic shift across European small businesses.

How OpenAI Defines EU Agent Use

The figures come from OpenAI’s own products and customers, rather than a survey of all companies operating in Europe. The report’s scope is limited to ChatGPT Enterprise and Business accounts in the EU. It does not measure use of rival AI services, internally developed systems or AI tools outside those accounts. The results are thus a vendor’s account of activity within its customer base, not a census of European business AI use.

OpenAI’s global Enterprise Signals material provides a comparison for its EU report. The report defines “frontier firms” by output tokens per active user, rather than revenue, productivity or other business outcomes. OpenAI says EU companies account for about 10% of frontier firms worldwide, compared with about 65% in North America. Those figures describe the distribution of OpenAI-defined high-use firms, not the share of all companies or the relative performance of entire regional economies.

OpenAI’s stated policy conclusion is that Europe should move from broad adoption toward deeper use by building workers’ skills and confidence and making advanced capabilities more accessible and affordable. That is the company’s interpretation of its data. The report does not establish that any particular policy would close the reported usage gap.

What the Data Cannot Establish

The report material does not give country-level account or user sample sizes, making it hard to judge how representative the national percentages are. OpenAI notes that smaller markets including Cyprus, Estonia and Lithuania have relatively high agent use within larger firms; without sample counts, a small number of large customers could affect a country’s result. The percentages should not be read as rankings of national AI readiness.

The reported 8.1-times growth has no absolute user baseline in the supplied material, and its comparison begins in early May, before ChatGPT Work launched. The report says the launch contributed substantially to growth, but does not quantify how much. Nor does it report how many agent runs correspond to a given volume of tokens, what work those runs completed, or whether customers achieved measurable productivity gains. These gaps leave the breadth, durability and business impact of the adoption trend uncertain.

Evidence Needed Beyond Token Counts

The October report provides a snapshot through September 30 for token shares and through the week ending October 5 for small-business user growth. OpenAI’s supplied material does not specify a follow-up publication date or a next measurement milestone. A later period of data could show whether use continues after the initial rollout of ChatGPT Work, but it would still need comparable definitions and baselines to establish a durable trend.

For a clearer assessment, readers would need absolute account and user counts, country-level sample sizes, task-level measures and results linked to business outcomes. Comparisons across providers and companies using non-OpenAI systems would also be needed to describe the wider EU market. Until those details are available, the confirmed finding is narrower: agents account for most of the output tokens in the covered OpenAI business accounts, while broader claims about European adoption and productivity remain unproven by this report alone.

Key Questions

What does OpenAI mean by saying agents produced 55% of EU output?

It means Codex and ChatGPT Work generated 55% of output tokens in the covered ChatGPT Enterprise and Business accounts by September 30, 2026. It does not mean agents handled 55% of tasks or users’ AI interactions.

Which products count as agents in the report?

OpenAI counts Codex, its coding agent, and ChatGPT Work, an agent mode launched in July 2026. The remaining 45% of reported output tokens came from standard ChatGPT.

Did small-business agent use increase 8.1 times across Europe?

OpenAI reports an 8.1-times increase in weekly active users of Codex or ChatGPT Work at businesses with 200 or fewer users, comparing the first week of May with the week ending October 5. The source material does not give the absolute user counts, and says growth followed the launch of ChatGPT Work in part.

Does the report show that agents improved productivity?

No. The report describes product use and token output, but the supplied material does not quantify time saved, work quality, productivity gains or financial returns. OpenAI itself cautions that tokens are an imperfect measure of business value.

Do the country figures represent all businesses in each EU country?

No. They cover OpenAI’s ChatGPT Enterprise and Business accounts, not every business or AI service in each country. The report material also does not provide country-level sample sizes, so the figures are not a complete measure of national adoption.

Source: ThorstenMeyerAI.com