OpenAI’s EU Business AI Headline: Why The Denominator Is Essential

🔍 Read the full analysis: OpenAI’s EU Business AI Headline: Why The Denominator Is Essential on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets delivered free with Prime

• Fast, free delivery on millions of items
• Prime Video, Amazon Music and more included
• Member-only deals all year

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

OpenAI reported on October 14 that its AI agents generated 55% of output tokens in the company’s ChatGPT Enterprise and Business accounts in the 27 EU member states by September 30. The figure highlights how much of those accounts’ generated text came from Codex and ChatGPT Work, but it does not show that agents handled most users’ tasks or created most of their business value.

OpenAI Economic Research’s report, Europe’s Agentic Opportunity: Moving from Breadth to Depth with AI Agents, counts output from Codex, OpenAI’s coding agent, and ChatGPT Work, an agent mode introduced in July 2026. Codex made up 34% of output tokens and ChatGPT Work 21%; ordinary ChatGPT accounted for the remaining 45%. OpenAI’s Global Affairs team published a related article, while the company’s Enterprise Signals page provides worldwide comparisons.

The report also says that among businesses with 200 or fewer users, weekly active users of Codex or ChatGPT Work increased 8.1 times between the first week of May and the week ending October 5. OpenAI says each member state recorded at least fivefold growth. Belgium, Italy and Luxembourg had the largest reported increases, at 12.1, 10.8 and 10.0 times, respectively. The company does not provide the absolute number of users behind those growth rates.

Agentic output shares differed across countries, from 85% in Lithuania and 81% in Cyprus to 22% in Slovenia and 26% in Austria. Germany’s share was 51%, ranking 17th among the 27 states. OpenAI defines “frontier firms” as the top 10% of firms by output tokens per active user. In September, EU frontier firms produced about four times as many tokens per user as typical EU firms; frontier firms outside the EU produced more than twice as many as the EU’s frontier group.

What the Token Share Shows

The report offers a view into how OpenAI’s business customers use its products, and its figures suggest agents are taking a growing share of model-generated text in those accounts. That can matter to companies planning staff training, software budgets and workplace processes. But the headline percentage is not a measure of how much work has been delegated or how much productivity has improved.

OpenAI describes tokens as an imperfect measure of business value and says agentic tasks can require substantially more computation. An agent may read files, plan, generate code or documents, check its work and revise it, producing more tokens than a short chat reply. The denominator matters: a share of output tokens answers how much text was generated by each product category, not how many interactions or tasks involved agents. Without task counts or outcome measures, the 55% figure cannot establish that most AI use—or most value—in these accounts is agentic.

The distinction is relevant to policymakers and business leaders weighing claims about Europe’s AI adoption. The report’s stated policy direction is to move from broader adoption to deeper use, including skills, confidence and access to advanced tools. Its data can inform that discussion, but the reported usage metrics alone do not show whether European firms are improving productivity or closing a competitiveness gap.

How OpenAI Defines Agent Use

The report covers only ChatGPT Enterprise and ChatGPT Business accounts in EU member states, rather than all European companies or all AI services. Its definition of agentic use includes Codex and ChatGPT Work. The latter launched in July 2026, after the May starting point used for the small-business growth comparison.

That timing affects interpretation. OpenAI’s chart shows Codex at roughly a fifth of EU output tokens in May and 34% by September 30, while ChatGPT Work contributed 21 percentage points after its launch. The source article estimates that Work accounts for more than half of the roughly 35-point increase in agentic share. OpenAI also says a major part of small-business user growth followed the Work launch. The figures therefore reflect both adoption and the arrival of a new product, not just a gradual shift in behavior.

OpenAI’s report frames Europe’s opportunity as turning broader adoption into deeper use. Its comparison of frontier firms is based on tokens per active user, not independently measured productivity. The report says EU companies account for about 10% of frontier firms worldwide, compared with about 65% in North America; those figures describe the composition of OpenAI’s identified frontier group, not the full business population in either region.

What the EU Figures Leave Out

The report does not provide account or user sample sizes by country, making it hard to judge how representative national percentages are. OpenAI notes that smaller countries including Cyprus, Estonia and Lithuania have relatively high agentic use within larger firms. A small number of large customers could influence a country’s result, but the published material does not establish whether that happened.

The source article’s comparison of country rankings finds little relationship between agentic share and small-business growth, but that is an analysis of OpenAI’s reported figures, not a conclusion established by the report. The country measures may reflect different customer mixes and starting points. They should not be treated as direct rankings of national AI readiness.

Other key measures are absent: the report gives no absolute small-business user counts, no count of agentic tasks, and no direct productivity or financial outcomes. The small-business comparison also begins before ChatGPT Work launched, and OpenAI attributes a major part of subsequent growth to that product. How much growth reflects a lasting change in work practices, rather than product availability and a low starting base, remains unclear.

What Future Adoption Data Must Show

OpenAI’s report sets out a direction for future work—building skills and confidence and making advanced capabilities more accessible and affordable—but the source material does not announce a specific next release date or a new measurement plan. The next useful evidence would include absolute user and account counts, country-level sample sizes, and time series that distinguish existing products from newly launched ones.

To assess the business impact, readers would also need measures beyond tokens: how many tasks agents complete, how often people review or revise the results, and whether use changes time, costs or output quality. Until those measures are available, the report is best read as a snapshot of activity within OpenAI’s own EU business accounts, not a census of European companies or proof of economy-wide productivity gains.

Key Questions

What does OpenAI mean by 55% of output?

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

Which tools count as agents in the report?

OpenAI counts Codex, its coding agent, and ChatGPT Work, an agent mode launched in July 2026. The report compares their output with ordinary ChatGPT.

Did small-business agent use increase 8.1 times?

OpenAI reports an 8.1-fold rise 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 October 5. It does not give the absolute user count, and says much of the growth followed ChatGPT Work’s launch.

Does the report prove that AI agents raised productivity?

No. The report measures product use and output tokens, not independently verified productivity, cost savings or business outcomes. Those effects remain unestablished by the figures provided.

Source: ThorstenMeyerAI.com