Future AI Pulse
Quick answer: OpenAI’s EU report describes growing use of Codex and ChatGPT Work within its customer sample. Token share measures activity in that sample, not the share of all European work performed by agents. Read the population, unit and definition before interpreting an adoption headline.
What the report says
In its October 14, 2026 release, OpenAI says Codex and ChatGPT Work account for more than half of output tokens in its EU ChatGPT Enterprise sample. It also reports more than eightfold growth since May in weekly active users of those tools among EU small and medium businesses.
The analysis defines those businesses as having 200 or fewer associated users. It defines frontier firms as the top 10% by output tokens per active user, with that sample limited to Enterprise and Business accounts. These are OpenAI’s definitions and platform measurements, not a census of every EU company or an independent productivity study.
Keep the numerator and denominator together
A percentage needs the group it describes. “More than half of output tokens” is different from “more than half of employees,” “companies” or “tasks.” Each statement would require a different measurement. Do not switch between them when summarizing a chart.
Growth also needs a starting point and a period. An eightfold increase can describe a change in a particular tool’s active users without establishing its share of the entire workforce. Preserve the original unit so readers can understand the result without guessing what grew.
A usage measure is not an outcome measure
More activity can reflect useful work, longer outputs, repeated attempts or a change in how people use a tool. A token count alone does not distinguish those possibilities. Measuring a business outcome requires another record, such as accepted work, error rates, elapsed time or customer impact.
This does not make usage data unhelpful. It makes it useful for a narrower question: where activity is occurring and how it changes under the report’s definitions. Treat a proposed benefit as a hypothesis unless the evidence actually measures it.
Read comparisons with the same definitions
Before comparing countries or groups, check that the accounts, period and metric align. A group selected for high usage will naturally differ from the typical user on a usage measure. That selection should remain visible when someone describes it as a frontier.
For a summary, distinguish the reported observation from the explanation you think might account for it. The data may show a gap without identifying which policy, industry mix or workplace practice caused it.
Summarize this adoption report using only its stated evidence. For each key figure, give the population, unit and time period. Preserve the report’s definitions of business size and frontier firms. Separate measured usage from proposed productivity benefits. List what the sample does not establish about the wider economy. Do not convert token share into a share of workers, firms or completed tasks.
Build a useful internal comparison
If your organization wants to learn from an adoption report, pair usage with the result of the work. Define a repeatable task and its acceptance criteria. Then record whether increased use changes completion, corrections and time to an accepted result.
A modest pilot with clear measurements can be more informative than setting a token-volume target. Keep failed attempts in the record and make changes to the task definition explicit, so later comparisons remain meaningful.
Our cost-per-deliverable guide offers a related outcome measure, and the metric-definition guide explains why precise units matter.
Illustrative workplace photograph: Yen Vu / Unsplash, under the Unsplash License. Not a depiction of the report’s sample.
Sources
Source checked October 15, 2026. Written with AI assistance. Interpretation and reading prompts are editorial analysis; no independent data audit is claimed.