What can one million sampled ChatGPT messages a month reveal about work?
OpenAI's new dataset maps consumer messages to 165 work activities across nearly 100 countries. It shows a broader mix over time, but measures classified messages—not workers, jobs, productivity or outcomes.
The short version
- What happened
- Signals v3.0 classifies a global sample of 1,000,000 consumer ChatGPT messages in each month from July 2024 through September 2026, from accounts with a declared age of at least 18.
- Who it affects
- Workers and educators may benefit from more transparent evidence about actual AI use, but policy decisions still need representative surveys and measured outcomes beyond one platform's prompts.
- What remains unproven
- Provider-specific consumer sample
Keep in mind
Provider-specific consumer sample
Excludes Enterprise and Codex
Messages are not unique users or tasks
Automated classifications can vary by language and context
Differential privacy adds noise
No productivity, wage, quality or employment outcome
Source analysis · 3 linked sources. Primary analysed source date: 26 October 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 4 min The evidence, methods, implications and limitations
The dataset opens a useful window—not a labour census
OpenAI has released quarterly data linking work-related consumer ChatGPT messages to 165 O*NET intermediate work activities across nearly 100 countries. The methods document says every monthly global sample contains 1,000,000 messages from accounts with a declared age of at least 18, covering July 2024 to September 2026. Enterprise messages and Codex use are excluded, as are deleted accounts and people who did not allow training use.
That denominator makes Signals unusually large and repeatable, but the unit is a message, not a person or completed task. One worker can send many prompts; another may never use ChatGPT or may use a competing tool. The sample cannot estimate the share of workers using AI, jobs changed, time saved, pay, output quality or displacement. It describes what selected consumer traffic looked like after classification.[1][2][3]
Work prompts became less concentrated in the top activities
Among 74 countries observed in every quarter, the ten most common activities accounted for an average 51.1% of each country's classified work messages in the third quarter of 2026, down from 56.0% in the first quarter of 2025. OpenAI interprets the spread as countries becoming more specialized in the work they bring to ChatGPT. Different occupational mixes or differences in access could also produce the pattern.
For teachers, analysts and managers, the dataset can support better questions: which activities appear often, which rise or fall, and where a local survey or controlled evaluation is needed. It cannot tell whether a message helped. O*NET is a US occupational framework applied to global traffic, so a label may travel imperfectly across languages, informal work, public services and occupations organized differently elsewhere.[1]
The distinction that matters
From private messages to public statistics
Each stage protects privacy or enables comparison, while also narrowing what the numbers can mean.
Sample
1m eligible consumer messages each month
Large provider-specific traffic view
Classify
Automated work and O*NET labels
Comparable activity shares with label error
Protect
Withholding, weighting and privacy noise
Safer aggregates with added uncertainty
Privacy protection and automated labels both shape the result
Researchers did not read raw message text. OpenAI says automated language-model classifiers operated on de-identified messages after personal information was removed. Daily weights adjust for changes in logging, small groups are withheld, and Gaussian noise provides message-level differential privacy before aggregated statistics are published. Fine-grained country-and-activity cells therefore have more sampling and privacy variation than global totals.
The categories—including work-relatedness, asking, doing and expressing—are inferred labels rather than user-reported facts. OpenAI produced the dataset from its own service and has a commercial interest in showing breadth of use. Transparent classifier validation by language, uncertainty intervals and outside replication would strengthen it. For now, Signals is a valuable public research input, provided analysts resist turning message shares into claims about worker benefit or economic transformation.[2][3]
Watch the background, then check the evidence
Illustrative photo · not a video frame
UNESCO · 11 min 27 sec · Background explainer
UNESCO Talks: Interview with 2024 Nobel laureate Professor Daron Acemoglu
YouTube loads only when you choose; playback starts when you press play.
Economist Daron Acemoglu discusses how choices about AI affect workers, inequality and society. Pair the interview with our comparison of task exposure, hiring and skills evidence.
Evidence check: An expert interview from 2024, offering a perspective on policy and work rather than a measured forecast of job losses.
Published 2024-10-14. Source checked 14 October 2026. Check captions on the publisher’s video page.
Read the AI and jobs guidePhoto: Dylan Gillis / Unsplash · Licence
Continue the story
Read next
Continue exploring work: compare the finding and its evidence limits.
Do agentic tools now make up most business AI use in the EU?
Within OpenAI's Enterprise and Business customer sample, yes: Codex and ChatGPT Work produced 55% of included output tokens at the end of September. That is not a measure of all EU firms, completed work or productivity.
Read the next reportExplore the full source and evidence record
What this means for people
- Workers and educators may benefit from more transparent evidence about actual AI use, but policy decisions still need representative surveys and measured outcomes beyond one platform's prompts.
Global context
Nearly 100 countries appear, but 74 were present in every quarter and the work taxonomy comes from US O*NET. Uneven access, language coverage and labour-market structure limit country comparisons.
What the evidence does not yet show
- Provider-specific consumer sample
- Excludes Enterprise and Codex
- Messages are not unique users or tasks
- Automated classifications can vary by language and context
- Differential privacy adds noise
- No productivity, wage, quality or employment outcome
What to watch next
- Classifier accuracy by language and country
- Uncertainty intervals for fine-grained cells
- Independent linkage to representative labour statistics
- Outcome studies on quality, time, pay and worker control
Living evidence record
Impact record IAI-09CEKY6
Evidence stage
Observed
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Not yet
Record status
Monitoring
Last checked
27 October 2026
Source trail
3 direct sources across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Related-source reporting disclosure
This record analyses 3 linked source records around the same underlying development. The extra records add method, date or context, but they do not by themselves constitute independent replication of every performance claim or predicted outcome.
Evidence trail
Sources used for this report
Links checked 27 October 2026
Research question, topics and regions
What OpenAI Signals v3.0 can and cannot show about the work activities people bring to consumer ChatGPT
Topics: Work · Labour data · ChatGPT · Privacy · Skills
Regions: Global · United States
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.