Activities, occupations, and movers

United States, Q3 2026: edit written materials or documents, create visual designs or displays, and develop marketing or promotional materials make up 29.8% of work-related messages.

Data as of Q3 2026 · OpenAI Signals v3.0 · O*NET 30.1

Back to the map

Mental processes · Making Decisions and Solving Problems

Edit written materials or documents

Share of work messages
17.1%
Rank
1 of 164
Since Q1 2025
−5.0 points

Down overall, but bumpy

22.2%17.1%

Around the world, Q3 2026

Loading the other countries…

Countries with fewer messages move around more, so read the pile, not any single dot.

Occupations whose tasks include this

Loading the occupations…

Reading the numbers

  • A share is one country’s work-related ChatGPT messages in one quarter, so the activities add up to 100%. A rising share grew relative to the others; it doesn’t mean more messages.
  • Consumer ChatGPT only: no enterprise accounts and no Codex, so business use, and coding in particular, is probably understated. OpenAI adds OpenAI adds small random amounts to the counts behind every number, a method called differential privacy, so it’s hard to tell whether any one message is in the data. Big shares barely move. Tiny shares, and countries with few messages, wobble more. and leaves out a country’s quarter when it has too few messages, so small shares and small countries wobble.
  • OpenAI reports O*NET work activities, not occupations, and publishes no worldwide figure. The occupation links come from O*NET’s task tables. Overlap is Wavelength’s own measure, not OpenAI’s: the share of work messages in the activities an occupation’s tasks list. It does not measure automation, exposure, or job risk.
  • OpenAI groups activities too rare to publish into “Other IWA” (2.4% of U.S. work messages in Q3 2026). It is in the table and the downloads, not drawn.

Sources and license

OpenAI Signals v3.0 (another site) (Chatterji, Cunningham, Deming, Hitzig, Johnston, Richmond, Ong, Shan, and Wadman, 2026), CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/ (another site)). Wavelength reshaped this data and combined it with O*NET. OpenAI does not endorse this page or how the data is used here.

This page includes information from the O*NET 30.1 Database (another site) by the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA), used under the CC BY 4.0 license (another site). O*NET® is a trademark of USDOL/ETA. Wavelength has modified and combined this information, and USDOL/ETA has not approved, endorsed, or tested these modifications.

Cite the OpenAI data
@online{openaiecon2026signalsdata,
  author = {Aaron Chatterji and Thomas Cunningham and David J. Deming and Zoe Hitzig and Drew Johnston and Alex Martin Richmond and Christopher Ong and Carl Yan Shan and Kevin Wadman},
  title = {{OpenAI} Signals v3.0},
  year = {2026},
  organization = {OpenAI},
  url = {https://openai.com/signals/data-download/},
  urldate = {2026-10-26},
  note = {Data download and methodology},
}

The Signal Newsletter

What changed for workforce programs, and what it means for yours.

A short weekly read, every Tuesday, for community college workforce and continuing education leaders: labor markets, funding, credentials, and program launches, with a practical take on each.

Free. No spam. Unsubscribe any time.