Data looking for accounted for 20.7% of messages on shopper ChatGPT plans in August, practically matching writing at 20.6%, based on OpenAI Alerts knowledge launched Oct. 26. It was the primary month within the collection, which begins in July 2024, when the reported share for looking for info was larger. Writing moved again in entrance in September, at 21.4% versus 20.2%.
Along with publishing the brand new model of their alerts (Alerts V3.0), OpenAI posted a weblog submit about how customers are bringing work to ChatGPT. Their knowledge relies on a month-to-month pattern of 1M messages from account holders with a declared age of 18+ on Free, Go, Plus, or Professional plans. They exclude Enterprise and Codex. As well as, they’ve created an automatic system that classifies every sampled message by topic space.
Over the months each cowl, Alerts additionally reveals a smaller rise in looking for info than a 2025 paper by OpenAI and Harvard researchers, which put the class at 24% of ChatGPT utilization by mid-2025.
Looking for Data Rose Whereas Writing Fell
Writing fell from 34.1% of messages in July 2024 to 21.4% in September 2026. Looking for info rose from 16.7% to twenty.2% over the identical months.
Sensible steering, which incorporates how-to recommendation and tutoring, has been the most important matter each month since March 2025 and was 28.1% in September.
In line with the discharge’s knowledge dictionary, looking for info covers requests for details, questions on services on the market, and recipe or cooking assist.
Work-related messages made up 32.2% of all messages in September. Amongst these, the share of messages looking for info remained between 10.7% and 12.1% every month from July 2024 via July 2026, earlier than rising to 13.9% in August and 15.2% in September.
What The Work Exercise Put up Provides
The work exercise submit organizes work-related messages into 165 actions from O*NET, a typical framework that teams duties shared throughout jobs. It presents the information by quarter, masking Q1 2025 via Q3 2026, throughout practically 100 international locations.
Throughout the 74 international locations with knowledge in each quarter, the ten most typical actions averaged 51.1% of every nation’s categorised work messages in Q3 2026, down from 56.0% in Q1 2025. The submit describes that as work-related use changing into much less concentrated in the most typical actions.
The submit’s graph reveals “Getting info,” which is one in all these 165 actions. In america, it represented 11.6% of labor messages throughout the first quarter of 2025. It additionally confirmed that it made up 9.9% of labor messages within the third quarter of 2026. These are primarily based on a separate classification system and characterize just one nation (the U.S.), so they don’t measure the identical factor as “looking for info” globally.
How It Compares With The 2025 Paper
Final September, I coated the OpenAI and Harvard paper on how individuals use ChatGPT. It reported that looking for info grew from 14% to 24% of ChatGPT utilization within the 12 months after July 2024.
Alerts makes use of the very same classifier prompts utilized by the analysis examine, based on the supplied knowledge dictionary, and each categorize the identical three classes of requests underneath “looking for info.” Alerts’ graphs additionally present the odds of the identical timeframe for looking for info; 16.7% in July 2024 and 18.3% in June 2025, which is the ultimate month coated by the analysis paper.
The 2 writing collection have fewer variations between them than the 2 seeking-information collection. The share decreased by 12 factors (from 36% to 24%) within the analysis paper whereas it decreased by 8.4 factors (from 34.1% to 25.7%) in Alerts throughout the identical timeframe.
Roughly 1.1 million conversations have been chosen by the researchers utilizing the sampling methodology the place they took just one message from every dialog. Every share was then averaged over 28-day intervals. Alerts selects 1 million messages monthly. This launch altered earlier months to be in line with that larger quantity. No motive was acknowledged or defined inside the launch weblog or the supplied knowledge dictionary concerning the distinction within the looking for info percentages.
Why This Issues
In OpenAI’s personal month-to-month knowledge on shopper ChatGPT messages, roughly twenty % of all messages are to look one thing up, which is roughly equal to the share for “writing.” Product questions are included on this share, nonetheless as a result of OpenAI mixed all three classes (truth/recipe/product) into one class, it can’t be decided what number of have been every.
Alerts does observe the varied matters of dialog that happen by way of ChatGPT; nonetheless it didn’t present any info concerning how typically ChatGPT accessed exterior sources corresponding to web sites or cited a website in its replies.
When citing percentages representing the usage of ChatGPT, please reference the supply and timeframe. There was an virtually six level distinction between the 2025 paper and Alerts concerning info looking for in mid-2025.
Trying Forward
To this point, every replace within the Alerts knowledge dictionary’s changelog provides 1 / 4 of information and revises earlier months. As a result of a determine from v3.0 could differ from the identical month’s determine in a later model, be sure you observe the model quantity everytime you embody a Alerts determine in a report.
Featured Picture: Tetiana Yurchenko/Shutterstock
