For one month, people asked ChatGPT to look things up slightly more often than they asked it to write. In August, information seeking made up 20.7% of consumer messages and writing made up 20.6%, per the Signals dataset OpenAI published on Oct. 26. Search Engine Journal covered the release on Oct. 27. Writing regained the top spot in September, at 21.4% versus 20.2% for information seeking.
The long-run direction matters more than the crossover. Writing dropped from 34.1% in July 2024 to 21.4% in September 2026, while information seeking climbed from 16.7% to 20.2% over the same span. Practical guidance, a bucket for tutoring and how-to help, has been the top topic monthly from March 2025 onward and reached 28.1% in September. The largest single use of ChatGPT is not lookup at all.
Know what the sample is before you quote it. OpenAI draws 1M messages each month from adults on its consumer tiers (Free, Go, Plus, and Pro), and it leaves out Enterprise and Codex. An automated system then assigns each message to a subject. Every figure is OpenAI’s own data, sorted by OpenAI’s own classifier, with no outside audit attached.
The “information seeking” label is wide. Per the release’s data dictionary, it combines factual questions, product and service shopping questions, and cooking help. Because OpenAI reports one merged number, nobody can tell how much of the 20.7% is commercial intent. That split is the part a search team would most want.
Work use shows a sharper shift. Work-related messages were 32.2% of the September total. Inside that group, the information-seeking share held between 10.7% and 12.1% monthly, from July 2024 until July 2026, then rose to 13.9% in August and 15.2% in September. A companion post sorts work messages into 165 O*NET activities, and its U.S. figure for “Getting information” slipped from 11.6% in Q1 2025 to 9.9% in Q3 2026. That series uses a separate classification and covers one country, so it cannot be set beside the global seeking-information line.
The new data also sits below an earlier benchmark. A 2025 paper from OpenAI and Harvard researchers reported information seeking rising from 14% to 24% of usage during the twelve months following July 2024. Signals runs the same classifier prompts, yet it shows 16.7% for July 2024 and 18.3% for June 2025, the paper’s last month. The writing series diverge less: a drop of 12 points (36% to 24%) in the paper, and 8.4 points (34.1% to 25.7%) in Signals.
Sampling differs between the two. The researchers took one message from each of roughly 1.1 million conversations and averaged shares over 28-day windows. Signals takes 1 million messages monthly and restated earlier months to match. Neither the release blog nor the data dictionary explains why the information-seeking numbers disagree by almost six points for mid-2025.
For discovery, the larger gap is what Signals leaves out. It tracks topics but does not say how often ChatGPT consulted outside sources or cited a website in a reply. A lookup message shows demand for answers; it does not show a click, a citation, or a visit to any publisher. Treating 20.7% as a traffic estimate would be a mistake.
The practical reading is narrow. Teams weighing ChatGPT as a channel can say that roughly one in five consumer messages is a lookup of some kind, and that the share is trending up from 16.7%. They should check that against their own referral logs and citation tracking before shifting budget. Any report that quotes a Signals figure should name the month and the V3.0 version, since OpenAI’s changelog shows each update revising earlier months.
Based on reporting by Search Engine Journal, published Oct. 27, 2026, on OpenAI Signals data released Oct. 26, 2026.
