ChatGPT Users Now Ask Questions as Often as They Write

Information seeking accounted for 20.7% of messages on consumer ChatGPT plans in August, nearly matching writing at 20.6%, according to OpenAI Signals data released Oct. 26. It was the first month in the series, which starts in July 2024, when information seeking ranked higher. Writing moved back in front in September, at 21.4% versus 20.2%.

The finding carries weight for marketers. Roughly one in five consumer messages to ChatGPT is now a lookup request, and that category includes questions about products and services for sale.

What does the new Signals release show?

OpenAI published Signals V3.0 alongside a blog post on how users bring work tasks to ChatGPT. The dataset draws on a monthly sample of 1 million messages from account holders aged 18 and over on Free, Go, Plus or Pro plans. Enterprise and Codex messages are excluded. An automated system classifies each sampled message by subject area.

The trend lines run in opposite directions. Writing fell from 34.1% of messages in July 2024 to 21.4% in September. Seeking information rose from 16.7% to 20.2% over the same period.

Practical guidance, a category covering how-to advice and tutoring, has been the largest topic every month since March 2025. It reached 28.1% in September. According to the release's data dictionary, seeking information covers requests for facts, questions about products and services for sale, and recipe or cooking help.

OpenAI combined those three request types into a single category. That decision means no one can determine how many messages were product questions versus factual lookups or cooking help.

How is ChatGPT used for work?

Work-related messages made up 32.2% of all messages in September. Within that subset, information seeking stayed between 10.7% and 12.1% each month from July 2024 through July 2026. It then rose to 13.9% in August and 15.2% in September.

The work activity post organizes work messages into 165 activities from O*NET, a standard framework that groups tasks shared across jobs. It presents quarterly data from Q1 2025 through Q3 2026 across nearly 100 countries.

Across the 74 countries with data in every quarter, the 10 most common activities averaged 51.1% of each country's classified work messages in Q3 2026, down from 56.0% in Q1 2025. OpenAI describes that decline as work-related use becoming less concentrated in the most common activities.

The graph for "Getting information," one of the 165 activities, shows a U.S. share of 11.6% of work messages in Q1 2025, falling to 9.9% in Q3 2026. These figures use a separate classification system and cover one country, so they do not measure the same thing as global information seeking.

Why does the Harvard paper disagree?

A 2025 paper by OpenAI and Harvard researchers reported that information seeking grew from 14% to 24% of ChatGPT usage in the year after July 2024. Signals shows a smaller rise.

The two projects use the exact same classifier prompts and the same three request categories, according to the data dictionary. Yet in June 2025, the final month the paper covered, Signals puts information seeking at 18.3%, nearly six points below the paper's 24%.

The writing series diverges less. The paper shows a 12-point drop, from 36% to 24%. Signals shows an 8.4-point drop, from 34.1% to 25.7%, over the same period.

Sampling differs between the projects. The researchers selected about 1.1 million conversations, taking one message from each, and averaged shares over 28-day periods. Signals selects 1 million messages per month, and this release revised earlier months upward to match that number. Neither the blog post nor the data dictionary explains the gap in information-seeking figures.

Signals tracks conversation topics but says nothing about how often ChatGPT accessed external sources or cited a website in its replies. For anyone citing these figures, the source, time frame and version number matter. Each Signals update adds a quarter of data and revises earlier months, so a figure from V3.0 may not match the same month in a later version.

Marketers tracking AI-driven discovery should therefore treat Signals as a leading indicator of query behavior, not a census of traffic referrals — at least until OpenAI publishes citation data alongside topic shares.