Information seeking caught up with writing on ChatGPT at 20.7 percent

OpenAI published a figure yesterday that every search team should carry into next week's plan. The source is OpenAI's own Signals v3.0 data, dated 26 October. Matt Southern at Search Engine Journal walked through what it says. The headline. In August 2026, information seeking reached 20.7 percent of consumer ChatGPT messages. Writing was 20.6 percent. It is the first time seeking has been higher.

The three figures to carry forward

The series starts in July 2024. Writing was 34.1 percent of consumer ChatGPT messages that month. By September 2026, writing was 21.4 percent. Seeking information rose from 16.7 percent to 20.2 percent over the same window.

Writing moved back ahead in September 2026, at 21.4 percent against 20.2 percent for seeking. Southern notes that August is the only month in the series so far where seeking was the larger share.

Practical guidance has been the largest category every month since March 2025. The group includes how-to advice and tutoring. It was 28.1 percent in September 2026. So the three biggest groups now share roughly two in three consumer messages between them.

How the data is built, and where it disagrees with earlier numbers

OpenAI selects one million messages a month from Free, Go, Plus and Pro consumer plans, age 18 and up. It assigns each one to a topic with a classifier model, then averages. Signals is the public version of the pipeline. OpenAI applied the sampling retroactively, so earlier months in the release reflect the same million-per-month baseline.

A 2025 OpenAI and Harvard research paper reported a higher share for information seeking, around 24 percent by mid-2025. The Signals release puts the same months closer to 18 percent. The two series use the same classifier prompts and the same category definitions, per the data dictionary. OpenAI does not explain the gap in the release blog.

Southern's note to anyone citing the figure: name the source and the time frame. The Signals figure for a given month can also change when a new release re-processes earlier months.

Why a search team should care

A share of consumer ChatGPT messages reading like a search query is now one in five, on OpenAI's own count. The series is still rising on the same count. A page that depends on classical Google clicks now has to earn visits a second way. It has to be quoted inside an AI answer.

Signals does not report how often ChatGPT reaches an outside site or cites one in a reply. That number is the one a GEO team actually needs. Until OpenAI releases it, the serve-side proof is a traffic log on the publisher side. A web team should be able to answer two questions each month. What is the Google share of referrals, and what is the ChatGPT share.

The second-source test matters more than the single figure. OpenAI's own paper, the Signals release and the Reuters Institute each measure something close but not identical. The honest reading has three parts. One or two in ten consumer ChatGPT messages are a search. The share is rising. The exact fraction will keep moving.

Source

ChatGPT Data: 1 In 5 Consumer Messages Now Seek Information, Search Engine Journal, by Matt G. Southern. OpenAI Signals data, OpenAI.