On 26 October OpenAI published the third release of Signals, its data on how people use ChatGPT, and it covers a million classified messages a month from July 2024 to September 2026. Within a day, everyone was sharing the drop in writing, from 34% of messages to 21%. Seeking information, OpenAI’s label for search, rose to 20% in the same period.
Writing fell and search caught up with it
Share of consumer ChatGPT messages, by month, July 2024 to September 2026.
Scroll sideways for the whole chart.
OpenAI Signals v3.0. Seeking information is OpenAI's label; it holds product questions, facts and recipes together.
I spent a day inside the files to see who is doing the searching and what they search for, and the surprise for me was the age split.
What the data counts
- It counts messages, not people. Every figure is a share of messages, so when a share rises, more of the conversation is about that thing. It says nothing about how many people or purchases sit behind it.
- It’s consumer ChatGPT only: Free, Go, Plus and Pro accounts, adults only. Business accounts aren’t in it, so anything about work is understated.
- OpenAI puts “which laptop should I buy” under the same label as “what’s the capital of Peru” and “how do I cook lentils”. So the figures here are about search in general, and the shopping part is inside them where nobody outside OpenAI can pull it out.
- OpenAI revises earlier months with each release, and July to September 2026 is out for the first time, so it will shift a little next quarter.
Everything else comes straight from the files, and I did the arithmetic myself. The data is published under a Creative Commons licence, so you can download it and check.
1. Who is searching
You might think “Gen Z searches in ChatGPT. Everyone else still googles.” but OpenAI’s own data says the opposite.
I took the 2+ years of ChatGPT data OpenAI published and checked, per age group, what share of their messages are searches?
Every age group from 25 up searches more than in 2024
Share of each age group's messages that are searches.
18 to 24
25 to 34
35 to 44
45 to 54
55 to 64
65 and over
OpenAI Signals v3.0, derived from the age by topic, topic and age tables. Adults, consumer plans, global.
People aged 55 to 64 now use ChatGPT as a search engine in more than a quarter of their messages.
I also checked whether this was just older people joining, but it wasn’t. If only the age mix had changed, search would still be where it started. Keep the July 2024 mix and change only how people use it, and search comes to 19.6% of messages, against 20.2% today. People over 35 changed what they use the assistant for.
These people hold the household budget, sign the PO and approve the vendor.
It’s the same picture in the US, the UK and Germany, and in most of the countries with enough data.
Share of each age group's messages that are searches, July to September 2026.
United States
United Kingdom
Germany
OpenAI Signals v3.0, derived from the age by topic, topic and age tables by country.
In July 2024, 61% of search messages came from accounts with a typically masculine first name, and by September 2026 that was down to 53%.
2. At work, people are researching vendors in the chat
OpenAI tags every work message in ChatGPT with the job activity it serves. The activities come from the list the US Department of Labor uses to describe jobs, and most work use is what you’d expect, like editing documents and writing marketing copy. Three of those activities describe a buyer. In the US, between early 2025 and the third quarter of 2026:
- Explaining how a product works: it’s down by 50%
- Finding information about products or services: it’s up 73%, from the 11th most common work activity up to the 5th.
- Working out what something should cost: more than 2x
At work, the buying questions moved to which one and at what price
Three buying activities in US work messages, as an index. Q1 2025 = 100.
Scroll sideways for the whole chart.
OpenAI Signals v3.0, work activities file (O*NET: obtain information about goods or services; determine values or prices of goods or services; explain technical details of products or services). United States, consumer plans only. Shares of US work messages from Q1 2025 to Q3 2026, in the same order: 1.7% to 2.9%, 0.3% to 0.7%, 1.2% to 0.6%.
Finding information about products went up in 70 of the 74 countries with data for every quarter, and most of Europe is ahead of the US.
Finding information about goods or services, share of work messages, July to September 2026.
United States
2.9%United Kingdom
3.2%Germany
3.3%France
3.7%Portugal
3.5%OpenAI Signals v3.0, work activities file. Consumer plans only.
Working out the price went up in 69 of those 74 countries too. Across all of them together, explaining how a product works fell by about half.
Fewer people now ask the AI assistant how your product works, more ask it which product to buy and what it should cost.
Buyers are moving from learning to choosing.
3. Countries that use ChatGPT more search with it more
In the third quarter of 2026, almost a third of all messages in Slovenia and Croatia were searches. In Indonesia it was less than half of that.
In some markets, search is twice as common as in others
Share of messages that are searches, July to September 2026, twelve markets.
Slovenia
31%Croatia
30%Czechia
28%Hungary
27%Spain
26%France
26%United Kingdom
23%Germany
23%United States
21%India
16%Brazil
16%Indonesia
14%OpenAI Signals v3.0, topic by country file. Each market is the mean of its three months.
The spread lines up with how much each country uses ChatGPT. OpenAI ranks countries by messages per person, and early in 2025 that rank told you little about how a country used it: the correlation between use per person and search share was 0.30. By the third quarter of 2026 it was 0.74. The countries with the heaviest use per person now put 24% of their messages into search, and the lightest quarter of countries 16%. Estonia, for example, ranks 4th of 173 countries for use per person, and nearly a quarter of its messages are searches.
In most of Europe, search is already what people use ChatGPT for most, and its share is still rising.
4. People are talking to the assistant more
OpenAI also sorts messages by what the person is doing: asking for information, asking the assistant to do a task, or expressing something with no task attached.
Messages with no task in them more than doubled
Messages outside work, by what the person is doing.
July 2024
September 2026
OpenAI Signals v3.0, intent file, non work messages, global.
People talk to the assistant and tell it what they want and why, and the question about your product comes up somewhere in the middle of that.
A brand can be named on most "best X for Y" questions and still get dropped as soon as the buyer says what matters to them.
No brand names in it
Signals is the best public view of how people use ChatGPT and I’d rather have it than not. It shows who is searching but not what the assistants say when buyer asks. Do they name you? how do they describe you vs a competitor? and do they still recommend you once the buyer says what matters to them? That’s where decisions are made and none of it shows up as a visit.
We do that research for brands. If you want to know what the assistants are telling your buyers at each step, and where you lose them, start with the free assessment.
Data: OpenAI Signals v3.0, OpenAI, 2026, released 26 October 2026, licensed CC BY 4.0. All figures are shares of classified consumer ChatGPT messages; age group figures are derived from OpenAI's published age by topic, topic and age tables. Country figures for a quarter are the mean of its three months. Full download: openai.com/signals/data-download.