At work, the buyer side of ChatGPT grew. The seller side did not.

OpenAI’s Signals release tags every work related ChatGPT message with the job activity it serves, using the US Department of Labor’s catalogue of work activities, 165 of them in the file. Most of the list is what you would expect: editing documents, making visuals, explaining a regulation. But a handful of those activities describe one side or the other of a purchase. Some are what a marketer or a salesperson does. Some are what a buyer does. I grouped them that way and watched both sides for seven quarters.

First, the frame. A purchase has stages: finding out what exists and what it does, comparing the options, deciding and settling the price, and buying. OpenAI’s activities map onto those stages well enough: obtaining information about goods or services and having a product’s technical details explained are research; evaluating the usefulness or performance of products is comparison; determining prices, selecting equipment and negotiating contracts are decision; purchasing goods or services is conversion. On the seller side: developing marketing materials and plans and analysing the market are marketing; promoting, selling and advising on products are sales; answering customer problems is service.

Two caveats before the numbers. These are consumer ChatGPT accounts only, so company accounts, where much purchasing happens, are not counted, and everything below is understated. And the figures are shares of work messages, not people.

How much of AI search is work, by market

Before the purchase, the search. In the first article I showed that it is people over 35 who turned the assistant into a search engine. How much of that searching is done at work?

In the United States, 18% of search messages are work related. In the UK, Germany, France, the Netherlands and Portugal it is 15% or 16%; in Japan 16%. In Singapore it is 23% and in India 24%. So in Europe and America, five of every six AI searches are personal; in the business hubs of Asia, one in four is a work search. For a B2B brand, the work share is the floor of the audience that matters, and it is higher in exactly the markets where procurement runs through English.

The buyer side grew

Now the purchase itself. Across the 76 countries with data at both ends, the four buyer stages together went from 5.0% of work messages in the first quarter of 2025 to 6.0% in the third quarter of 2026. The seller side, all three groups together, went from 7.2% to 7.4%.

Both lines sit on the same chart for a reason. For every seven messages sent on the selling side at work, marketing, sales and service together, buyers now send six, up from five. The two sides are converging, and all of the movement is on the buyer’s side.

And it grew at the end of the journey

Look inside the buyer side and the growth is not evenly spread.

Research, the largest stage, went from 3.2% to 3.7%. Comparison did not move: 1.3% then, 1.3% now. Decision doubled, from 0.4% to 0.9%. Conversion, the actual purchase through the assistant, tripled from almost nothing: 0.03% to 0.09%.

Put differently, the share of buying activity that happens after the research step rose from 36% to 39%.

The assistant is being used deeper into the purchase every quarter: the price check, the choice between finalists, the negotiation, and now the first orders.

Meanwhile the marketer’s own use is flat. Developing marketing materials is the third largest work activity in the data everywhere, about 6% of work messages, and it has not grown in seven quarters. Where the seller side does grow, it is where it meets the buyer: sales and service activities doubled from a small base.

Marketers use the assistant to write at the buyer. Buyers use it to choose. Only the second use decides the sale, and the first never sees it.

Where buying at work runs hottest

Bulgaria leads at 9.5% of work messages, then Croatia 8.0%, Finland 7.9%, Romania and Slovakia 7.8%, Poland 7.5%, Czechia 7.2%. The same Central and Eastern European markets that lead consumer search lead buying at work. Among our usual markets, France is at 6.9%, Portugal 6.5%, Spain 6.4%, Canada 6.4%, Germany 6.2%, the United States 6.1%, the UK and the Netherlands 6.0%. The US is a little below the European median on volume but deeper into the journey: 42% of its buying activity sits past research, against 36% to 38% in Germany and the UK.

The other end of the scale is instructive too. In Nigeria, Argentina, Pakistan and Brazil the seller side is 11% to 13% of work messages: there, ChatGPT at work is above all a marketing copy writer.

What this means if your buyer is at a desk

Your buyer’s research used to leave tracks: a search, a visit, a download, a form. The research now happens in a chat, and the assistant answers in full sentences with names in them. Then the comparison happens there. Then, increasingly, the price check and the shortlist.

By the time a form arrives, if one arrives, the assistant has already said which three vendors are worth a call and roughly what each should cost.

Your marketing team is using the same assistant to write the material that is supposed to shape that conversation. The data says the writing side is standing still and the choosing side is growing. The question that decides your pipeline is no longer how much you publish. It is what the assistant says when your buyer asks it which one to get.

We research exactly that: what the assistants tell your buyers at each stage, who they recommend instead, and why. If you want it for your category, start with the free assessment.

Data: OpenAI Signals v3.0, OpenAI, 2026, released 26 October 2026, licensed CC BY 4.0: the work activity file (O*NET intermediate work activities, by country and quarter) and the topic by work status files. Shares of classified consumer ChatGPT work messages; medians across the countries present in both quarters. The grouping of activities into stages is ours. Full download: openai.com/signals/data-download.