Britain asks AI: what do people ask when they talk to AI?

In September 2026, 51% of British consumer ChatGPT messages were someone wanting to know something (22.7% seeking a fact, 28.5% asking for guidance or explanation), the share of work-related requests halved from 62% to 31% since July 2024, and Britain ranked 41st of 147 countries for using the assistant to look things up. Analysis by Campaign Salience of OpenAI Signals v3.0 (26 October 2026).

What is OpenAI Signals data, and what does it leave out?

OpenAI Signals v3.0, released on 26 October 2026 under a CC BY 4.0 licence, is the closest thing that exists to a census of how the public uses an AI assistant. It samples one million consumer ChatGPT messages a month from July 2024 to September 2026, adults only, and labels each one automatically: what it is about, whether it is work, and whether the person is asking something, delegating something or simply expressing something. Nobody reads the messages. Noise is added to every published figure so that no individual message can be recovered.

It describes the consumer app and nothing else. Enterprise accounts, the API and coding tools are excluded, as is anyone who opted out of AI training. It gives shares, never counts, and no geography below country level. The data tells us a great deal about what questions ordinary people ask in the AI answer economy, and almost nothing about how professionals use AI at work.

What do British people ask ChatGPT?

In September 2026, 51% of all British consumer ChatGPT messages were someone wanting to know something: 28.5% practical guidance (how to do it, learn it, decide it) and 22.7% seeking a specific fact. A third were requests to make something, mostly text: writing 22%, images and other media 7%, technical help 4%. One message in ten (10%) was classified as self-expression, with no question or task in it at all.

These resemble the questions once put to a very logical advice columnist, answered with encyclopaedic reach: practical, sense-making questions ("what does this error message mean", "how do I change a fuse"), tutoring, health and self-care questions, and requests for specific facts and figures. People are asking their AI assistant things they used to type into a search engine or ask a knowledgeable friend. The difference is that they expect an answer inside the AI, not a list of weblinks to click on and go check. The answer an AI assistant gives is often the start of a dialogue, which is a different kind of interaction from a search engine request.

Is ChatGPT a tool or a conversation partner?

Among British non-work related messages, the share of messages that delegate a task (write this, draft me a version of that) fell from 37% in July 2024 to 22% in September 2026, while statements that neither ask nor instruct, which OpenAI classifies as "expressing", rose from 12% to 33%. Questions held at around half of non-work messages throughout, 50% in July 2024 and 45% in September 2026.

People increasingly talk with their assistant rather than instruct it. They react to what it says, mull it over, think out loud and carry on a conversation. AI has moved from being a tool to being an interlocutor, and the public's habits have followed. Material that reaches people through AI answers is paraphrased, discussed and argued with. This might explain the zero-clicks phenomenon that is revolutionising the internet. AI robots are being despatched across the web to fetch data that can be synthesised into meaning making answers.People are staying put, waiting on their AI to provide that meaning, rather than clicking on a weblink to find out for themselves.

Has work use of ChatGPT fallen in Britain?

In July 2024, 62% of British consumer ChatGPT messages were work-related; by September 2026 the figure was 31%, slightly below the United States at 33%. This is not evidence that fewer people use AI for work. Much of the professional audience is likely to have moved to employer-provided enterprise accounts, agents and APIs, which are outside the sample. More likely is that the composition of the consumer audience has changed: the early adopters who used the entry-level app for work have been joined by a much larger public using AI instead of a search engine. That shift also explains the fall in delegation and the rise in dialogue inside AI.

Does Britain use AI to look things up more than other countries?

Seeking information made up 23% of British messages in September 2026, in line with Germany (23%), France (25%) and Ireland (26%), above the United States (20%), and well above Brazil (16%), India (15%) and Nigeria (14%). In the large emerging markets, work tasks and making things remain far more prominent, perhaps reflecting what a free system does to the marginal cost of image-making, writing and delegation. Among non-work messages, 45% of British messages are questions, against 38% in India and 35% in Nigeria.

This is not a story of unusually high British adoption. On OpenAI's own per-capita ranking of countries by message volume, Britain stood 41st in the quarter to September 2026. It is a story about a particular use: Britons treat the consumer assistant disproportionately as a place to look things up. The dominant request has moved from "help me to" to "tell me".

Where does Britain rank on wanting to know?

On the share of messages that seek a fact or ask for guidance or explanation, averaged over the quarter to September 2026, Britain ranks 41st of 147 countries at 52.5%, in the same band as the Netherlands (37th), Germany and Canada (38th) and Kuwait (34th), and ahead of the United States at 49th (51.6%). Central and Eastern Europe dominates the top of the table: Slovenia (62%), Croatia (61%) and Slovakia (60%), then Bosnia, Czechia, Montenegro, Greece, Romania, Serbia and Bulgaria at 58%. Spain (57%), the Nordic countries and Ireland (56%) sit just below. Small countries carry more noise; the full ranking is in the table.

What do British people use ChatGPT for at work?

The third of British use that is work-related related to editing or writing tasks. In the quarter to September 2026, 19% of British work messages were about editing written material, 7% creating visual designs, 5% developing marketing material, 4% careers and education advice, 3% researching products or services, with resolving computer problems and programming at 3% each. By topic, 39% of work messages were about writing, 24% practical guidance and 12% seeking information. The United States looks the same.

The activities that describe a journalist, campaigner or policy researcher (gathering information for news stories, advocating for community needs, assessing the impact of regulations, researching social issues, interviewing, presenting research) sum to 1.2% of British work messages; every research and analysis activity in the O*NET list combined comes to 4%.

The professionals who do that work are either on enterprise accounts, working through agents and APIs, or not yet using AI for it, and the data cannot distinguish the three. What it can say is that the analytical professions are nearly invisible in the consumer product while the public's information-seeking grows. The risk is a divide between professionals with access to higher-end, paid-for models and a consumer public served by the entry-level one.

Do British people use AI for news?

They say no, and they mean it. The Reuters Institute's Digital News Report 2026 ranks Britain lowest of 48 markets for using AI chatbots for news (4% weekly) and lowest for trust in chatbot answers about news (6%), with trust in news itself down five points to 30% and half the population (50%) sometimes or often avoiding the news. Set against that, Signals shows the British public among the most likely in the world to ask an assistant for a fact.

Both can be true if the public draws a line between news and AI answers. News is a branded, institutional product, something the BBC does, something half of them now avoid. Asked what pay is like in a town, whether rents are rising or how a benefit works, Britons type it into an assistant without experiencing the answer as news. The facts that journalism once printed as civic content have moved to a medium the public does not file under "journalism", even though some AI answers are assembled from news sites.

The industry is making room for this. In the Reuters Institute's January 2026 survey of 280 digital leaders, publishers said they plan to scale back service journalism (−42) and general news (−38), the listable, answerable material they expect AI to commoditise. That is exactly the content the British public increasingly asks for. Britain, with its loyalty to a handful of legacy news brands, looks set for a hybrid: the BBC or the Mail for headlines, the assistant for the facts beneath them.

Do British people trust AI answers?

Stated trust in chatbot answers about news is 6%, concern about misinformation is 77%, among the highest of any market, and 73% of Britons say they distrust news on social media. Yet the Signals record shows the same public asking an assistant for facts more often than almost anyone. Low trust does not stop use; it did not stop social media use either. The British public is taking a growing share of its practical facts from a surface it says it does not trust and cannot see sources for, while telling pollsters it is worried about exactly that. That is a demand for provenance, for facts that are dated, sourced and attributed where they are read, that few are currently meeting.

What does this mean for communicating with the public?

Three things follow for anyone whose job is to put facts in front of people.

Answers, not links. The dominant public mode is a plain-language question expecting a direct answer. A fact you want people to hold has to be answerable in that form, dated, and attributed in the same sentence as the figure, because an AI surface does not show its workings and readers are no longer clicking through to check.

A place where meaning is made. The growth in consumer AI is in information and reflection, not task delegation. People argue with their assistant, interrogate it, react and reason with it. The answer layer is a persuasion environment as much as a utility, and what is retrievable there shapes what people come to believe.

A widening gap. The public's share of consumer AI is rising while visible professional use falls. What ordinary people are told by the answer layer and what professionals know from primary sources, or from better models, can diverge. The material that closes that gap has to be retrievable, current and quotable by consumer assistants. The organisations that will matter are the ones whose evidence is already there when the question is asked.

How this analysis was done, and how to query it

All figures are shares of messages from OpenAI Signals v3.0 (public release tables, 26 October 2026, CC BY 4.0), computed by Campaign Salience. "All British messages" figures are the topic shares for country GB; the asking / doing / expressing figures are for non-work messages; the work-activity figures are shares of work-related messages, with ONET activity names from the ONET 28.3 IWA Reference. The country ranking averages July, August and September 2026 for every country with data in all three months. Each table on this page is downloadable in CSV and JSON above, with its source table named in the vintage line. See also how we make a report something AI assistants cite.

Updated 5 November 2026. Licence: CC BY 4.0, attribution to Campaign Salience; underlying data OpenAI Signals v3.0, CC BY 4.0.