Three intentions behind a message
OpenAI's September 2025 study groups consumer messages into three intentions. Asking accounts for 49%, doing for 40% and expressing for 11%. The study analyzes 1.5 million conversations, but the percentages in this chart describe messages, not that many distinct people.
These categories describe what a person wants from an exchange. Asking seeks information or guidance. Doing requests an output or a completed task. Expressing includes reflection, exploration and play. A single person's work can move through several intentions as a question develops.
An answer and an output need different briefs
Consider a writer preparing an article about a city's changing population. Asking why two population estimates differ creates a different task from requesting a finished introduction. The first calls for an explanation of definitions, dates and measurement. The second needs a reader, a purpose and a set of claims the writer is ready to defend.
Citation's interpretation is that these intentions make a useful planning tool. Before opening a conversation, finish the sentence: I need to understand, I need to produce, or I need to explore. That small choice helps define what a satisfactory response would look like. It also makes the next review step easier to name.
For an explanation, check whether it addresses the question and whether its evidence supports the answer. For a draft, check the same evidence alongside structure, tone and relevance. A polished paragraph can still contain an unsupported claim; a useful explanation can still need substantial editing before publication.
Keep exploration separate from evidence
An exploratory exchange can be valuable without producing a factual conclusion. A writer might try several analogies, rehearse an objection or discover that the original question was too broad. Those activities can shape the brief even when none of the generated text belongs in the final article.
Treat an interesting possibility as a lead to investigate. If a conversation suggests that commuting patterns explain a change in local spending, look for evidence that measures both things in a comparable place and period. Keep the suggestion and the established finding distinct in your notes so that one does not quietly become the other during drafting.
A practical way to use the three categories
- Ask to clarify. Identify unfamiliar terms, competing explanations and the evidence you still need. Record unresolved questions beside the draft.
- Do with a defined brief. Specify the intended reader, available sources and desired output. Keep the claims tied to material you have checked.
- Explore without committing. Try alternative structures or interpretations, then decide which deserve further research. Preserve uncertainty until the evidence resolves it.
These are suggested editorial practices, not outcomes tested by the usage study. The distribution helps describe activity. It does not identify which intention produces the best work, how much time any step takes or whether someone ultimately uses the response.
How to read this chart
The source covers consumer ChatGPT plans and does not describe enterprise or API activity. This page reproduces its published intention shares on a common percentage scale. The study edition is September 2025; that publication date should not be treated as the observation period for every message.
Read OpenAI's study summary and the research paper for the study design and definitions.
Questions about the data
Does asking mean searching the web?
Not necessarily. Asking describes the intention to obtain information or guidance. It does not establish that a web search happened or that the response included a source.
Can these percentages be compared directly with Copilot goals?
The products, populations and classification units differ. This chart measures consumer ChatGPT messages. The Copilot report measures classified user goals in commercial conversations. A similar percentage does not establish the same behavior.