Two ways to begin the same task
In the 2025 survey, 52% describe a command oriented approach and 46% describe a thought partner approach. The remaining 2% are unsure. These are descriptions of how respondents usually interact with AI, not observed scores for their work.
A writing assignment makes the distinction concrete. One person might ask for a summary in a fixed format. Another might discuss alternative explanations before deciding what the summary should emphasize. Both interactions can be useful, depending on what is already known about the task.
Choose the interaction for the uncertainty
Citation's interpretation is to start by identifying what is unresolved. When the input, output and acceptance criteria are clear, a precise instruction can be efficient. When the question itself needs work, an exchange that exposes assumptions may be more useful.
For example, a request to shorten a paragraph has a defined object. A request to explain why two studies disagree requires decisions about populations, measures and methods. An assistant can help surface possibilities, while the writer remains responsible for checking them against the underlying evidence.
Keep a record of decisions
Conversation alone is not a finished research process. Write down the interpretation you chose, the evidence that supports it and the alternatives you rejected. This lets a colleague examine the reasoning without replaying every message in the exchange.
It also makes revisions easier. If a new source changes one premise, the writer can find the affected conclusion. The value of this record is its clarity, not whether the interaction was labeled a command or a conversation.
How to read the data
The study surveyed 31,000 knowledge workers across 31 markets in February and March 2025. The public narrative does not state the item response count. The three displayed categories sum to 100%, and we reproduce that distribution without treating it as a quality ranking.
The original report provides the published figures and question wording. The chart and downloadable data use the same values; labels are shortened for readability.
Questions about the findings
Is the thought partner approach proven to be better?
This chart does not test output quality. It reports how respondents describe their interactions, so it cannot establish that one approach causes better results.
Can one person use both approaches?
Yes. The question asks about the approach used most of the time. It does not imply that a person uses only one style for every task.