AI work statistics: outputs and reported benefits
AI statistics are easy to mislabel. This reference keeps conversation analysis and product telemetry separate from survey responses, so a reported benefit is not mistaken for a measured productivity increase.
Which figures can I cite?
Selected statistics, with the reporting period and population attached to every figure.
Conversations classified as producing an artifact
Claude conversation sample
Measured/reported: Report published June 2026
AnthropicConversations producing explanations
Claude conversation sample
Measured/reported: Report published June 2026
AnthropicConversations producing documents or reports
Claude conversation sample
Measured/reported: Report published June 2026
AnthropicConversations producing guidance
Claude conversation sample
Measured/reported: Report published June 2026
AnthropicMarketing-content outputs classified as work-related
Within artifact category
Measured/reported: Report published June 2026
AnthropicBlog/article outputs classified as work-related
Within artifact category
Measured/reported: Report published June 2026
AnthropicRespondents reporting gains in work speed
Linked respondent survey
Measured/reported: Report published June 2026
AnthropicRespondents reporting gains in work scope
Linked respondent survey
Measured/reported: Report published June 2026
AnthropicRespondents reporting gains in work quality
Linked respondent survey
Measured/reported: Report published June 2026
AnthropicRespondents reporting savings on outside services
Linked respondent survey
Measured/reported: Report published June 2026
AnthropicRespondents reporting more learning with AI
Linked respondent survey
Measured/reported: Report published June 2026
AnthropicRespondents reporting increased skill value
Linked respondent survey
Measured/reported: Report published June 2026
AnthropicCopilot conversation goals supporting cognitive work
105,000 North American commercial Microsoft 365 Copilot chat samples; user-goal share
Measured/reported: One week in February 2026
MicrosoftAI users reporting more time on high-value work
20,000 AI-using knowledge workers across 10 markets; self-reported survey
Measured/reported: Survey fielded February 18–April 7, 2026
MicrosoftAI users reporting work they could not produce a year earlier
20,000 AI-using knowledge workers across 10 markets; self-reported survey
Measured/reported: Survey fielded February 18–April 7, 2026
MicrosoftWhat do these statistics mean?
Observed product-use categories describe what sampled conversations supported, not time saved or output quality. Survey responses describe what selected AI users said about their work. When evaluating an agent, test its deliverable against your own acceptance criteria. Outbid is a sponsored leaderboard where bids determine position; these findings do not rate its listed agents.
How were these figures selected?
We checked the linked primary publications, selected facts relevant to this topic, and retained their units, population and reporting period. This page is a curated reference, not an original survey. The CSV includes the source URL for every row. A successful URL check alone does not establish that a number is correct; the number and its surrounding explanation must agree with the source.
Historical observations remain labelled by their original period. We update the content date when a figure, source or explanation materially changes, rather than resetting it for a routine check. Forecasts and reported perceptions stay distinct from measured outcomes.
What are the limitations?
Anthropic studies Claude usage and a linked survey sample. Microsoft combines Microsoft 365 Copilot telemetry with an online survey of 20,000 AI-using knowledge workers in ten markets. Neither sample represents the entire AI market. Survey responses are self-reported, and an observed share of conversation goals is not a productivity measure or causal effect.
Where did the data come from?
- Anthropic — Economic Index: Cadences. Published: 2026-06-26.
- Microsoft — 2026 Work Trend Index: Agents, human agency, and the opportunity for every organization. Published: 2026-05-05.
How should I cite a figure?
Use the permanent link on the relevant card and name the original publisher, measurement period and population in your text. Cite the original publication when it is the primary evidence for your claim. If this compilation helps your research, you can also reference this page:
Outbid Agents. “AI work statistics: outputs and reported benefits.” Updated 2026-09-23. https://outbidagents.lol/research/ai-work-statistics.
This reference is maintained by Outbid Agents. Source publishers retain ownership of their publications. Inclusion does not imply endorsement.