What People Use AI For

About This Chart

Data Visualization Method

• Data: Real-world Claude.ai conversations, classified against O*NET work tasks (Anthropic Economic Index). These are observed conversations, not a survey in which users describe how they use AI, and each classification reflects the task at hand rather than a user's actual occupation.
• Period and area: A global snapshot of roughly one million conversations, observed from February 5–12, 2026.
• Selection: Around 3,260 raw task labels were grouped into work areas and detailed tasks. Unclassified and non-task conversations (about 7.03%) were excluded, so the chart does not represent the full 100% of Claude.ai usage.
• Processing: aistockQ Charts re-grouped similar O*NET tasks into a two-level structure — work area and detailed task — and shortened long task names into compact labels for readability.
• Chart form: A three-stage flow, AI Usage → Work Area → Detailed Task. Every ribbon is drawn at equal width on purpose; usage rank is carried by color and ordering, from higher (lighter blue) to lower (darker blue).

Creator's Note

Thousands of individual tasks cannot all be read at once when ribbon thickness carries their exact share — a few large categories fill the screen and everything below them disappears. I wanted a single view where the structure of AI use, the makeup of each work area, and the rough order of tasks could all be scanned at a glance.

So every ribbon is set to the same width, and the usage hierarchy moves into color and arrangement instead. Where exact proportion is the point, ribbon thickness matters; here, the author's reading is that color and ordering carry the rank clearly enough on their own.

Data Sources

• Anthropic Economic Index — the underlying dataset of classified Claude.ai conversations and their O*NET task shares.
• Anthropic Economic Index · March 2026 Report — the accompanying report describing method and findings.

Limitations

This reflects Claude.ai usage only. The people using Claude.ai may differ from generative-AI users as a whole in country, occupation, and the kind of work they bring, so the distribution should not be read as AI usage across all platforms.

The observation window is short — about eight days — so seasonal or one-off patterns can sit inside it. Linking each conversation to an O*NET task also relies on automated, model-based classification rather than fully manual human coding.

The work areas, detailed tasks, short labels, and the Other grouping are aistockQ's own re-classification for this chart, not Anthropic's official final categories. It is best read as Anthropic raw data recompiled by aistockQ.

What This Chart Doesn't Show

Ribbon width is not usage share. All flows are drawn at equal width, so a thicker or thinner ribbon does not mean more or less use. Rank is shown instead through color and order — lighter blue and a higher position for higher usage, darker for lower.

The percentages behind the data are task-classification shares, not user shares. A task making up, say, 11% of classified usage does not mean 11% of people, 11% of developers, or 11% of all AI use do that task.

Other and Other Tasks appear in gray, outside the blue rank gradient. This does not make them small — as a combined residual of many separate professional tasks, Other is in fact a large share. Gray simply marks it as a separately grouped category, not a low-ranked one.

Finally, this is a February 2026 snapshot rather than annual data, drawn from a single platform. It should not be treated as a complete measure of how AI is used everywhere.