The fourth Anthropic Economic Index report released country and region breakdowns of its new "economic primitives" for the first time (Anthropic Economic Index report: Economic primitives). This brief applies them to one country. It describes India as already the world's largest exporter of IT services and home to one of the fastest-growing AI user bases, and sets out to show how Claude.ai use there differs from the global pattern, for readers concerned with AI policy, investment and deployment in India. It is a short post with figures, by Ruth Appel, published on 16 February 2026.
Data and method
The brief uses the same sample as the January 2026 report: Claude.ai Free, Pro and Max conversations from 13 to 20 November 2025, about one million globally (Executive summary and Methodology; Figure 1 gives N = 975,160), of which 58,098 came from India (Figure 4). Countries and Indian states are assigned by IP geolocation; tasks are mapped to O*NET tasks and occupation groups; the primitives are computed as in the full report, by asking Claude questions about each conversation. Country rankings include only countries with at least 200 observations.
What it finds
Large in total, small per person. India accounts for 5.8% of Claude.ai use, second to the United States (Figure 1). Adjusted for working-age population, it ranks 101st of 116 countries with enough observations, below Singapore and Malaysia. The brief notes that India's per-capita use is about what its per-capita income predicts, given the strong global correlation between the two.
Concentrated in four states. Maharashtra (15.5%), Tamil Nadu (13.2%), Karnataka (12.7%) and Delhi (10.5%) together hold over half of India's use (Figure 2). The brief reads this as adoption driven by the established technology workforce rather than broad consumer uptake.
Software-heavy. India ranks first of all countries in the share of use on software-related tasks, 45.2% of O*NET-mapped tasks, ahead of Vietnam (42.1%) and Egypt (39.2%). Educational tasks also appear among the most common (Table 1, Figure 3).
Differences on the primitives (Figure 4, India against the global average):
- Tasks take 14.8 minutes with AI and an estimated 3.8 hours without, a 15x speedup, against 15.4 minutes and 3.1 hours (12x) globally.
- Use is 51.3% work, 20.9% coursework and 27.8% personal, against 46%, 19.3% and 34.7% globally.
- Autonomy delegated to Claude averages 3.60 on the 1 to 5 scale, against 3.38.
- 84.6% of tasks could have been completed by the user alone, against 87.9%.
- Prompts need an estimated 12.2 years of education to understand and responses 12.5 years; India ranks in the top 10% of countries on the education level of Claude's responses.
How the brief interprets the results
The Implications section draws five conclusions: broader impact will require use beyond software and IT services; investing in AI can bring substantial productivity gains, which the brief says India is already extracting (the 15x estimated speedup); closing the gap between total and per-capita use requires addressing income, digital infrastructure and awareness outside the IT sector; AI autonomy appears to be serving Indian users well; and training in effective AI use could pay off, given the global correlation between prompt sophistication and response quality. For the fourth conclusion the brief cites India's higher autonomy score, longer baseline task times and the share of tasks users could or could not complete alone; the figures it reports for India are averages, and it does not break outcomes such as task success down by autonomy level. For context, the full report found that countries with higher usage per capita delegate less decision-making autonomy to Claude (Chapter 3), and it describes its primitives, including the time and education estimates, as directionally accurate estimates made by Claude rather than exact measures (Chapter 2).
Limitations
The brief inherits the full report's limits: one week of data, Claude-estimated primitives validated for direction rather than precision, and Claude.ai only (no API traffic). It does not report uncertainty for the India-versus-global differences, so it is not clear which are statistically meaningful. Its sample reflects people already using Claude, concentrated in the IT sector, so the primitives describe that population rather than Indian workers generally.
Research questions
- Are the India-versus-global differences larger than sampling noise? The brief gives point estimates only (Figure 4). Confidence intervals, for example by bootstrapping over conversations, would show which of the nine primitive differences are robust.
- Do the differences survive controlling for task mix? India's use is more software-heavy than the global average. Comparing primitives within the same O*NET tasks would show whether Indian users delegate more autonomy and bring longer tasks, or simply bring more coding tasks.
- Is higher delegated autonomy associated with better outcomes for Indian users? The brief draws this conclusion from India's average figures (Implications). Comparing task success rates across autonomy levels within India would test it directly.
- Does use spread beyond the four leading states over time? Tracking state shares across future Economic Index samples would answer whether adoption is broadening.
How it connects
- Source report. All data and methods come from Anthropic Economic Index report: Economic primitives, which Anthropic summarized in Anthropic Economic Index: New building blocks for understanding AI use.
References
- Appel, R. (2026). India Country Brief: The Anthropic Economic Index. Anthropic, 16 February 2026. https://www.anthropic.com/research/india-brief-economic-index (figure numbers in this note refer to the brief).
- Appel, R., Massenkoff, M., McCrory, P., et al. (2026). Anthropic Economic Index report: Economic primitives. https://www.anthropic.com/research/anthropic-economic-index-january-2026-report.