India is already one of the main markets for Claude.ai, but that absolute leadership hides a reality: adoption is highly concentrated and most of the population still doesn't use AI regularly. What does that tell you about policy, investment, and talent development? Here I explain the technical findings and the practical implications.
Executive summary
Anthropic's Economic Index report analyzes ~1 million conversations with Claude.ai from November 2025 and shows that India accounts for 5.8% of global use, second only to the United States. However, when adjusted for working-age population, India falls to rank 101 out of 116 countries with sufficient observations. In practice, that means the total volume is driven by population size, not by high per-capita use.
Indian users employ Claude.ai more for professional work, give the AI more autonomy, and assign it tasks that would take a long time without assistance. That produces notable productivity gains, but also shows adoption is concentrated in four states and in tech-related roles.
What Claude.ai usage in India shows
- Global share: India = 5.8% of total
Claude.aiusage in the sample. - Geographic concentration: Maharashtra, Tamil Nadu, Karnataka and Delhi account for more than half of usage. That matches IT hubs: Mumbai, Chennai, Bangalore, Hyderabad and Delhi NCR.
- Occupational profile: 45.2% of tasks mapped to the
O*NETcatalog are software-related, the highest share among analyzed countries. - Types of use: 51.3% for work, 20.9% for coursework, 27.8% for personal use.
With those numbers, adoption looks driven by the existing tech workforce rather than broad diffusion across other sectors.
Economic primitives: key metrics and how to read them
Anthropic introduces so-called "economic primitives" to measure how humans and AI collaborate. Some values relevant for India:
- Productivity speed: tasks that would take on average 3.8 hours without AI are completed in 14.8 minutes with
Claude.ai. That's a speedup of ~15x. Globally the speedup is ~12x (3.1 hours -> 15.4 minutes).
This difference suggests that Indian users bring more complex or higher-friction tasks and gain larger relative benefits from using AI.
- Work orientation: 51.3% of use is for work (vs 46.0% global), confirming a strong professional bias.
- AI autonomy: average score 3.60 on a 1–5 scale (vs 3.38 global). Indian users are more willing to delegate decisions to the AI.
- Human-only capability: 84.6% of tasks could be completed by a human alone (vs 87.9% global). India has more tasks that require capabilities humans don't have without AI.
- Proxy education level: human prompts estimated at 12.2 years of education and AI responses at 12.5 years. India sits in the top 10% for sophistication of AI output.
Technically, these metrics combine time measures, task classification with O*NET, and ordinal scales for autonomy. They're useful to quantify where AI creates value and how usage patterns differ across countries.
Geographic and occupational concentration: why it matters
Adoption is highly focused. Four states concentrate most of the usage and nearly half of tasks are software-related. That has several consequences:
- Risk of inequality: productivity benefits concentrate in specific regions and professions.
- Opportunity to scale: if practices and training spread to other regions and sectors, India could multiply its economic impact.
- Dependence on the IT sector: right now AI diffusion is a natural extension of the software services and outsourcing business.
Think of an SME in a secondary city or a manufacturing worker — the chance they're using Claude.ai today is low. That's a gap that public policies and companies can try to close.
Implications for policy, investment and technical training
The evidence points to concrete actions:
- Invest in infrastructure and access: closing digital gaps will let more people take advantage of tools like
Claude.ai. - Training programs in prompt engineering and productive AI use: the correlation between prompt sophistication and response quality implies high returns to training.
- Support adoption outside IT: targeted incentives for sectors like manufacturing, education and health can diversify economic impact.
- Consider regulatory frameworks that build trust: higher levels of autonomy also require standards for safety and accountability.
For investors: the observed productivity gains (15x on complex tasks) point to businesses and tools that amplify those capabilities as high-return opportunities.
Methodology and technical limitations
- Data: ~1 million
Claude.aiconversations in the window Nov 13–20, 2025. Includes Free, Pro and Max plans. - Geographic assignment: IP-based geolocation.
- Occupational classification: mapping tasks to
O*NETand grouping toSOClevels. - Country inclusion: only countries with >= 200 observations were reported to keep statistical stability.
- Limitations: short time window, possible bias toward early or professional users, and dependence on automated task classification.
In technical terms, this means the figures are robust for detecting macro patterns but should be complemented with longitudinal studies and sectoral sampling to design precise policies.
Final reflection
India is already using AI at the frontier of complex professional tasks and is getting higher-than-average returns. But the challenge now is to turn that absolute leadership into broader, more equitable adoption. Investing in infrastructure, training and practical regulatory frameworks is not just a technical issue — it's a strategic decision to let millions more benefit from the same productivity gains.
Original source
https://www.anthropic.com/research/india-brief-economic-index