ANTHROPIC (CLAUDE) · LLMs · QUALITATIVE RESEARCH AT SCALE · GLOBAL STUDY

What Nearly 81,000 AI Interviews Reveal for M&E and International Development: From Small Samples to Global Listening at Scale

In December 2025, Anthropic (Claude) conducted one of the largest qualitative studies ever: nearly 81,000 participants across 159 countries and 70 languages, all in one week. Using a conversational AI interviewer, participants shared what they want from AI, what they fear, and how AI is already impacting their lives. This is not just a technical achievement—it is a new model of large-scale qualitative research with profound implications for M&E and international development.

Provider: Anthropic (Claude) Focus: Large-scale qualitative research & global listening AI Methods: Conversational AI interviewer, automated thematic classification Data: Nearly 81,000 participants, 159 countries, 70 languages

PRIMARY SOURCE Anthropic (Claude) – Official Research Publication

This case study is based on the official Anthropic (Claude) research publication:

What 81,000 people want from AI – Anthropic (Claude)

The following write-up integrates insights from this official source for the M&E and international development community.

PRIMARY SOURCE: Read the Full Anthropic (Claude) Study Explore the Interactive Quote Wall

What is this case study about?

In December 2025, Anthropic (Claude) conducted one of the largest qualitative studies ever attempted: nearly 81,000 participants across 159 countries and 70 languages, all within a single week. Using a conversational AI interviewer called "Anthropic Interviewer" (a version of Claude prompted to conduct open-ended interviews), participants shared their hopes, fears, and lived experiences with artificial intelligence.

This case study translates the findings from this landmark research for the monitoring, evaluation, and international development community. It examines what worked, what didn't, and what the results mean for how we understand human-centered development in an age of AI.

Nearly 81,000

Participants

159

Countries

Languages

1 Week

Data Collection Period

67%

Global Positive Sentiment

Part 1 — What Worked: 7 Key Advantages and Breakthroughs

1. Scale + Depth Combined (A Historic Shift)

Traditionally in M&E: scale means surveys, depth means interviews. This study achieved both simultaneously—open-ended, adaptive interviews at massive global scale. It breaks the long-standing trade-off between depth and volume, representing a new form of social science: qualitative research at scale.

2. Real Human-Centered Insights (Not Just Data Points)

Instead of structured responses, participants expressed personal stories, aspirations, and emotions. Key aspirations included: professional excellence (18.8%), personal transformation (13.7%), life management (13.5%), time freedom (11.1%), financial independence (9.7%), and societal transformation (9.4%). The data reveals human motivations, not just metrics.

3. AI Is Already Delivering Tangible Impact

A striking result: 81% of participants said AI has already helped them. Main areas of impact included productivity gains (32%), cognitive partnership (17%), learning improvements (10%), and technical accessibility (9%). AI is not theoretical—it is already changing lives today.

4. Expanding Access and Inclusion (New Opportunities)

AI is helping people learn without formal education, overcome disabilities, and access knowledge previously out of reach. Examples include individuals learning programming without prior experience, users overcoming learning disorders, and people in crisis using AI for support. AI acts as a capability multiplier, especially for underserved groups.

5. Always-Available, Non-Judgmental Support

A recurring theme: AI offers 24/7 availability, unlimited patience, and no social judgment. This enables faster learning, honest expression, and continuous support. A new type of human-AI interaction model emerges—one that fills gaps where traditional support systems are unavailable or uncomfortable.

6. Multilingual and Global Reach

The study captured voices from high-income countries and low- and middle-income regions across diverse cultural contexts. Key insight: 67% of respondents globally expressed positive sentiment toward AI. Even more notable: higher optimism in developing regions, where AI is seen as a path to opportunity and growth.

7. A New Form of Social Science

This approach introduces qualitative research at scale, with capabilities including automated thematic classification, multi-dimensional coding of responses, and extraction of representative quotes. This is not just faster research—it is a new methodology for understanding people at scale.

Part 2 — The Reality Check: 10 Critical Challenges for M&E

1. Representativeness: Who Is Missing?

Participants were existing AI users (Claude users). This implies digitally connected populations and early adopters. The most vulnerable groups—those without digital access or AI familiarity—may still be excluded from this type of research.

2. Self-Selection Bias

Participation was voluntary: only those interested in AI responded. Perspectives may skew more positive or engaged than the general population. This affects the validity of conclusions when generalizing beyond the sample.

3. The "Light vs Shade" Reality (Benefits = Risks)

The study revealed paired tensions: learning versus cognitive atrophy, productivity versus work pressure, emotional support versus dependency, economic opportunity versus job loss. The same AI capability creates both value and risk simultaneously. Notably, these tensions often coexist within the same person.

4. Reliability Concerns Are the Number One Risk

Top concern: 26.7% worry about AI unreliability—confident but incorrect answers, hallucinations, and misleading conclusions. This is a critical issue for decision-making in M&E, where accuracy is paramount.

5. Jobs and Economic Disruption

22.3% fear job impacts, making this the strongest predictor of negative sentiment. Especially relevant for writers, freelancers, and knowledge workers.

6. Loss of Human Agency

21.9% are concerned about autonomy, including over-reliance on AI and reduced independent thinking. This raises fundamental questions about human decision-making in development contexts.

7. Cognitive Atrophy (Less Thinking Over Time)

16.3% worry about reduced thinking ability, observed especially in students and knowledge workers. Educators were 2.5-3 times more likely than average to report having witnessed cognitive atrophy firsthand, presumably in their students.

8. Emotional Dependency and Social Impact

AI provides support, but some users replace human interaction, risking isolation or dependency. Particularly visible in crisis contexts and among vulnerable individuals. This is the most entangled tension, with the strongest co-occurrence of light and shade in the same person.

9. Validation and Interpretation Challenges

At this scale, human review is limited and AI classification dominates analysis. This creates risks of oversimplification, loss of nuance, and misinterpretation.

10. Ethical, Governance, and Data Risks

Participants raised concerns about privacy (13.1%), misinformation (13.6%), and governance gaps (14.7%). These issues are not solved by scale—they are amplified.

How Perspectives Vary Around the World

Clear regional patterns emerged. Globally, 67% of interviewees expressed net positive sentiment toward AI. South America, Africa, and much of Asia view AI with more optimism than Europe or the United States. Respondents from Sub-Saharan Africa (18%), Central Asia (17%), and South Asia (17%) were the most likely to say they had no concerns—roughly double the rate in North America (8%), Oceania (8%), and Western Europe (9%).

Wealthier, more AI-exposed regions want AI to manage the complexity of life; developing regions want AI to create more opportunity. The vision of AI for entrepreneurship resonates most in Africa, South and Central Asia, the Middle East, and Latin America & the Caribbean—framed as a capital bypass mechanism. Learning using AI is disproportionately important in Central and South Asia, where users describe education as a primary lever for breaking cycles of poverty.

Final Insight for M&E and International Development

The Bottom Line: AI does not just improve research—it redefines how we understand people. But the same system that enables scale introduces new risks around bias, representation, interpretation, and ethics. The future is not traditional M&E versus AI, but human-centered M&E powered by AI. Your role becomes designing better questions, ensuring inclusion, interpreting complex realities, and acting on insights. AI makes it possible to listen to thousands of voices, but the real challenge remains: Are we understanding them correctly—and acting responsibly?

Key Lessons for Evaluators and M&E Practitioners

Scale + Depth Is Now Possible

The traditional trade-off between sample size and qualitative depth has been broken. AI-enabled interviews can achieve both, opening new possibilities for mixed-methods research.

Representativeness Remains Critical

AI studies risk excluding the most vulnerable. M&E practitioners must actively design for inclusion and acknowledge sampling limitations.

Light and Shade Are Entangled

Benefits and risks co-emerge from the same AI capabilities. Evaluations must capture both sides and avoid one-sided narratives.

Reliability Is the Top Concern

For decision-making contexts, AI hallucinations and inaccuracies are unacceptable. Validation protocols are non-negotiable.

Context Shapes AI Perception

Developing regions show higher AI optimism. M&E must account for local context and avoid imposing Western-centric AI narratives.

Human Interpretation Remains Essential

AI classification accelerates analysis but cannot replace human judgment, nuance, and contextual understanding.

Frequently Asked Questions

How was this study different from traditional surveys?

Instead of structured multiple-choice questions, Anthropic Interviewer (Claude) conducted open-ended, adaptive conversations. This allowed participants to express personal stories, emotions, and nuanced perspectives—not just select from pre-defined categories. The AI interviewer adapted follow-up questions based on responses, bridging the trade-off between depth and scale.

What are the main limitations of this approach for M&E?

Key limitations include: self-selection bias (only existing AI users participated), representativeness gaps (digitally connected populations only), potential for AI classification errors, and the fact that participants were already AI users, which may skew positive. For M&E, these limitations mean findings cannot be generalized to non-AI users or digitally excluded populations without caution.

How can M&E practitioners apply these methods?

The methods can be adapted for large-scale qualitative data collection in development contexts, such as beneficiary feedback, stakeholder consultations, and participatory evaluations. However, practitioners must address representativeness, ensure data privacy, validate AI classifications with human review, and design for inclusion of non-digital populations.

What does "light and shade" mean for development programming?

"Light and shade" refers to the finding that benefits and risks are tightly bound—the same AI capability that enables learning can also cause cognitive atrophy; the same tool that provides emotional support can lead to dependency. For development programming, this means interventions must be designed to maximize benefits while actively mitigating risks, not assuming one without the other.

Main Reference & Original Source

PRIMARY SOURCE – Anthropic (Claude) (2026, March 18). What 81,000 people want from AI. Anthropic Research. https://www.anthropic.com/features/81k-interviews

Citation: Huang, S., Carter, S., Eaton, J., Pollack, S., Callender III, D., Makagiansar, N., Gonzalez, M., Carr, S., Hong, J., Handa, K., McCain, M., Millar, T., Julapalli, M., Yun, G., Alt, A., Larsson, C., Leibrock, J., Gallivan, M., Sumers, T., Durmus, E., Kearney, M., Shen, J. H., Clark, J., Stern, M., & Ganguli, D. (2026). What 81,000 People Want from AI. Anthropic.

Interpretation and analysis by EvalCommunity: The insights, lessons, and expert commentary presented in this article are offered by EvalCommunity as an independent interpretation of the Anthropic (Claude) research for the M&E and international development community. They are based on the official Anthropic publication "What 81,000 people want from AI." EvalCommunity does not claim ownership of the original research or findings. We encourage readers to consult the original source for the complete context, methodology, and limitations.

Resources for Further Learning

Advance your skills in AI-enhanced evaluation

Learn how to integrate AI-assisted qualitative methods, large-scale listening, and mixed-methods research into your own M&E practice. Access courses, templates, and expert guidance.

EvalCommunity Services AI in M&E Course