Anthropic Education Report: The AI Fluency Index

As artificial intelligence (AI) tools become integrated into daily routines at an unprecedented pace, understanding their impact goes beyond mere adoption. An essential question arises: are individuals developing the necessary skills to use these tools effectively? Previous Anthropic Education Reports have examined how university students and educators utilize Claude, an AI tool. Students have leveraged it for report creation and lab result analysis, while educators have employed it to develop lesson materials and automate routine tasks. Recognizing that any user of AI is likely to improve their skills, this report aims to delve deeper into how individuals develop “fluency” with AI technology over time.

Measuring AI Fluency

To quantify AI fluency, we utilize the 4D AI Fluency Framework, developed by Professors Rick Dakan and Joseph Feller in collaboration with Anthropic. This framework defines 24 specific behaviors that exemplify safe and effective human-AI collaboration. Out of these, 11 behaviors can be directly observed during interactions with Claude on Claude.ai or Claude Code. The remaining 13 behaviors, which include being transparent about AI’s role in work and considering the consequences of sharing AI-generated content, occur outside the chat interface and are more challenging to track. These unobservable behaviors are crucial dimensions of AI fluency, and we plan to employ qualitative methods to assess them in future studies.

Research Methodology

For this study, we focused on the 11 directly observable behaviors. Using our privacy-preserving analysis tool, we examined 9,830 conversations that included multiple exchanges with Claude on Claude.ai over a 7-day period in January 2026. We measured the presence or absence of the 11 behaviors, with each conversation potentially displaying multiple behaviors. To ensure the reliability of our sample, we verified whether our results were consistent across different days of the week and various languages, which they were. This analysis led to the creation of the AI Fluency Index, serving as a baseline measurement of current human-AI collaboration and a foundation for tracking behavioral evolution over time.

Key Findings

Our initial study revealed two primary patterns in Claude’s usage: a strong correlation between AI fluency and the iterative refinement of conversations, and notable changes in users’ fluency behaviors when creating outputs such as code or documents.

Iteration and Refinement

One of the most significant findings is the relationship between iteration and refinement and other AI fluency behaviors. An impressive 85.7% of conversations in our sample demonstrated iteration and refinement, where users built upon previous exchanges to enhance their work instead of accepting the first response and moving on. These conversations exhibited substantially higher rates of other fluency behaviors. On average, conversations characterized by iteration and refinement displayed 2.67 additional fluency behaviors, compared to a non-iterative rate of 1.33. Notably, these conversations were 5.6 times more likely to involve users questioning Claude’s reasoning and four times more likely to identify missing context.

Creating Outputs

In contrast, 12.3% of conversations involved the creation of artifacts, including code, documents, and interactive tools. In these instances, users collaborated with AI in a markedly different manner. We observed higher rates of behaviors related to “description” and “delegation.” For example, users were more likely to clarify their goals, specify formats, provide examples, and iterate when working on artifacts. However, this increased directiveness did not correlate with higher levels of evaluation or discernment. In fact, users were less likely to identify missing context, check facts, or question the model’s reasoning in artifact conversations.

Implications for AI Fluency

These patterns raise several important considerations. Users may perceive polished outputs as complete, reducing the necessity to question further. Additionally, tasks involving artifacts may prioritize aesthetics or functionality over factual precision. It’s also possible that users evaluate artifacts through different channels, such as testing code or sharing drafts, rather than within the initial conversation. As AI models continue to produce refined outputs, the ability to critically evaluate these results will become increasingly vital.

Developing Your Own AI Fluency

AI fluency is a skill that can be developed over time. Based on our findings, we have identified three key areas where users can enhance their skills:

• Staying in the Conversation: Engaging in iteration and refinement is crucial. Treat initial responses as starting points—ask follow-up questions, challenge any aspects that seem off, and refine your queries.
• Questioning Outputs: When working with AI-generated artifacts, maintain a critical mindset. Just because an output appears polished does not mean it is free from errors or omissions.
• Expanding Your Skills: Continuously seek to understand the broader implications of AI in your work. This includes being mindful of AI’s role and the consequences of sharing AI-generated outputs.

Frequently Asked Questions

Note: The findings in this report provide a foundation for understanding and enhancing AI fluency, which is essential as AI continues to evolve and integrate into various aspects of life.