Anthropic’s AI Fluency Index is a baseline study of behaviors visible in sampled Claude.ai conversations—not a test of how AI-literate the public is. It found that iteration was common, while in-chat checks of AI reasoning and missing context appeared less often, especially when conversations produced artifacts such as code, apps, or documents. The results describe patterns in one platform’s conversations; they do not show that any behavior causes better outcomes.
What is the AI Fluency Index?
The AI Fluency Index is an Anthropic report that counts 11 observable behaviors in Claude.ai conversations. Its motivating question is whether people are developing the skills to use AI well as it becomes part of everyday life. Anthropic describes the report as a baseline for studying how AI collaboration develops over time.
As an Amazon Associate I earn from qualifying purchases.
The report page lists February 23, 2026, as its original publication date, while its embedded BibTeX record gives February 16, 2026. The precise publication date is therefore inconsistent across the page’s metadata. Read Anthropic’s AI Fluency Index.
How did Anthropic measure AI fluency?
The study analyzed 9,830 Claude.ai conversations with several back-and-forths during January 20–26, 2026. Each of 11 indicators was coded as present or absent, so a conversation could count toward multiple behaviors. Anthropic says it used a privacy-preserving analysis tool and binary classifiers: Claude Sonnet 4 classified behaviors, while Claude Haiku 3.5 detected language. A screening process excluded greetings, one-word exchanges, test messages, and pure chitchat; a manual review of 200 excluded chats found none that qualified for an indicator. The report says personally identifiable information was not included in the analysis.
#1 Best Overall
Anthropic checked whether results were broadly stable across days and six languages: English, French, Spanish, Chinese, Japanese, and German. Most rates varied by 1–5 percentage points across days and by no more than 3 percentage points across language groups. These checks support consistency within the analyzed sample, not representativeness of all Claude users, AI users, or the public.
Which AI fluency behaviors did the report track?
The Index uses the 4D AI Fluency Framework, developed by Professors Rick Dakan and Joseph Feller in collaboration with Anthropic. The framework defines 24 behaviors; the report measures 11 that can be observed in Claude.ai or Claude Code interactions. The remaining 13 include behaviors outside the chat interface, such as disclosing AI’s role in work and considering the consequences of sharing generated output.
Rank #2
In the table, the percentage is the share of analyzed Claude.ai conversations in which Anthropic identified that behavior. It is not the percentage of people who possess a skill, and the indicators can overlap.
| Behavior | Conversations |
|---|---|
| Iterates and refines | 85.7% |
| Clarifies goal before asking for help | 51.1% |
| Provides examples of what good looks like | 41.1% |
| Specifies format and structure | 30.0% |
| Sets interaction mode | 30.0% |
| Communicates tone and style preferences | 22.7% |
| Identifies when AI may be missing context | 20.3% |
| Defines audience | 17.6% |
| Questions AI reasoning | 15.8% |
| Consults AI on approach before execution | 10.1% |
| Checks important facts and claims | 8.7% |
Anthropic’s report says conversations with several back-and-forths exhibited more than twice as many AI fluency behaviors as quick exchanges. That comparison describes the conversations in its sample; it does not establish why the behaviors differed.
Does iteration make AI use better?
In the sample, conversations with iteration also showed higher rates of other measured behaviors. For example, goal clarification appeared in 54.5% of iterative conversations, compared with 30.9% of conversations without iteration. Questioning AI reasoning appeared in 17.9% versus 3.2%.
This is an association, not evidence that adding follow-up messages causes better judgment or outcomes. More complex tasks, user expertise, or other factors could influence both iteration and the other behaviors. The study also cannot tell whether a person became more fluent over time, because it does not track individuals’ skill development.
Why might people scrutinize AI artifacts less?
When Claude conversations produced an artifact—such as an app, code, document, or interactive tool—users were less likely than in non-artifact conversations to question AI reasoning (3.1 percentage points lower) or identify missing context (5.2 points lower). The companion discussion guide also reports a 3.7-point decline in checking facts.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Anthropic suggests that polished output may look finished, or that users may review it elsewhere. Those are possible explanations, not conclusions established by the data. The study observes chat behavior and cannot determine whether checks happened outside the conversation. It does, however, make independent review a useful practical consideration when an AI-generated artifact will be relied on or shared.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the Index cannot tell you
- It is not a public-opinion or population-wide skills survey. The sample consists of Claude.ai conversations meeting the study’s screening criteria.
- It does not measure all 24 framework behaviors. Some behaviors occur beyond the interface and are not directly visible in conversation data.
- It does not establish causation. The observed relationships do not show that iteration produces discernment, or explain why artifact conversations show less in-chat scrutiny.
- It does not capture every kind of review. Users may check facts or context outside Claude, which the conversation sample cannot establish.
Anthropic says initial analysis found consistency with Claude Code conversations, but calls that finding preliminary and notes that Claude Code has a different user base and functionality. The report identifies cohort analysis, qualitative study of behaviors that cannot be seen in chat, and causal questions as areas for future research.
How educators and teams can use the report
Anthropic’s discussion guide is designed for leadership groups, faculty teams, and professional learning communities. It suggests setting aside 45–60 minutes, asking participants to read or skim the report first, and choosing two or three sections for discussion. Optional activities include sending at least three follow-ups to improve an answer, inspecting an AI-generated artifact together for gaps, and writing a short preamble describing the desired collaboration and when the AI should push back. These are suggested exercises, not interventions shown by the study to improve fluency. See Anthropic’s discussion guide.
In an August 20, 2026 article, Anthropic described Claude Academy as combining Claude-specific learning with product- and model-agnostic instruction. The company says its teaching approach has shifted from emphasizing specific behaviors toward broader, durable mindsets, including human agency, deliberate delegation, practice, and verification proportionate to the stakes. It also says learners can access Academy courses and track completion and badges at academy.claude.com. These are Anthropic’s descriptions of its own educational service and approach, which may change. Read Anthropic’s account of its AI teaching approach.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
