The State of AI Fluency 2026
The first benchmark of AI skills measured from real behaviour — not self-reported. Findings from 412 completed conversational assessments.
The Average AI Fluency Score Is 52/100
The average professional scores 52.1 out of 100 on measured AI fluency. Only 12% reach Advanced or Expert.
Across 412 completed assessments, scores follow a broad curve centred just above 50 — squarely in the “Developing” tier: professionals who get real work done with AI, but without the principles to do it consistently. The median is 53. Because this sample is self-selected (people curious enough to measure themselves), the true workforce average is likely lower.
The AI Literacy Ladder: From Bystander to Native
49% of professionals are AI Literate — past adoption but below fluency. Only 11% are AI Native.
Beyond the five scoring tiers, a simpler four-level model captures the journey from not using AI at all to making it integral to how you think. The AI Literacy Ladder places every professional on this spectrum — and the data shows the biggest cluster sitting at “Literate but not yet Fluent.”
The AI Literacy Ladder maps the journey from non-engagement to AI-native thinking. Deep dive: The State of AI Literacy 2026.
The Understanding Gap: AI Usage Outpaces AI Knowledge
The two weakest skills of all 11 measured are AI Fundamentals (5.0/10) and Tool Landscape (5.1/10). Professionals have learned to use AI faster than they’ve learned how it works.
Skills tied to daily output — task decomposition, domain application, evaluating outputs — lead the table. Skills tied to understanding — how models work, which tools exist and when to use them — sit at the bottom. That ordering has a consequence: when AI fails in an unfamiliar way, the average professional has no mental model to fall back on.
Average score per criterion (1–10 scale), strongest to weakest. Weakest two highlighted.
The AI Safety Deficit
Safety & Responsibility is the second-weakest dimension of AI fluency: 45/100.
Professionals integrate AI into their workflow (53.3) faster than they build the habits that make that integration safe — knowing data boundaries, adjusting scrutiny to stakes, thinking about downstream impact. For organisations, this is the quiet risk: adoption metrics look healthy while the safety layer lags.
Average dimension scores (0–100). The two lagging dimensions highlighted.
The Ten Tribes: AI Skill Personas, Ranked
24% of professionals are Enthusiasts. 21% are Dabblers. Nobody — in 412 assessments — has ever scored as an Oracle.
AISA classifies every candidate into one of ten personas based on their skill profile shape, not just their score. The biggest skill cliff in the data sits between the Enthusiast (avg 53) and the Builder (avg 73) — the jump from using AI eagerly to building with it deliberately. The Oracle — principles-level mastery of how AI works — remains unclaimed.
Each persona positioned at its average score (0–100). Circle size = share of population.
What the Top 2% of AI Users Do Differently
Experts don’t just use AI more — their Safety score is nearly double the average (88 vs 45).
Only 7 of 412 assessments reached the Expert tier. Their profile is distinctive: the population’s weakest dimensions — Safety and Technical Understanding — are precisely where Experts pull furthest ahead. Expertise in AI isn’t heavier usage; it’s understanding the machine and respecting its failure modes.
Dimension averages: all professionals (grey) vs Expert tier (gradient).
AI Readiness Inside Organisations: The Spread Problem
Within a single company, employee scores ranged from 15 to 97. An organisational average hides the risk that matters.
The chart below shows eight employees of one mid-size company (anonymised) who each completed an assessment. Same employer, same tools available, same AI policies — and an 82-point spread. This is why AI readiness can’t be inferred from adoption dashboards or a team survey: the gap between an organisation’s strongest and weakest AI users is usually wider than the gap between companies.
Measured AI fluency scores of 8 employees at one organisation (anonymised).
Methodology: Measured, Not Self-Reported
Surveys ask people to rate their own AI skills. AISA watches them work. Every data point in this report comes from a completed conversational assessment — an average of 18 exchanges in which candidates demonstrate, not declare, how they work with AI. An independent evaluator scores 11 criteria across 5 dimensions against the published AISA Rubric, and every assessment passes a two-stage scoring process with a final holistic review.
The rubric independently covers 93% of the fluency markers identified in Anthropic’s AI Fluency Index (2026), plus four dimensions it could not measure — full comparison here. Sample: 412 completed assessments, February–June 2026, self-selected professionals across roles and industries. Self-selection means this sample skews AI-curious — the true workforce average is likely lower than reported here.
AI Fluency Statistics 2026: The Key Numbers
Every statistic below comes from measured assessment scores, with the plain-English meaning taken directly from the published scoring rubric. Free to cite with attribution.
- ▸37% of professionals treat AI as a black box — they use the vocabulary (tokens, models, training) but cannot apply how the technology actually works.
- ▸1 in 5 professionals who have made AI a consistent part of their workflow cannot explain how it works (22% of integrated users score below functional on AI fundamentals).
- ▸36% can name multiple AI tools but show no evidence of knowing when to use which — tool awareness without tool judgement.
- ▸Nearly 1 in 4 (23%) know AI’s limitations only from headlines — hallucinations, bias — not from anything they’ve experienced or tested themselves.
- ▸1 in 5 (20%) have no systematic process for verifying AI output before using it.
- ▸27% show inconsistent or no safe-AI practice — unaware of data boundaries and risks beyond the obvious.
- ▸Fewer than 2 in 5 (39%) demonstrated an active AI learning habit — naming something concrete they’d recently learned and acted on. Just 2% showed a systematic learning practice.
- ▸0 of 412 professionals have reached the Oracle level — principles-first understanding of how AI systems behave.
Per-skill percentages are of assessments where that skill was measured (n = 325–412 per criterion).
How to cite this report
AISA (2026). The State of AI Fluency 2026. Retrieved from https://aisa.to/state-of-ai-fluency. Statistics and charts may be reproduced with a link to this page. For interviews or custom data cuts, get in touch.
AI Fluency Benchmark — FAQ
What is the average AI fluency score?
Across 412 completed AISA assessments (2026 baseline), the average AI fluency score is 52 out of 100, with a median of 53. The sample is self-selected AI-curious professionals, so the true workforce average is likely lower.
What is a good AI fluency score?
A score of 60+ reaches the Proficient tier (top 38% of professionals). 69+ puts you in the top quartile, and 80+ in the top 10%. Only 1.7% of professionals reach the Expert tier (92+).
How is AI fluency measured?
AISA measures AI fluency through a conversational assessment: candidates demonstrate how they actually work with AI while an independent evaluator scores 11 criteria across 5 dimensions (Prompting, Critical Thinking, Technical Understanding, Workflow, and Safety). Scores reflect demonstrated behaviour, not self-assessment.
How is this different from AI readiness surveys?
Surveys ask people to rate their own AI skills. AISA observes real behaviour in a structured assessment and scores it against a published rubric — measured evidence rather than self-perception. Self-reported skill levels consistently overestimate measured proficiency.
Where do you stand?
Get your AI fluency score, your persona, and a personalised growth plan — in one 25-minute conversation.
See also: The State of AI Literacy 2026 — the companion report on AI understanding and knowledge gaps.
Next measurement: the Q3 2026 edition publishes in October.
The Science Behind AISA
In 2026, Anthropic published the AI Fluency Index — the largest empirical study of AI fluency to date, analysing nearly 10,000 conversations. AISA covers 93% of the behaviours Anthropic identified as markers of AI fluency and goes even deeper with 4 additional dimensions. The U.S. Department of Labor's AI Literacy Framework (TEN 07-25) defines what every worker needs to know about AI — AISA covers 100% of its 25 sub-competencies.Read our analysis: Anthropic's AI Fluency Study & AISA · DOL AI Literacy Framework & AISA
AISA's framework is developed by a team with deep roots in tech, behavioural science, and AI product leadership — the rubric is informed by backgrounds spanning the Metropolitan Police, Harvard, Crowdbotics (Silicon Valley), and the European School of Economics.
