Data report · Updated quarterly
AISA AI Fluency Index, November 2026: The Understanding Gap
The state of AI skills, measured in nearly 3,400 in-depth assessments
The average professional scores 46 out of 100 on AI fluency, and 70% score below Proficient. In almost every occupation the weakest area is the same: understanding how AI works and which tool fits the job. It is also what most separates the top 10% from everyone else.
The AISA AI Fluency Index reports how people actually work with AI, measured in conversation rather than asked in a survey. This edition draws on nearly 3,400 AISA assessments taken since March 2026, by individual professionals and by teams assessed through their employer. It adds results by occupation and seniority, what people told us before they were assessed, and how far self-belief and measured skill drift apart.
This is the third edition of the AISA AI Fluency Index, which we publish every quarter, after the June 2026 baseline and the August 2026 edition. The AI Literacy Report 2026 is its companion.
It shows where AI skills really stand, so .
The short version
Key findings on AI fluency, November 2026
- 01The understanding gap. Technical Understanding, knowing how AI works and which tool fits the job, is the lowest of the five dimensions at 41 out of 100 and the weakest area for 44% of people. It is the weakest dimension in 12 of the 14 largest occupations.
- 02It is also what separates the best. The top 10% lead everyone else most on AI fundamentals and on building AI into their workflow, and least on iterating with AI and checking its output.
- 03Fluency is still rare. The average score is 46 out of 100. 70% of people score below Proficient, and 1.1% reach Expert.
- 04Self-belief runs ahead of skill. Three in four people overestimate their score, by 18 points on average. Those who expect to score 80 or more average 50.
- 05Occupation shapes the profile. Product people (57), executives and founders (56) and engineers (54) lead; educators (38) and students (35) trail. Each occupation has its own AI persona.
- 06Directors know where they stand. Directors and heads of function predict their own score within 4 points on average. Managers and team leads score below senior individual contributors.
- 07Daily habit shows in the score. People who use AI several times a day average 60, against 33 for everyone else. People whose main AI tool is Claude average 59, against 42.
- 08Assessed at work, judged more accurately. People assessed through their employer overestimate their score by 5 points, against 19 for individuals, and score higher on all five dimensions.
46/100
Average AI fluency score
70%
Score below Proficient
1.1%
Reach Expert
+18
Points the average person overestimates
44%
Weakest in Technical Understanding
Score distribution
What is the average AI fluency score?
The average AI fluency score is 46 out of 100, and the median is 46. Most people sit in the Developing tier: they use AI regularly but without the consistent, checked practice that marks Proficient work. A score of 63 puts you in the top 25%, and 76 in the top 10%.
Share of people by score
Share of people in each tier
- Emerging0–27 · 23.6%
- Developing28–59 · 46.2%
- Proficient60–79 · 22.9%
- Advanced80–91 · 6.2%
- Expert92–100 · 1.1%
What is a good AI fluency score?
| Score | Tier | Where it places you |
|---|---|---|
| 46+ | Developing (the median) | Top 50% |
| 60+ | Proficient begins | Top 30% |
| 63+ | Proficient | Top 25% |
| 76+ | Proficient | Top 10% |
| 80+ | Advanced begins | Top 7% |
| 92+ | Expert begins | Top 1.1% |
This edition's headline
The understanding gap: the weakest AI skill in almost every occupation
Of the five dimensions of AI fluency, Technical Understanding scores lowest: knowing how AI works, where it breaks, and which tool fits the job. It is the weakest area for 44% of people and the weakest dimension in 12 of the 14 largest occupations.
Average score by dimension, out of 100
The gap is wide inside the dimension itself. On AI fundamentals, 45% of people score 4 or below out of 10 while 17% score 7 or above: the widest split of any skill. On the tool landscape, 43% score 4 or below.
What separates the top 10% of AI users?
The people in the top 10% (scores of 76 and above) lead everyone else on every dimension, but by the most on Technical Understanding: 38 points, against 30 on Safety. On individual skills, their biggest leads are AI fundamentals and building AI into the workflow; their smallest are iterating with AI and checking what it produces.
| Skill, out of 10 | Top 10% | Everyone | Lead |
|---|---|---|---|
| AI FundamentalsAmong the biggest leads | 7.5 | 4.8 | +2.7 |
| Workflow IntegrationAmong the biggest leads | 7.8 | 5.1 | +2.7 |
| Output EvaluationAmong the smallest leads | 7.2 | 5.0 | +2.2 |
| Iterative DialogueAmong the smallest leads | 7.1 | 5.0 | +2.1 |
Safety divides people more than any other dimension: it is the strongest dimension for 30% of people and the weakest for 26%.
Occupations
AI fluency by occupation: which roles score highest?
Product people score highest on AI fluency at 57 out of 100, followed by executives and founders (56), engineers (54) and consultants (54). Half of all product people reach Proficient. Educators (38) and students (35) score lowest.
Average AI fluency score · share at Proficient or above
Every occupation has its own AI fluency profile
Beyond the average, each occupation leads in different places and gathers a different AI persona. Technical Understanding is the weakest dimension in twelve of these fourteen occupations; for engineers and for executives it is Critical Thinking, by a narrow margin.
Product56.7
- Edge:
- Building AI into workflows and breaking work down for it
- Signature persona:
- The Conductor, at 2.7 times its share of the population
- Self-estimate:
- overestimates by 8 points
Executives & founders55.8
- Edge:
- Building AI into workflows and knowing the tool landscape
- Signature persona:
- The Conductor and The Architect, each about 2.5 times their usual share
- Self-estimate:
- overestimates by 10 points
Engineering53.9
- Edge:
- AI fundamentals and managing context
- Signature persona:
- The Builder (27% of engineers) and The Architect
- Self-estimate:
- overestimates by 16 points
Consulting53.5
- Edge:
- Building AI into workflows and decomposing tasks
- Signature persona:
- The Conductor, at 2.5 times its usual share
- Self-estimate:
- overestimates by 12 points
Data & analytics50.8
- Edge:
- AI fundamentals and checking AI output
- Signature persona:
- The Builder
- Self-estimate:
- overestimates by 13 points
Research & academia49.3
- Edge:
- Knowing where AI falls short
- Signature persona:
- The Sceptic, at 2.5 times its usual share
Design & creative46.5
- Edge:
- Applying AI to their own domain
- Signature persona:
- The Enthusiast
- Self-estimate:
- overestimates by 14 points
HR & people46.4
- Edge:
- Building AI into workflows and using it safely
- Signature persona:
- The Enthusiast (42% of HR)
- Self-estimate:
- overestimates by 6 points
Sales & partnerships46.4
- Edge:
- Building AI into workflows and iterating with it
- Signature persona:
- The Copy-Paster and The Tactician, each about twice their usual share
- Self-estimate:
- overestimates by 13 points
Finance & investment45.6
- Edge:
- Knowing where AI falls short and checking its output
- Signature persona:
- The Tactician (3 times its usual share) and The Sceptic
- Self-estimate:
- overestimates by 21 points
Marketing & content45.3
- Edge:
- Prompt design
- Signature persona:
- The Tactician and The Enthusiast
- Self-estimate:
- overestimates by 20 points
Operations & management43.1
- Edge:
- Safety: the only occupation whose strongest dimension it is
- Signature persona:
- The Tactician
- Self-estimate:
- overestimates by 17 points
Education37.5
- Edge:
- Closest to average on the tool landscape and domain application
- Signature persona:
- The Dabbler
- Self-estimate:
- overestimates by 29 points
Students34.9
- Edge:
- Closest to average on iterating with AI and prompt design
- Signature persona:
- The Dabbler (48% of students)
- Self-estimate:
- overestimates by 32 points
Where does your occupation's average leave you?
A free 20 to 40 minute conversation scores your AI fluency against the same rubric as everyone in this report, with a full report and a certificate.
Seniority and job titles
AI fluency by seniority and job title
AI fluency rises with seniority, with one exception: managers and team leads score below senior individual contributors. Directors and heads of function are the most accurate judges of their own AI skills of any group, within 4 points of their measured score.
| Level | Average score | Self-estimate gap |
|---|---|---|
| C-level, founders and VPs | 55.4 | +12.1 points |
| Directors and heads of function | 53.5 | +4.3 points, most accurate |
| Senior individual contributors | 50.6 | +14.5 points |
| Managers and team leads | 47.8 | +12.4 points |
| Individual contributors | 46.6 | +16.7 points |
| Juniors, interns and assistants | 41.3 | +22.8 points |
| Students | 34.9 | +31.9 points |
Engineering reverses the pattern: engineering managers, leads and directors average 61.7, against 54.6 for engineers without those titles.
Does having AI in your job title mean you are AI fluent?
Mostly. People with AI in their job title average 62.5, against 49.4 for other professionals, and 66% reach Proficient. Even so, one in three score below Proficient, and they still overestimate their score by 8 points.
AI in the job title
62.5/100
66% Proficient or above · 23% Advanced or above
Other professionals
49.4/100
35% Proficient or above · 8% Advanced or above
Self-assessment
Do people overestimate their AI skills?
Yes. Before the assessment, people predict their own score. 74% overestimate it, 44% by more than 20 points, and only 29% land within 10 points. Predicted and measured scores correlate at r = 0.23, in line with Zhang et al. (LAK26), who found self-reported AI literacy tracks objective measures at r = 0.07 to 0.24.
To set these measured results against what the major 2026 surveys of AI at work report, see AI in the Workplace Statistics 2026, our review of those reports in one place.
| Measured tier | Predicted | Measured | Gap |
|---|---|---|---|
| Emerging | 58.7 | 17.8 | +40.8 |
| Developing | 59.8 | 42.7 | +17.1 |
| Proficient | 68.1 | 68.2 | −0.1 |
| Advanced | 76.6 | 83.9 | −7.4 |
| Expert | 74.8 | 94.8 | −20.1 |
Higher expectations, barely higher scores
Measured score by the score people expected
Who knows their level best?
HR and people teams (+6), product people (+8) and executives (+10) come closest to their measured score. Students (+32), educators (+29), finance (+21) and marketing (+20) overestimate the most. People assessed through their employer overestimate by 5 points, against 19 for individuals.
Self-knowledge also follows learning. The more deliberately someone keeps learning about AI, the closer their estimate:
| No learning habit | +37.1 points |
| Emerging habit | +20.9 points |
| Active learner | +4.2 points |
| Deliberate learner | −4.2 points |
Asked before the assessment
AI habits and AI fluency: what people told us
Before their assessment, people answered one question about how they use AI. Their answers line up with their measured AI fluency. The sharpest split is frequency: people who use AI several times a day average 60, against 33 for everyone else.
“How often do you use AI tools in a typical work week?”
“What best describes your AI journey so far?”
“Which AI tool do you use most?”
“How would your colleagues describe your AI usage?”
“Do you pay for an AI tool?”
Where people work
Organisations
AI fluency inside organisations
In each organisation shown, at least 35 points separate the most and least AI-fluent person, and the widest gap is 69 points. In each, about one in three people reach Proficient. One training plan for everyone starts most people in the wrong place.
Lowest to highest score inside each organisation · average marked
These are real organisations that assess their people with AISA. To protect their privacy, we show them as Organisation A, B, C and D.
People assessed through their employer average 51.0, against 45.8 for individuals, and lead on all five dimensions, most on Safety (+5.9) and Workflow (+5.3).
Personas
The 10 AI fluency personas
Every person is matched to one of ten AI personas from the shape of their profile. The Dabbler is the most common, at one in three people; The Oracle is the rarest.
| Persona | Share of people | Average score |
|---|---|---|
| The Dabbler | 32.2% | 26.2 |
| The Enthusiast | 22.1% | 51.9 |
| The Builder | 15.8% | 70.8 |
| The Copy-Paster | 7.0% | 29.9 |
| The Tactician | 6.6% | 59.1 |
| The Sceptic | 5.3% | 42.5 |
| The Architect | 4.2% | 87.7 |
| The Conductor | 3.6% | 70.4 |
| The Bystander | 2.9% | 11.3 |
| The Oracle | 0.3% | 80.0 |
The 11 skills
The 11 AI skills, from strongest to weakest
Breaking work down for AI is the strongest skill on average, and knowing the tool landscape the weakest. The two Technical Understanding skills have the most people scoring 4 or below.
| Skill | Average /10 | Score 7+ | Score 4 or below |
|---|---|---|---|
| Task DecompositionWorkflow & Application | 5.3 | 19.2% | 30.6% |
| Context & Memory ManagementPrompting & Communication | 5.2 | 20.8% | 35.0% |
| Prompt DesignPrompting & Communication | 5.2 | 16.1% | 30.9% |
| Domain ApplicationWorkflow & Application | 5.2 | 14.2% | 28.5% |
| Workflow IntegrationWorkflow & Application | 5.1 | 19.3% | 38.3% |
| AI Safety & ResponsibilitySafety & Responsibility | 5.1 | 12.9% | 31.5% |
| Iterative DialoguePrompting & Communication | 5.0 | 12.2% | 34.9% |
| Output EvaluationCritical Thinking | 5.0 | 11.7% | 32.9% |
| Limitation AwarenessCritical Thinking | 4.9 | 11.3% | 37.5% |
| AI FundamentalsTechnical Understanding | 4.8 | 16.9% | 45.1% |
| Tool LandscapeTechnical Understanding | 4.7 | 12.0% | 42.8% |
AI skills travel together. The closest pair is Workflow Integration and Task Decomposition (r = 0.88); the most independent is Tool Landscape and AI Safety (r = 0.64), so knowing many tools says the least about using them safely.
The fast but careless profile is rare: of the people strongest at building AI into their work, 4.5% are weak at checking its output.
Growth
How AI fluency grows: learning habits and retakes
61% of people show no active habit of learning about AI. Average scores climb steeply with the habit, from 25 with none to 78 for deliberate learners.
Average score by learning habit · share of people
21% of people demonstrate more skill in the conversation than they put into words when describing their own work.
What happens when people take AISA again?
Scores hold: 98.9% of people who retake land in the same or an adjacent tier, and first and second scores correlate at r = 0.80. On the second attempt, scores rise by 6.1 points on average, 66% improve and 32% move up a tier. Among people who retake at least two weeks later, scores rise by 4.8 points and 64% improve.
Change over time
What changed since the August 2026 AI Fluency Index
The average moved from 48 to 46, and the share below Proficient from 67% to 70%. The ranking of skills held: Tool Landscape is still the weakest skill and Task Decomposition the strongest.
| Measure | August 2026 | November 2026 |
|---|---|---|
| Average AI fluency score | 48 | 46 |
| Median score | 48 | 46 |
| Below Proficient | 67% | 70% |
| Reach Expert | 1.3% | 1.1% |
| Average self-assessment gap | +19 | +18 |
| Overestimate by the Emerging tier | +40 | +41 |
| Underestimate by the Expert tier | −27 | −20 |
| The Dabbler, most common persona | 28.5% | 32.2% |
| The Builder | 17.2% | 15.8% |
| Workflow & Application, highest dimension | 50 | 47 |
| Technical Understanding, lowest dimension | 41 | 41 |
| Weakest skill: Tool Landscape | 4.8 | 4.7 |
| AI Fundamentals | 5.0 | 4.8 |
| Strongest skill: Task Decomposition | 5.4 | 5.3 |
Method
How AI fluency is measured
Every number in this report comes from AISA assessments: 20 to 40 minute adaptive conversations in which people show how they actually work with AI. Each is scored against the published AISA rubric of 11 skills in five dimensions, every score is tied to evidence from the conversation, and every result is calibrated before it is issued.
Measured, not self-reported
Scores come from demonstrated behaviour in conversation, never from a self-rating survey.
Integrity-checked
Every assessment is checked for answers that are not the person's own; 8.4% were flagged and are marked on their report.
Stable on retake
98.9% of people who retake land in the same or an adjacent tier.
Cross-referenced
The rubric covers 93% of the markers in Anthropic's AI Fluency Index and all 25 sub-competencies of the U.S. Department of Labor's AI literacy framework.
This edition covers assessments taken from March to November 2026 by individual professionals and by teams assessed through their employer, across engineering, product, design, data, marketing, sales, HR, finance, consulting, operations, education and students. Occupation and seniority come from the role each person described. Each edition includes every assessment to date. Full methodology
AI Fluency Index FAQ
What is the average AI fluency score?
The average AI fluency score is 46 out of 100 in the November 2026 edition of the AISA AI Fluency Index, based on nearly 3,400 AISA assessments. The median is also 46, which sits in the Developing tier.
What is a good AI fluency score?
A score of 60 or more is Proficient and places you in the top 30% of professionals. 80 or more is Advanced (top 7%), and 92 or more is Expert, a level reached by 1.1% of people assessed.
Which occupations are the most AI fluent?
Product people score highest at 57 out of 100, followed by executives and founders (56), engineering (54) and consulting (54). Education (38) and students (35) score lowest. Each occupation also has a distinct profile: product people and executives are most often Conductors, engineers are most often Builders.
Do people overestimate their AI skills?
Yes. Three in four people overestimate their AI fluency, by 18 points on average. People who expect to score 80 or more average 50. Predicted and measured scores correlate at r = 0.23, consistent with Zhang et al. (LAK26), who found self-reported AI literacy correlates with objective measures at r = 0.07 to 0.24.
What is the weakest AI skill?
Technical understanding: knowing how AI works and which tool fits the job. The Tool Landscape and AI Fundamentals skills have the lowest averages of the 11 skills measured, and Technical Understanding is the weakest dimension for 44% of people and in 12 of the 14 largest occupations.
How often is the AISA AI Fluency Index published?
Quarterly. Each edition includes every AISA assessment to date. The November 2026 edition is the latest; earlier editions are the August 2026 edition and the June 2026 baseline.
Archive
Past editions of the AISA AI Fluency Index
- August 2026 editionThe first full Index: score distribution, the prediction gap, personas and roles.
- State of AI Literacy 2026AI literacy as measured in AISA assessments.
- June 2026 baselineThe first State of AI Fluency, now part of the August edition.
Related AISA research
How to cite this report
AISA (2026). AISA AI Fluency Index, November 2026: The Understanding Gap. https://aisa.to/ai-fluency-index/november-2026
Download the full report (PDF)Benchmark your team against the Index
See every person, team and function on the same AI fluency scale, set against everyone in this report.
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 criteria. 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.
Official Member Of
