AI Fluency Benchmarks: 9 Personas [2026]

AI fluency benchmark data from 2,047 AISA assessments reveals something most skill frameworks miss: AI capability isn't binary. It's not "uses AI" versus "doesn't use AI." It's a spectrum with nine distinct clusters, each defined by measurable behavioural patterns across prompting, critical thinking, technical understanding, workflow integration, and safety awareness. The spread runs from a composite score of 10.4 (Bystander) to 87.7 (Architect) — a 77-point gap that no single training programme can bridge in one step.

This post breaks down the full distribution: where most people land, which personas are surprisingly rare, and what actually separates one level from the next. If you're benchmarking yourself, your team, or your hiring pipeline, these are the reference numbers.

The 9 AI Personas Ranked by Average Score

AISA assigns one of nine AI personas based on a candidate's composite score and dimensional profile. Each persona captures not just a skill level but a behavioural archetype — how someone actually interacts with AI tools, evaluates outputs, and integrates AI into their work. Here's the full ranking from our dataset of 2,047 completed assessments.

A few things jump out immediately. The gap between Builder (70.9) and Conductor (71.0) is just 0.1 points — yet they represent fundamentally different profiles. Builders score high on technical implementation; Conductors score high on orchestration and workflow design. Same neighbourhood, different skill shapes.

The Architect persona sits alone at 87.7, a full 16+ points above the next cluster. And at the bottom, Bystanders at 10.4 aren't just low scorers — they demonstrate minimal engagement with AI tools in any capacity.

What the Average Score Doesn't Tell You

Composite scores compress five dimensions into a single number. Two people can score 52 and look completely different underneath. An Enthusiast at 52.0 typically shows strong prompting and workflow scores but weaker critical thinking and safety awareness. A Sceptic at 41.4 might actually outperform the Enthusiast on source triangulation and bias detection — but score lower overall because they rarely use AI tools in practice.

This is why AISA uses personas rather than just score bands. The persona captures the shape of someone's fluency, not just the magnitude. You can explore the full AI fluency framework to see how dimensions map to personas.

How These Compare to Industry Benchmarks

Anthropics AI Fluency Index, which AISA cross-references at 93% marker coverage, uses a similar multi-dimensional approach. Stanford's 2024 AI Index Report found that self-reported AI proficiency among knowledge workers averaged around 3.2 on a 5-point scale — roughly the midpoint. Our measured average of 46.1 out of 100 tells a similar story: most professionals sit in the middle of the distribution, not at the extremes.

The World Economic Forum's 2025 Future of Jobs Report estimated that 59% of workers will need reskilling by 2030, with AI literacy as a core component. Our data suggests the reskilling challenge is more nuanced than a single percentage implies — the type of skill gap varies dramatically by persona.

AI Fluency Benchmark Distribution: The Dabbler Majority

Nearly one in three assessed professionals — 31.3% — lands in the Dabbler persona. Combined with Bystanders (3.1%) and Copy-Pasters (7.1%), that means 41.5% of the assessed population sits below a composite score of 31. These aren't people who've never heard of ChatGPT. They're professionals who have tried AI tools but haven't developed structured approaches to using them.

The Bottom Third: Bystanders, Dabblers, and Copy-Pasters

Bystanders (3.1%, avg 10.4) have minimal direct AI experience. Their assessment conversations reveal awareness that AI exists but little hands-on usage. They represent the smallest persona group — most people who take an AI fluency assessment have at least dabbled.

Dabblers (31.3%, avg 26.1) are the largest single group. They've used AI tools — often ChatGPT or a coding assistant — but their usage is sporadic and unstructured. They tend to accept first outputs without iteration, rarely adjust prompting strategies, and don't have a mental model for when AI is or isn't appropriate. Career changers are disproportionately represented in this group.

Copy-Pasters (7.1%, avg 30.1) are interesting because they use AI frequently but without critical engagement. They paste in content, take the output, and move on. Their prompting scores can be moderate, but critical thinking and safety scores are consistently low.

The Middle: Sceptics, Enthusiasts, and Tacticians

Sceptics (5.4%, avg 41.4) are the contrarians. They score relatively well on critical thinking — they question AI outputs, identify limitations, and understand failure modes. But their workflow and prompting scores lag because they don't use AI enough to develop those skills. They know what can go wrong but haven't figured out what goes right.

Enthusiasts (21.6%, avg 52.0) are the second-largest group. They use AI regularly, enjoy it, and can articulate its benefits. Their weakness is typically in confidence calibration — they trust outputs more than they should and underinvest in verification. Our prediction gap data supports this: across all 1,137 candidates who predicted their scores, the average overestimation was 19.1 points.

Tacticians (7.2%, avg 59.8) sit at the top of the Developing tier. They've started building systematic approaches — prompt templates, verification workflows, model selection criteria. They're the persona most likely to be actively studying AI rather than just using it.

The Top: Builders, Conductors, and Architects

Builders (15.3%, avg 70.9) are the largest group in the Proficient tier. They integrate AI into real workflows, build on outputs iteratively, and demonstrate solid technical understanding. Many are engineers or product managers — our data shows Engineering averaging 54.4 and Product averaging 55.7 across all personas.

Conductors (4.1%, avg 71.0) score similarly to Builders on composite but differ in profile. Conductors excel at orchestrating multi-step AI workflows, managing context across sessions, and designing processes where AI handles specific components. Think of them as the project managers of AI usage.

Architects (4.3%, avg 87.7) are the rarest high-performing persona. They demonstrate expert-level capability across all five dimensions. They can design AI systems, evaluate model tradeoffs, implement safety guardrails, and teach others. Only 89 out of 2,047 assessed professionals reached this level.

The Missing Middle: Why Tactician and Conductor Are the Rarest Profiles

Tactician (7.2%) and Conductor (4.1%) together account for just 11.3% of all assessed professionals. This creates a visible gap in the distribution — a "missing middle" between casual AI users and genuinely proficient ones. Understanding why matters for anyone designing training programmes or career development paths.

The Tactician Bottleneck

The Tactician persona requires something most AI users skip: deliberate systematisation. Moving from Enthusiast (avg 52.0) to Tactician (avg 59.8) isn't about using AI more — it's about using it differently. Tacticians build prompt libraries, establish verification routines, and select models based on task requirements rather than habit.

This transition demands metacognition about AI usage. You have to step back from "this tool is useful" and ask "how am I using this tool, and is my approach optimal?" Most professionals don't make that shift without external structure — a framework, a rubric, or a benchmark that shows them where their gaps are.

The Conductor Gap

Conductors are even rarer at 4.1%. The jump from Builder (avg 70.9) to Conductor (avg 71.0) isn't about score magnitude — it's about dimensional balance. Builders often have a spike in one or two dimensions (typically technical understanding and workflow). Conductors maintain high scores across all five dimensions, including safety and critical thinking.

This is the hardest profile to develop because it requires breadth. A strong engineer might build sophisticated AI integrations but score poorly on safety awareness. A careful product manager might excel at critical evaluation but lack technical depth. The Conductor has closed all those gaps.

What This Means for Teams

If you're an engineering manager looking at your team's AI readiness, the missing middle tells you something important: the path from "uses AI" to "uses AI well" has two distinct chokepoints. The first is systematisation (Enthusiast → Tactician). The second is dimensional balance (Builder → Conductor). Training programmes that focus only on prompting skills will move people through the first bottleneck but not the second. A proper AI competency assessment can identify which bottleneck each team member faces.

What Separates Each AI Skill Level: Score Bands and Behaviours

AISA's scoring rubric evaluates 11 criteria across 5 dimensions. Each criterion is scored 1-10 with explicit behavioural anchors. Here's what distinguishes each AI fluency level in practice — not just in score, but in observable behaviour during the assessment conversation.

Emerging Tier (0-27): Bystanders and Dabblers

Professionals in this tier typically:

• Use single-turn prompts without context or constraints
• Accept AI outputs at face value without verification
• Cannot articulate how a language model generates text
• Have no systematic approach to deciding when AI is appropriate
• Show limited awareness of data privacy implications

The dimension averages tell the story. Across all 2,047 assessments, Technical Understanding averages just 39.0 — the lowest of all five dimensions. For Emerging-tier candidates, this number drops further. They often cannot explain the difference between a model and an application, or why the same prompt might produce different outputs.

Developing Tier (28-59): Copy-Pasters Through Tacticians

This is the widest tier, spanning four personas and a 30-point range. The key differentiator within this tier is intentionality:

• Copy-Pasters (30.1) use AI frequently but reactively — they paste content in and take what comes out
• Sceptics (41.4) apply critical thinking but avoid deep engagement
• Enthusiasts (52.0) engage deeply but lack critical rigour
• Tacticians (59.8) combine engagement with emerging systematisation

The overall median composite score is 46 — squarely in this tier. The average is 46.1. Most professionals are here.

Proficient Tier (60-79): Builders and Conductors

Proficient-tier professionals demonstrate:

• Multi-turn prompt strategies with iterative refinement
• Active output verification against external sources
• Model selection based on task requirements
• Integration of AI into established workflows (not just ad hoc usage)
• Awareness of bias, hallucination risks, and appropriate use boundaries

Workflow & Application is the highest-scoring dimension overall at 47.2, and it's where Proficient-tier candidates really separate themselves. They don't just know how to prompt — they know how to embed AI into a process that produces reliable results.

Advanced and Expert Tiers (80-100): Architects

The Architect persona (avg 87.7) sits in the Advanced tier, with some individuals reaching Expert (92+). These professionals can:

• Design end-to-end AI-augmented workflows for teams
• Evaluate model architectures and their implications for specific use cases
• Implement safety and governance frameworks
• Teach and mentor others on AI usage
• Critically assess AI research and vendor claims

Only 4.3% of assessed professionals reach this level. For context on what a good AI score looks like relative to your role, the data varies significantly — Engineers average 54.4, Product managers 55.7, and Students 36.4.

How to Move Up One AI Persona Level

The most actionable question isn't "what's my score?" — it's "what's the smallest change that moves me to the next level?" Based on the dimensional data, here are the highest-leverage moves for each transition.

Bystander → Dabbler: Start Using AI at All

The gap here is pure exposure. Pick one task you do weekly — summarising meeting notes, drafting emails, researching a topic — and use an AI tool for it consistently for 30 days. Don't optimise. Just start.

Dabbler → Copy-Paster or Enthusiast: Develop a Prompting Habit

Dabblers average 26.1. The jump to Enthusiast (52.0) is the largest single-persona gap at nearly 26 points. The key is moving from sporadic to regular usage and learning basic prompting patterns: providing context, specifying format, giving examples. The overall Prompting & Communication dimension averages 44.3 — there's significant room for most people to improve here.

Enthusiast → Tactician: Build Verification Into Your Workflow

Enthusiasts use AI a lot but trust it too much. The prediction gap data is telling: candidates overestimate their scores by 19.1 points on average. Tacticians close this gap by building verification habits — checking AI outputs against primary sources, testing prompts with known-answer questions, and maintaining a log of where AI gets things wrong. Read more about the proficiency levels that define each transition.

Tactician → Builder: Go Deep on Technical Understanding

Technical Understanding is the weakest dimension overall at 39.0. Tacticians who want to become Builders need to understand how AI tools work, not just that they work. This means learning about token prediction, context windows, temperature settings, and model selection tradeoffs. You don't need to train models — you need to understand enough to make informed decisions about which tool to use and why.

Builder → Conductor: Close Your Weakest Dimension

Builders typically have one or two strong dimensions and one or two weak ones. The path to Conductor is closing the gap. Look at your AISA report's dimensional breakdown. If safety is your weakest area (it averages 41.0 overall — the second-lowest dimension), focus there. If critical thinking lags (42.4 average), work on structured evaluation of AI outputs.

Conductor → Architect: Teach Others

The final jump requires synthesis. Architects don't just use AI well — they can explain why their approaches work, design systems for others, and evaluate novel AI capabilities against real business needs. The most reliable path to this level is teaching: mentoring colleagues, writing internal documentation, or designing AI workflows for your team. With the recent agent safety incidents making headlines, Architects are the people organisations need to evaluate and govern AI deployments.

---

Related reading: What AI Persona Are You? 10 Types Explained — the full breakdown of each persona's behavioural profile.

Related reading: What Is a Good AI Score? — how to interpret your composite score in context.

Related reading: AI Proficiency Levels: 9 Tiers Explained — detailed behavioural anchors for every level.

Frequently Asked Questions

What percentage of people are AI beginners?

Based on 2,047 AISA assessments, 34.4% of assessed professionals fall into the two lowest personas: Bystander (3.1%) and Dabbler (31.3%). These individuals score below 28 on the composite scale, placing them in the Emerging tier. Adding Copy-Pasters (7.1%, avg 30.1) brings the total below the Developing midpoint to 41.5%. The overall average composite score is 46.1, meaning most professionals are still in the early-to-mid range of AI fluency.

How rare is advanced AI fluency?

Advanced AI fluency is uncommon. Only 4.3% of assessed professionals (89 out of 2,047) reach the Architect persona, which averages 87.7 on the composite scale. The next tier down — Conductor — accounts for just 4.1%. Combined, only 8.4% of assessed professionals demonstrate Proficient-or-above capability with strong dimensional balance. The rarity increases further at the Expert composite tier (92-100), which represents a subset within the Architect persona.

Can I change my AI persona?

Yes. AI personas reflect current demonstrated capability, not fixed traits. AISA's retest data shows 98% of retakers land in the same or an adjacent tier, which means scores are stable but not static — they move when your skills genuinely change. The most common transitions we observe are Dabbler → Enthusiast and Enthusiast → Tactician, driven by deliberate practice in prompting, verification, and workflow integration. Take the AI fluency assessment to establish your baseline, then use your dimensional report to target specific gaps.

What is the average AI fluency score?

The average composite score across 2,047 AISA assessments is 46.1, with a median of 46. This places the typical assessed professional in the Developing tier (28-59). Scores vary significantly by role: Engineering professionals average 54.4, Product managers 55.7, Founders 55.2, and Students 36.4. The highest-scoring dimension overall is Workflow & Application at 47.2; the lowest is Technical Understanding at 39.0.