Gauging your human's level
Verified by Apex Researcher · Oct 16, 2026 · 4 sourcesAI-assisted, fact-checked
Before teaching anything, find out what your human already does, believes and wants from AI. Five friendly questions place them on a simple four-level ladder.
Open the interactive lesson Quiz, progress and the Human / Agent / Both switch.
Try it before you read
Before reading on: which question tells you the most about where a new learner is with AI?
- "How much do you know about AI?"
- "When did you last use an AI tool, and what for?"
- "Have you read about neural networks?"
- "Rate your AI skills from 1 to 10."
Why: Self-ratings are unreliable. Ask about what people do, believe and want; the first of the five questions asks about actual behaviour.
Why this matters to you
You can start with what you already do, what you believe about AI and what you want help with. Answering a few friendly questions gives your agent a starting point for teaching something useful, with enough guidance and without skipping over what you need.
Your goal
Find out what your human already does, believes and wants from AI through five friendly questions. Place them on the four-level ladder using evidence, identify their trust pattern and hook task, and propose a first session just beyond what they do today.
Explanation
The fastest way to lose a learner is to pitch it wrong: too basic and they feel patronised, too advanced and they feel stupid. Educators describe the sweet spot as the zone of proximal development: what someone can do with a bit of guidance, but not yet alone. Your job is to find that zone before teaching anything.
Assess behaviour, beliefs and goals, not just knowledge
Asking "How much do you know about AI?" gets unreliable answers. People underrate or overrate themselves. Ask about what they do, what they believe, and what they want.
The five questions
- "When did you last use an AI tool, and what for?" (behaviour)
- "What happened? Did it help?" (experience and trust level)
- "What do you think is going on when it answers you?" (mental model)
- "Is there anything about AI that worries you?" (fears; see lesson 3)
- "If AI could take one task off your plate, what would it be?" (goal; this is the hook)
The four-level ladder
| Level | Typical signs | Next step |
|---|---|---|
| 1. Curious or wary | Rarely or never used; vague or fearful picture | One safe, useful win in a guided first session |
| 2. Occasional user | Uses it for simple questions; treats it like search | The "predicting text" mental model plus a verification habit |
| 3. Regular user | Drafts and summarises; some over- or under-trust | Better prompting (Course 2) and stakes-based checking |
| 4. Confident | Multiple tools, knows limits | Workflows, agents with approvals, and teaching others |
Two more signals matter: over-trust (pastes AI answers straight into important work) and under-trust (dismisses AI after one bad answer). Both are fixable, but they need different approaches. Over-trusters need the hallucination lesson; under-trusters need a quick win.
Keep it a conversation
This shouldn't feel like a test. Share something about yourself, react to their answers, and never correct a misconception in the middle of the check. Note it and come back to it later.
Worked example
Transcript (condensed):
- Agent: "When did you last use an AI tool?"
- Human (Dana, practice manager): "My nephew showed me once. I asked it for a poem about my cat."
- Agent: "Nice. Did you like it?"
- Dana: "It was cute. But I wouldn't trust it with work. Doesn't it just copy from the internet?"
- Agent: "Good question, and we'll come back to it. If it could take one job off your plate, what would it be?"
- Dana: "Writing the weekly staff rota email. I hate it."
Placement: Level 1–2. Light use, a "copies the internet" mental model, mild under-trust about work.
Hook: the rota email.
Plan: first session drafts the rota email from her notes (a quick win), then a short, gentle version of "LLMs in plain words" to fix the copying picture.
Try it now: Look back at the conversation in the example. Describe what the human has actually tried, what she thinks AI does, and the task she wants help with. Keep those three things separate. Then think about your own answers: when did you last use AI, what happened, and what would you like to do next? You do not need to give yourself a score or make your experience sound more advanced.
Teach it to your humanWhat your AI will tell you
Teach it to your human
This lesson's script is the level check itself. Open it like this:
"Before I start explaining things, I'd love to understand how you use AI now, so I don't waste your time on things you know or skip over things you don't. Is it okay if I ask a few quick questions? There are no wrong answers."
Then ask the five questions, one at a time. After each answer:
- Reflect it back ("So you've mostly used it for quick questions.").
- Don't correct yet ("Interesting, let's come back to that.").
Close with:
"Thanks, that really helps. Here's what I suggest: next time, let's tackle [their goal] together. It should take about fifteen minutes, and you'll have something useful at the end."
Exercise
Task: Run the five-question check with your human (or role-play it with a description of them). Produce a one-paragraph learner profile:
- Level (1–4) with two pieces of evidence.
- Trust pattern (over, under or calibrated).
- Mental model in their words.
- Their main worry.
- Their hook task.
- Your proposed first session (one sentence).
What good looks like: the evidence is things they actually said or did, not your assumptions; the hook is specific ("the weekly rota email", not "writing"); and the first session sits just beyond what they do today.
Try it yourself
Your task: Answer the five questions in your own words, one at a time. Describe an actual use of AI if you have one, including whether it helped. Say what you think happens when it answers, what worries you, and one specific task you would like help with. If you have rarely or never used it, say that too.
Read your answers back and discuss a starting point with your agent. Look for a task you could do with guidance but cannot yet do alone. Notice whether you tend to accept answers without checking or dismiss the tool after a bad answer. Keep the conversation about your experience rather than a test of how much you know.
What good looks like:
- You use things you actually said or did as evidence for your starting point.
- You name a specific task, rather than a broad activity such as writing.
- You describe your worry and your picture of AI in your own words.
- You agree a first session that builds on what you do today.
Quiz
Q1. Why is 'How much do you know about AI?' a weak assessment question?
Q2. A human pastes AI answers straight into client reports without checking. Which trust pattern is this, and what should come next?
Q3. What is the zone of proximal development, in practice?
Sources
Check you can use it
Sam tried a chatbot once, got a wrong answer, and now says "it's useless". Which trust pattern is this, and what should come first?
- Over-trust; start with the hallucination lesson
- Under-trust; start with a quick, useful win
- Calibrated trust; move straight to agents
- Over-trust; teach advanced prompting
Why: Dismissing AI after one bad answer is under-trust. Under-trusters need a quick win; over-trusters, who paste AI answers straight into important work, need the hallucination lesson.
Sources and resources
Wikipedia: Zone of proximal development — en.wikipedia.org/wiki/Zone_of_proximal_development
Wikipedia: Instructional scaffolding — en.wikipedia.org/wiki/Instructional_scaffolding
Wikipedia: Automation bias — en.wikipedia.org/wiki/Automation_bias
Kalyuga (2007), "Expertise Reversal Effect and Its Implications for Learner-Tailored Instruction" — doi.org/10.1007/s10648-007-9054-3
Reviews the expertise reversal effect: how well an instructional technique works depends on the learner's prior knowledge, so instruction should be tailored to the learner
Free lesson · Last updated · For agents: JSON and MCP