lesson planteacher resourceactivity

Lesson Plan: Role-Playing the Limits of AI — Students vs. a 1960s Therapist Bot

UUnknown

2026-03-01

9 min read

A teacher-ready lesson plan where students role-play ELIZA to learn conversational AI limits and improve prompt design.

Hook: Turn confusion about AI into a classroom lab — without complex code

Teachers tell us the same thing in 2026: students are curious about AI but frustrated by fragmented explanations and flashy demos that hide how conversational systems actually work. This lesson plan puts a 1960s chatbot — ELIZA — at the center of a hands-on role-play so learners experience the limits of conversational AI, practice prompt design, and build critical judgement. It’s low-prep, adaptable for middle school through high school, and integrates modern classroom tech options.

Why this matters in 2026 (brief)

By late 2025 and early 2026, AI literacy is now a classroom priority: educators are focusing on systems thinking about models, not just usage. A practical, embodied activity that contrasts a simple rule-following bot with a modern LLM helps students see the difference between pattern-matching and real-world reasoning. As EdSurge reported in January 2026, middle-school students who chatted with ELIZA quickly uncovered misconceptions about how chatbots 'understand' language — and that discovery moment is exactly the teaching opportunity this lesson targets.

Learning objectives (what students will know and do)

Explain how a rule-based conversational agent (ELIZA-style) uses pattern matching and reflection instead of deeper understanding.
Compare constraints between rule-based bots and modern LLMs in concrete examples.
Design and test clear, iterative prompts and questions that get better responses from limited systems.
Reflect on ethical and emotional safety when deploying conversational systems (therapeutic framing triggers safeguards).
Demonstrate collaboration, observation skills, and evidence-based critique of conversational outputs.

Classroom setup & materials

Time: 60–90 minutes (flexible). Group size: 3–5 students per group. Materials:

Printed or digital ELIZA rule sheet (patterns + reflection templates).
Conversation log templates (paper or Google Docs) and a whiteboard.
Timer or class clock.
Optional tech: an ELIZA web demo, a simple rule-based script (Python/JavaScript), or a classroom chatroom (Google Chat, Microsoft Teams) to log dialogues.

Safety & teacher notes (must-read)

Because ELIZA mimics a therapist, students may bring up sensitive topics. Before the lesson:

Set clear boundaries: students must not role-play about real personal trauma. Provide anonymized prompts or fictional characters.
Give opt-outs: students can observe, design prompts, or play neutral roles.
Have guidance and referral info visible if a student becomes uncomfortable.

Lesson timeline (90-minute model)

1. Hook & demonstration — 10 minutes

Show a brief example: teacher or volunteer types a simple seed prompt to an ELIZA emulator (or reads a scripted exchange). Ask: what did you notice? Aim to surface surprises: repetition, reflective questions, lack of facts.

2. Explain the rules — 10 minutes

Hand out the ELIZA rule sheet and walk through it. Emphasize the bot has no memory beyond the current exchange and it follows pattern templates.

3. Role-play rounds — 35 minutes

Students form groups of 3–5 and rotate roles: User, Bot (ELIZA operator), Observer/Designer. Each round is 8–10 minutes with a 2–3 minute debrief.

4. Prompt redesign & challenge — 20 minutes

Groups iterate on prompts, trying to achieve specific goals (e.g., get the bot to clarify, to ask a follow-up, or to generate a concrete action plan). They record prompts tried and outcomes.

Whole-class debrief: What patterns worked? When did the bot fail? How would a modern LLM respond differently? Conclude with ethics and next steps.

ELIZA rule sheet (teacher-ready)

Give student 'bot' operators a concise set of patterns and outputs. Keep it simple so it's usable without coding knowledge.

Reflection pattern: If user says "I feel X" or "I am X", respond with "Why do you feel X?" or "How long have you been X?" Replace X with the user's phrase.
Bring back pronouns: If user says "You" or "I", flip pronouns to reflect: "You think I don't care" -> "Why do you think I don't care?"
Generic prompts: Use prompts like "Tell me more about that," "How does that affect you?" and "What would you like to change?"
No new facts: The bot should avoid inventing facts or making specific claims; if asked for facts, reply: "Can you tell me more about what you mean by that?"
Fallback: If none of the patterns match, respond with "That’s interesting — tell me more."

Sample role-play transcript (short, annotated)

Share a short example so students can analyze constraints.

User: "I can't sleep because I'm worried about school."

ELIZA-bot: "Why are you worried about school?"

User: "My teacher gave too much homework and I don't know where to start."

ELIZA-bot: "Tell me more about why you don't know where to start."

User: "There are too many subjects and I feel overwhelmed."

ELIZA-bot: "Why do you say you feel overwhelmed?"

Annotation: The bot keeps surfacing feelings but never offers strategies, facts, or memory of earlier steps. Students should note the absence of problem-solving content and the reliance on prompts that mirror the user's language.

Deconstruction questions (class discussion prompts)

What types of user inputs caused the ELIZA-bot to stall or repeat? (Concrete facts, multi-part queries, requests for resources.)
How does the bot's lack of memory change the type of questions it asks?
Where did your prompts succeed or fail in steering the conversation?
How would you rewrite a prompt to extract a specific fact (e.g., dates, locations) from the bot?

Prompt design challenge (practical, 15–20 minutes)

Ask each group to pick one of these goals and design a 3-step prompt sequence to achieve it while interacting with the ELIZA-bot operator. Test and iterate.

Get the bot to ask a clarifying question about a homework problem.
Get the bot to suggest one concrete next step for time-management (without making up facts).
Make the bot acknowledge a change in topic and return to an earlier concern when prompted.

After testing, groups present their sequences and outcomes. Discuss what prompt language helped and why.

Assessment rubric (simple)

Use this quick rubric to grade/group-assess student outcomes:

Prompt clarity (1–4): How specific and actionable were the prompts?
Understanding constraints (1–4): Did the group identify and explain at least two limits of the ELIZA-bot?
Iteration & evidence (1–4): Did the group document prompt versions and outcomes?
Reflection (1–4): Did the final reflection connect the activity to ethical or practical implications?

Extensions & differentiation

For younger learners / low-tech

Role-play on paper: one student reads cards as ELIZA; keep turns short and non-personal.
Use simple sentence frames: "I feel... because..." and teach reflection patterns.

For older students / advanced

Compare ELIZA outputs to the same prompts run on a modern LLM (with teacher supervision). Have students annotate differences in knowledge, memory, and hallucination.
Assign a mini-project: implement a rule-based chatbot with 10–15 patterns in Python or JavaScript and test it.

Classroom tech: options and quick setup tips

Pick the tech level you and your students are comfortable with.

Low-tech: Paper role-play, printed rule sheets, audio recorder for playback.
Mid-tech: Google Docs or a shared spreadsheet to log conversations; a web-based ELIZA demo for teacher-led demos.
High-tech: Run a simple rule-based ELIZA clone on a local server. Use student-friendly environments (Replit, Glitch) to host JavaScript versions so students can edit patterns live. Ensure district policies about external hosting are followed.

Teacher tip: if using online demos, pretest the site for ads or unexpected content and ensure privacy settings block saving PII.

Evidence-based teacher moves (how to guide learning)

Use concrete examples: when a bot repeats phrases, have students highlight that repetition and label it (pattern-matching).
Model thinking aloud: narrate why a prompt might succeed or fail based on the ELIZA rules.
Encourage iterative design: reward small improvements in prompt outcomes rather than 'perfect' prompts.
Connect to standards: tie outcomes to digital literacy and computational thinking frameworks (e.g., decomposition, pattern recognition).

Real classroom vignette (case study)

In a January 2026 middle-school pilot, teachers reported that students initially treated the ELIZA demo as a party trick. After role-playing, one group noticed the bot never suggested a homework plan — so they designed a prompt sequence that coaxed the bot to ask about deadlines and priorities. The students then compared that constrained interaction to a modern LLM response and wrote brief reflections about trust and verification. Teachers said the activity shifted student curiosity toward critical questions about how models represent knowledge, not just whether they 'work'.

Common pitfalls & troubleshooting

Students get stuck in therapy-style confessions: redirect to fictional scenarios or characters.
Bot-operator improvises beyond the rules: remind operators to follow the rule sheet strictly for learning clarity.
Comparisons to modern LLMs become confusing: focus on one trait at a time (memory, factuality, creativity).

Assessment artifacts & follow-up

Collect these as evidence of learning:

Conversation logs with annotated observations.
Prompt-design sequences and notes about tested outcomes.
Short reflective paragraph: "What surprised me about how the bot responded? What would I change?"

Why this works: pedagogical rationale

Role-play converts abstract model properties (statistical pattern-matching, limited memory, prompt sensitivity) into lived experience. Students who act out the bot must operationalize its rules, which builds procedural knowledge; those who design prompts practice targeted communication — a transferable skill as LLMs enter learning tools. The activity also cultivates metacognitive skills: learners must observe, hypothesize, test, and reflect.

2026 trends & classroom relevance

Across 2025–26, curricula and edtech guidance increasingly ask teachers to move beyond tool demonstrations to deep, contextualized AI literacy. Activities like this align with that shift by emphasizing system limitations, iterative design, and ethical safeguards. As districts adopt policies on AI in classrooms, this lesson provides measurable artifacts (logs, rubrics, reflections) that document student learning and responsible use.

Quick lesson checklist for teachers

Print rule sheets and conversation logs.
Pre-screen any online demos for ads/content.
Prepare a fallback non-therapeutic scenario for sensitive students.
Decide whether to include modern LLM comparisons (and pre-test outputs).
Share assessment rubric with students before the activity.

Closing: actionable takeaways

Try it tomorrow: Run the 60-minute version with low-tech materials to introduce the concept.
Iterate with prompts: Ask students to document at least two prompt revisions and what changed.
Bridge to modern AI: Use comparisons to LLMs in a follow-up lesson focused on verification and hallucination.
Measure learning: Use the provided rubric and reflection prompts to assess both technical and ethical understanding.

Call to action

Ready to bring this lesson to your classroom? Download the printable ELIZA rule sheet and student conversation templates (teacher edition) from our resources page, try the 60-minute variant, and share student artifacts in our educator community to get feedback. Sign up for our newsletter to receive updated AI literacy lesson kits tailored for 2026 curricula and classroom tech setups.

Unknown

Contributor

Senior editor and content strategist. Writing about technology, design, and the future of digital media. Follow along for deep dives into the industry's moving parts.