All activities Professional learning
Teachers climb eight rungs of prompts for the same classroom task, predict each mock AI output before reading it, and weigh what every added detail gains and what it costs.
A chatbot predicts its answer from patterns it learned plus the words you give it, so your prompt is the context it predicts from. Adding detail usually narrows the output toward what you want, but not always for free. In this unplugged session, teams climb a "ladder" of eight prompts for one real teacher task (a 4th-grade water-cycle exit ticket). At each rung, they predict what the AI will produce, read a printed mock output, and score the tradeoff. Two rungs are traps: one where a prompt's built-in misconception gets repeated back as fact, and one that works well but pastes private student information into the tool. Teachers leave with a prompting habit they can model for students: add the context that serves the learning, and stop before it costs accuracy or privacy.
Hook
Ask the room to finish each phrase aloud as you add words: "The…" (anything goes). "The water…" "The water in the puddle…" "On a hot, sunny afternoon, the water in the puddle…" (dried up, evaporated). Say: "Every word I added narrowed your prediction. A chatbot does something similar: it predicts its answer from patterns it learned plus everything in your prompt. Your prompt is its context. Today we test when more context helps, and when it costs something."
Facilitator noteIf people ask how the prediction works under the hood, save it for a follow-up session. Keep the focus on the prompt as context.
Explore
One person on each team is the "chatbot." Flip Rung 1 ("water cycle questions"). The chatbot has 60 seconds to write the questions they'd produce, with no follow-up questions allowed, because a chatbot answers what it's given. Teammates then read Output 1 on Handout B and compare. Ask: "What did the chatbot have to guess? Grade level? How many? What kind of thinking?" Log Rung 1 on Handout C.
Facilitator noteHuman "chatbots" usually make the same guesses as the mock output: generic recall questions at a mixed level. That's the point: vague prompts force guessing, and the guess is the most common pattern.
Practice
Teams flip Rungs 2–8 one at a time. For each rung: predict in one sentence what will change in the output, read the matching output on Handout B, then score the rung on Handout C: what improved, what it cost (time, variety, steering, privacy), and whether it was worth it. Rotate the reader for each rung.
Facilitator noteWatch Rung 5. Some teams won't notice that the output repeats the prompt's misconception (clouds as water vapor). If they miss it, ask: "Is every answer in the key correct?"
Model
Think aloud about Rung 5: "My prompt assumed clouds are water vapor. The output built an answer key on my mistake and marked it true. A more specific prompt made the output worse because my specifics were wrong. The model tends to follow the framing I give it." Then Rung 6: "This output is genuinely useful. And I just typed two students' names, disability plans, and quiz failures into a tool. Watch me rewrite it." Write on chart paper: "Make one version with shorter sentences and a word bank, and one with three questions instead of five, including one drawing question." Same instructional benefit, no identifying information. Show the district's guidance slide.
Facilitator noteEmerging-tech angle (Teacher ELE 1.3): many tools now add memory, file uploads, and saved custom instructions, which means more context travels with every prompt. The privacy question gets bigger, not smaller.
Debrief
Each team puts a sticky note on the rung where they'd stop for this task and says why. Discuss: "Which rung gave the biggest jump in quality for the least effort?" (Usually Rung 3: stating the evidence of learning you want.) "Which rungs made the output worse or riskier?" "Would writing these five questions yourself have been faster?" For some teachers, it would, and that's a legitimate answer.
Facilitator noteConnect to the 2026 ELE question: the best prompt is the one that states the evidence of learning you want. Rung 3 is Evidence Before Tools in prompt form.
Transfer
On the back of Handout C, each person writes a three-rung ladder for a real upcoming task: a vague prompt, a prompt that states the audience and the evidence of learning, and a prompt that adds one helpful constraint. Beside it, they write one thing they'll never put in a prompt and one sign that they should stop prompting and just write it themselves. Then plan how you'd run a student version: students climb a ladder for their own study task and explain which rung helped their learning.
Facilitator noteFor students under 13, the teacher runs any live prompts on a projector; the ladder itself works on paper at any age.
The session is designed to run unplugged: prompts are cards, outputs are printed, and the human "chatbot" in Explore makes the prediction idea concrete. Printed mock outputs also guarantee every team sees the same results, which live tools can't.
After the Practice step, the facilitator runs Rungs 1, 3, and 5 live in a district-approved chatbot on the projector. Participants compare the real outputs with the mock ones: Does the real tool repeat the cloud misconception or correct it? Results vary by tool and day, and that variation is itself a lesson. Never demonstrate Rung 6 with real names; if you show it, use the rewritten version. In a virtual session, share the cards and outputs as slides revealed one at a time, with a shared scorecard per breakout room.
Does it need a screen? Running three rungs live lets teachers check the paper claims against a real tool and see that outputs vary, and that some tools correct a wrong premise while others repeat it. The ladder itself is better on paper, because fixed outputs keep the discussion on the tradeoffs rather than on whatever the tool happened to say today.
What you should be able to see or collect if it worked.
Teachers predict how each added prompt detail changes AI output, preparing to teach students that a prompt is context for prediction and still needs checking.
Use your three-rung ladder for a real task this week and keep the outputs. Then teach students one ladder climb for a study task, and collect their best prompt plus a sentence explaining which rung improved their learning and what they checked in the output.
