Interactive teaching tool · Prompt literacy · Responsible AI
Prompt Lab
Treat prompting as specification design, not a search for magic words. Build a prompt, inspect what it leaves implicit, and plan how a human should review the output before it is trusted or used.Private by design · this page sends nothing anywhere
Build the specification
Generated prompt
Add a task, then build the prompt.
TRACE prompt structure
TaskReferences & evidenceAudience & contextConstraints & outputEvaluation & reviewTRACE turns an informal request into an inspectable specification. The point is not to make a prompt longer; it is to make the evidence boundary, assumptions, output requirements, and review criteria visible.Audit an existing prompt
Paste a prompt you already use. This local heuristic checks for visible specification signals. It cannot determine whether the prompt is good, safe, or factually grounded.Revision suggestions
Plan the output review
Choose what the AI-generated output contains or will be used for. The tool generates a review plan before the output is published, submitted, scored, or used to make a decision.Human review plan
Prompt literacy: what students should learn
1. Prompt quality ≠ factual reliability
A precise instruction can produce a precise-looking error. Students should evaluate evidence separately from instruction-following.2. Evidence boundaries matter
Specify what sources are allowed, what counts as missing evidence, and when the system should stop rather than infer.3. Role prompts do not create expertise
“Act as an expert” changes framing; it does not grant credentials, access to authoritative data, or professional accountability.4. Uncertainty should stay visible
Ask the system to separate sourced facts, inference, assumptions, and unresolved questions instead of smoothing them together.5. Review depth should match consequence
A brainstorming prompt and an eligibility decision should not share the same verification standard or automation level.6. Privacy is upstream of prompting
No wording trick makes restricted data appropriate to disclose. Tool approval, data minimization, consent, and institutional policy come first.Compare the instruction
Under-specified
“Summarize this research and tell me what it means.”More inspectable
“Using only the supplied article, summarize the authors’ research question, method, findings, and stated limitations for a graduate seminar. Cite the section supporting each finding. Separate the authors’ claims from your interpretation. If the article does not support a requested conclusion, say so.”Suggested classroom sequence
Predict → Prompt → Inspect → Verify → Revise. Ask learners to predict what evidence a good answer would require before prompting. After generation, compare the output with sources or calculations, identify unsupported confidence, and revise the specification. This keeps the cognitive work visible instead of treating the model output as the endpoint.Method note. The tool develops patterns visible in my earlier prompting and research workflow—explicit tasks, source-bounded review, evidence checks, revision criteria, and placement/context instructions—into a reusable teaching framework. The risk layer is informed by NIST's Generative AI Profile and UNESCO's guidance for generative AI in education and research, which emphasize risk management, privacy, validation, and human agency. This is an educational aid, not a legal, policy, or compliance determination. NIST AI RMF · UNESCO guidance