Teaching · Practical guide
AI Literacy: Make, Check and Explain an AI-Assisted Answer
Being able to open a chatbot is not the same as knowing when to trust its output. In this lesson you will inspect an AI-assisted answer, identify evidence it needs and explain what a person must decide. All situations and records below are fictional.
What AI literacy requires
AI literacy combines understanding, practical use and judgement. A learner should be able to describe what a system receives, what it produces, which claims need checking and what harm an error could cause. Learning a list of product names is less useful because products and interfaces change. The ability to ask an appropriate question and verify the response transfers between tools.
UNESCO’s student competency framework includes a human-centred mindset, ethics, techniques and applications, and system design. This lesson uses a smaller introductory task: inspect one generated answer without assuming that fluent language proves understanding. A generative system can assemble a plausible explanation while misstating a fact or omitting an important condition. Treat the response as material to examine.
Separate three activities. Generation produces a candidate answer. Verification compares a claim with independent evidence. Decision-making determines what to do in context. Asking a chatbot to confirm its own answer is still another generated response; it does not automatically supply independent verification. A calculator, an original source or a reproducible test may provide a stronger check, depending on the claim.
Read the task before evaluating the answer
Start by identifying the user’s purpose. A brainstorming task can tolerate speculative ideas that are clearly labelled. A calculation used in an assessment needs a correct method and units. A message sent to a real person also requires appropriate tone, factual accuracy and permission to disclose its contents. “Useful” therefore depends on the task, not just on how polished the answer sounds.
For a simple inspection, write four headings: inputs, output, evidence and responsibility. Under inputs, note whether the material is public, fictional or private. Under output, state the exact deliverable. Under evidence, list what must be checked. Under responsibility, identify who can approve use. This short record makes the human role visible rather than leaving it implicit.
Worked example: a confident answer with two problems
Fictional task: summarise a training attendance record. There were 40 registered participants and 32 attended. A generated draft says: “Attendance reached 32%, showing that participants disliked the course. The organiser should cancel the next session.”
First isolate the numerical claim. Attendance is 32 ÷ 40 × 100 = 80%, not 32%. The draft confused a count with a percentage. Second isolate the interpretation. Attendance alone does not establish participants’ preferences; there may be scheduling conflicts or other reasons. Third inspect the recommended action. Cancelling a session requires information about objectives, constraints and feedback that the record does not contain.
A defensible revision is: “32 of 40 registered participants attended, giving an attendance rate of 80%. The record does not explain non-attendance. Before changing the next session, collect relevant feedback and consider scheduling constraints.” This revision keeps what the data support, removes the invented explanation and leaves the decision with the organiser.
Record the check as: attendance percentage—recalculated—corrected to 80%; dislike claim—no supporting evidence—removed; cancellation recommendation—insufficient context—replaced with a request for information. The log is an account of your reasoning, not a guarantee that every possible issue has been found.
Protect information and recognise limits
Practise with invented records. A class list can contain names, identifiers, performance information or personal circumstances that are unnecessary for a demonstration. Removing names does not necessarily make a record anonymous when the remaining details identify a person. Follow the rules for the particular institution and approved account before using real material.
Do not infer someone’s ability, honesty or emotional state from a short text or a numerical score generated by a tool. Ask what observable evidence supports the conclusion. If the evidence is missing, say so. An acceptable outcome of an AI task is a clearly explained uncertainty, a decision to consult a subject expert or a decision not to use AI for that step.
Explain assistance honestly. A brief account could say that AI helped draft a summary while you recalculated the figures and removed unsupported causal statements. This describes the process without implying that the draft was independently researched or professionally approved.
Practice: run a 20-minute literacy check
Minutes 0–4: inspect this fictional draft: “18 of 30 invited colleagues replied to a survey, so the response rate is 18%. The other colleagues oppose the proposal.” Identify the count, denominator and unsupported interpretation. Minutes 4–9: calculate the percentage and write what the record does and does not establish.
Minutes 9–14: create a four-heading record covering inputs, output, evidence and responsibility. Minutes 14–20: produce a corrected two-sentence summary and a short explanation of your checks. You can complete the activity without an AI account; the skill being practised is evaluation.
Example answer: “18 of 30 invited colleagues replied, a response rate of 60%. The available record does not establish the views of non-respondents.” The input is fictional survey data; the output is a summary; the calculation verifies 18 ÷ 30 × 100; the author remains responsible for any decision based on the summary.
Self-review: can you distinguish a calculation from an interpretation? Did you avoid treating absent data as evidence of a preference? Can you explain exactly what you changed and why? Keep these checks with the revised answer so another reader can follow your reasoning.
Reusable prompt
Select and copy this template, or customise your own using the free builder.
Inspect this fictional AI-assisted answer: [text]. Task: [purpose]. Identify numerical claims, factual claims, interpretations and proposed actions. For each, state the independent evidence needed. Do not claim verification you have not performed. Return a corrected draft that marks missing information and a short inputs/output/evidence/responsibility record. Do not infer personal characteristics.Open the prompt builder →
Pause · Think · Check
Test your understanding
Choose one answer per question, then check your reasoning. This is practice; scores are not saved.