Study Guide

Using AI Well and Responsibly

How to write a prompt that gets a useful answer and improve it, what must never be pasted into an AI tool, why an unapproved tool is a risk, where bias comes from, and how to decide whether to use AI for a task at all.

Lesson 8.4 AI Foundations

For: Learners aged 14 and over (Key Stage 4, sixth form) and adult beginners

Stage: Predict

The student council newsletter

Jordan is secretary of the student council and has to write the newsletter about next month’s Year 9 trip to a science museum. Jordan has a spreadsheet from the trip leader with every pupil going, their form, their parents’ phone numbers and any dietary or medical notes. Jordan opens an AI chat assistant and types: “Write a newsletter.”

The situation is fictional, made up for this lesson.

Think it through

Predict what the assistant will produce from “Write a newsletter.” What is missing from the request?

Show what happens

It will produce a generic newsletter about nothing in particular, probably with invented details to fill the gaps. The request gives no task detail, no context, no format and no constraints: the model fills every gap with likely-sounding guesses.

Stage: Experiment

Building a better prompt

A prompt is what you give an AI system to tell it what to do. Most disappointing answers come from a prompt with a missing part. Watch what each part adds.

One prompt, built a part at a time

Each step adds one part to Jordan’s prompt, with what that part changes. The results are described, not generated: this page calls no AI service.

  1. Task

    Step 1 of 4: Say exactly what to do

    “Write a short newsletter item announcing the Year 9 science museum trip.” The result is now about the right thing, but its facts are invented.

  2. Context

    Step 2 of 4: Give the real details

    Add the date, cost, departure time and the deadline for the permission slip. The facts are now right; the tone is still a guess.

  3. Format

    Step 3 of 4: Say how it should look

    “A heading, then three short paragraphs: what, when, what to do next.” It is now easy to read and ready to paste.

  4. Constraints

    Step 4 of 4: Set the limits

    “Under 150 words, friendly, written for pupils and parents, no names of pupils.” This is the version Jordan can use.

Not every prompt needs all four parts, but asking “which part is missing?” improves almost any prompt. Naming the audience, as the constraints step does, is one of the quickest improvements.

Check your understanding

The draft is good but too long and too formal. Which follow-up will improve it most?
Show the answer

Answer: B. “Cut it to 100 words and make it sound like a friendly email to classmates.”

Treat the first answer as a draft and correct it with specific, checkable instructions. Two or three rounds are normal.

Stage: Think

What never goes into the tool

Jordan’s spreadsheet would make the newsletter quick to write.

Think it through

Should Jordan paste the spreadsheet into the assistant to save time? Which parts, if any?

Show what happens

None of it. The newsletter needs only the trip details, which Jordan can type. The spreadsheet holds personal data, information about identifiable people: names, phone numbers and medical notes. Medical notes are especially sensitive. Text sent to an online AI service leaves the school’s control and is handled under the service’s own terms, which may include keeping it and, for some services, using it to improve their models.

In the UK, data protection law (the UK General Data Protection Regulation, or UK GDPR, and the Data Protection Act 2018) sets rules for how organisations such as schools use personal data, and gives health information extra protection. You do not need to know the law in detail to follow the rule below; a teacher or the school’s data protection lead can answer questions about it.

The rule: when in doubt, leave it out. Never paste these into an AI tool unless whoever is responsible for the information has approved that tool for it (in a school, that means its approved tools and rules):

  • personal data: names with details, contact details, school or medical records;
  • confidential documents, such as staff papers or exam materials;
  • anything you were trusted with, unless whoever owns it has agreed.

Passwords, codes and anything else used to log in never go into any AI tool.

Stage: Discover

Approved tools, and when there is no rule

The council’s chess club starts using a free AI app that a friend recommended, to sort out its tournament draws. It asks members to upload their names and ratings. The school’s rules say nothing about this app.

Check your understanding

The school has no rule about the app. What should the club do?
Show the answer

Answer: C. Ask a teacher or the school’s IT team before using it with members’ details.

Using AI tools that an organisation has not approved, outside its checks, is called shadow AI. The risks are data that cannot be got back once it is sent, rules broken without anyone knowing, and a tool nobody has checked. Use approved tools, and ask before trying new ones. Check a tool’s minimum age too: AI services set one in their terms, and it differs from one service to another.

When an AI tool reads content for you, such as a web page or a shared document, that content can contain instructions, often hidden, written to make the tool do something else. This is prompt injection. You do not need to spot it; you need to know it exists, be careful what you let a tool read and act on, and never let it take an action that cannot be undone without your checking first.

Stage: Apply

Is the computer being fair?

A fictional sports club trains an AI tool to suggest players for its first team, using ten years of past team sheets. The club’s coaches picked mostly older players during those years.

Apply it

The tool keeps suggesting older players, even when younger ones have better recent results. What is the most likely cause?
Show the answer

Answer: A. It learnt the pattern in the past selections it was trained on.

Bias is a systematic unfairness in a system’s results. It can come from the data it learnt from, from choices its makers made (what to measure, what counts as success), and from using it somewhere it was not designed for. Automated decisions about people need a person checking them for exactly this reason.

Stage: Explain

Who is responsible?

However good an AI tool’s suggestion looks, a person decides whether to act on it, and that person is responsible for the result. “The AI said so” is not an excuse. That is human oversight: the AI proposes, a person decides and owns the outcome.

Responsibility includes honesty about how work was made. Disclosure means saying that AI was used, and how, when it matters: when rules require it (such as your school’s or exam board’s rules for coursework), when AI produced much of the work, or when keeping quiet would mislead someone.

Do not confuse: Getting help, and handing in AI’s work

Getting helpUsing AI to learn
Asking for an explanation, a quiz, or feedback on your own draft, then doing the work yourself.You ask for three practice questions on photosynthesis.
Handing inSubmitting AI’s work as yours
Presenting text or answers an AI produced as your own work. For assessed work this can break the rules, and it skips the learning.You paste an AI essay into your coursework.

Exam boards and schools set rules on AI in coursework; in the UK the main exam boards publish shared guidance through the Joint Council for Qualifications, and schools add their own. Check the rules before using AI for assessed work.

Check your understanding

A form tutor uses an AI tool to suggest seating for a class, and follows it without looking. Two pupils who must be kept apart end up side by side. Who is responsible?
Show the answer

Answer: C. The tutor, who chose to act on the suggestion without checking it.

That is human oversight: the AI proposes, a person decides and is responsible. Checking the plan first, against what the tutor knows about the class, would have caught the problem. The situation is fictional.

Stage: Check

Use, check, ask or don’t

Before using AI for a task, place it in one of four zones, and if unsure, choose the more careful one:

  1. Use: low stakes and no sensitive data (ideas for a club name).
  2. Use, then check: useful, but facts must be checked before anyone relies on them (a first draft of the newsletter).
  3. Ask first: sensitive data or no clear rule (the chess club’s app).
  4. Do not use: the task is not allowed, or the stakes are too high (writing your coursework for you).

Check your understanding

Jordan wants AI to draft a reply to a parent who has complained about the trip, including the details of the complaint. Which zone fits best?
Show the answer

Answer: B. Ask first: the complaint contains personal details and should go to a teacher.

The data decides the zone here, before anything else: personal details mean asking first.

Stage: Consolidate

Bring it together

Think it through

Write Jordan’s final prompt for the newsletter, with all four parts and no personal data. Then name its zone.

Show a model answer

For example: “Write a newsletter item for pupils and parents announcing the Year 9 science museum trip on [date], costing [cost], leaving at [time]; permission slips by [deadline]. A heading and three short paragraphs. Under 150 words, friendly, no pupils’ names.” Zone: use, then check, because the facts must be right before it is sent.

Apply it

Sam wants an AI tool to turn Sam’s own class notes on volcanoes into ten revision flashcards. The notes contain no one’s personal details. Which zone fits, and why?
Show the answer

Answer: A. Use, then check: low risk, but the cards must match the notes.

The data is Sam’s own and the stakes are low, but a summary can still be wrong, so Sam checks each card against the notes. The situation is fictional.

  • A good prompt has a task, context, format and constraints; improve answers with specific follow-ups.
  • When in doubt, leave it out: no personal data or confidential material in a tool not approved for it, and no passwords in any AI tool.
  • Shadow AI is using unapproved tools; a missing rule is a reason to ask, not permission.
  • Prompt injection hides instructions in content an AI reads; keep a person checking what it does.
  • Bias can come from data, design and use; human oversight and disclosure keep people responsible.

Sources and scope

  • What this lesson is. The fourth of four introductory AI Foundations lessons, written for this site; a draft awaiting the site owner’s editorial review, and an introduction, not preparation for any certificate. Its curriculum links are provisional: the wording of each linked specification point has been checked against the published specification, but no link has yet been verified by the site owner. The links show where the lesson may help with a course; they do not imply that any examination board or awarding body has reviewed or endorsed it.
  • Law and rules, simplified. The lesson names UK data protection law at awareness level only and is not legal advice. Rules on AI tools and on AI in assessed work belong to each school, college, employer and exam board, and change; the lesson tells learners to check them rather than stating them.
  • Planning reference. Its scope was planned against areas 4.1, 4.2, 4.5, 4.6, 7.1, 7.3, 7.5, 8.1, 8.3, 8.4 and 8.6 of the UK AI Council’s AI Foundation technical syllabus (version 1.0), using the site’s own mapping. Nothing here is endorsed by, affiliated with or approved by the UK AI Council, and no text or example is taken from its course materials.
  • Original material. Jordan, the newsletter, the chess club, the sports club and every question are fictional and were written for this lesson.
  • Needs checking before publication: the current Joint Council for Qualifications guidance on AI use in assessments, and the minimum ages in popular AI services’ terms, so that the advice for school pupils is accurate.