Week 7: Follow-up Questions and Improvement

 

Welcome to Week 7 of the AI Literacy Course.

In Lesson 6, you created a RUKF prompt, tested it, and saved the first response. This week, you will decide what to do with that response.

A first AI response is not automatically a finished answer. It is material that you can evaluate, question, verify, revise, accept, or reject.

The central principle is:

Do not ask AI to “make it better” until you have decided what “better” means.

A useful follow-up identifies a specific weakness and requests an observable change. Sometimes, however, another prompt is not the right solution. You may need to verify a claim externally, restart with a better prompt, consult a qualified person, or stop.

Learning goals

By the end of this lesson, learners should be able to:

• Evaluate a first AI response before prompting again

• Identify the highest-priority weakness

• Write a targeted follow-up request

• Distinguish clarification, depth, challenge, reformatting, and practice

• Decide when external verification is necessary

• Recognise when to restart or stop

• Compare versions using consistent criteria

• Record follow-ups, checks, and decisions in a revision log

1. The first response is a starting point

AI responses can appear polished and complete. Clear headings, confident language, and detailed explanations may create the impression that the task is finished.

But fluency is not the same as quality.

The first response may contain:

• useful information;

• unclear language;

• unsupported claims;

• missing details;

• incorrect assumptions;

• unsuitable advice;

• a structure that does not support the task.

Pause before writing another prompt.

Ask:

• Does the response support my original purpose?

• What do I understand better?

• What remains unclear?

• What appears unsupported?

• What important information is missing?

• What decision still belongs to me?

A smoother answer is not necessarily a better answer.

2. Evaluate before prompting again

Use five categories to examine the first response:

Category	Meaning
Useful	Supports the original purpose
Unclear	Difficult to understand or open to different interpretations
Unsupported	Makes a claim without adequate evidence
Missing	Leaves out information needed for the task
Inappropriate	Unsuitable for the audience, context, risk level, or intended use

One part of a response may be useful while another is unsupported or inappropriate.

For example, an AI-generated explanation of diabetes may define a term clearly but include treatment advice that should come from a qualified healthcare professional.

Do not try to repair every weakness at once. Identify the weakness that matters most for your purpose.

3. Choose one priority

A targeted improvement begins with one priority.

Possible targets include:

• a paragraph;

• a factual claim;

• an explanation;

• an example;

• a missing perspective;

• the language level;

• the structure;

• an assumption;

• a source problem.

Changing several things at once makes comparison difficult. If the next version changes its examples, structure, length, conclusion, and level of detail, you may not know which change helped.

One target at a time makes the effect easier to inspect.

Weak follow-up:

“Make it better.”

This does not explain what is wrong or what should change.

Stronger follow-up:

“Explain the difference in the second paragraph using one workplace example. Keep the distinction between correlation and causation.”

The stronger follow-up names the target, requests a change, and explains what important meaning must be preserved.

4. Write a targeted follow-up

A practical follow-up can contain three parts:

Target + requested change + condition for success

Target

Identify the exact part that needs attention.

“The second paragraph…”

Requested change

State what should be changed.

“…needs a clearer explanation…”

Condition for success

Explain what the improved version should include or preserve.

“…using one workplace example without removing the distinction between the two concepts.”

Combined follow-up:

“Explain the second paragraph more clearly using one workplace example. Preserve the distinction between correlation and causation.”

Not every follow-up needs all three parts in separate sentences. The important point is that the requested improvement should be visible and testable.

5. Choose the right follow-up move

Different weaknesses require different kinds of follow-up.

Clarify

Use clarification when the meaning is difficult to understand.

You may ask for:

• a definition;

• a distinction;

• an example;

• a simpler restatement;

• an explanation of a technical term.

Example:

“Explain the second paragraph in simpler language without removing the distinction between training and inference.”

Simpler language can improve accessibility, but check that important meaning has not disappeared.

Deepen

Use a depth follow-up when an explanation is too general or superficial.

You may ask for:

• key factors;

• assumptions;

• supporting evidence;

• causes and consequences;

• relevant conditions;

• limitations.

Example:

“List the key claims, assumptions, and evidence that support this conclusion.”

A longer answer is not automatically deeper. Real depth should make important claims and assumptions easier to examine.

Challenge

Use a challenge follow-up to test an interpretation or conclusion.

You may ask:

• What assumption does the conclusion depend on?

• Under what conditions might the conclusion change?

• Is there another explanation supported by evidence?

• What is an important limitation?

• What evidence would weaken the conclusion?

Example:

“Identify one serious limitation of this explanation and describe when that limitation matters.”

An alternative view should not be added merely to create the appearance of balance. Two claims do not necessarily have equal evidence.

Reformat

Use reformatting when the content is potentially useful but difficult to compare, review, or apply.

Example:

“Present the three alternatives in a table using the same criteria: cost, benefit, limitation, and evidence needed.”

Changing the format may improve usability, but it does not correct factual errors.

Turn the response into practice

A follow-up can also change passive reading into active learning.

Examples:

“Ask me one question at a time and wait for my answer before giving feedback.”

“Create three retrieval questions from this explanation, but do not show the answers yet.”

This approach will be developed further in Lessons 8 and 9.

6. Know when another prompt is not the answer

Some weaknesses cannot be solved by continuing the conversation.

Use this guide:

Situation	Appropriate next action
One explanation is unclear	Ask for clarification
The answer is too superficial	Ask for depth
An assumption needs testing	Ask for a challenge or limitation
Useful content is difficult to compare	Request a clearer format
A factual claim lacks evidence	Verify it outside the chatbot
The original prompt framed the wrong problem	Restart
Risk is high or professional judgement is needed	Stop and consult a qualified person
The purpose and criteria have been met	Stop prompting and complete the work yourself

External verification

Asking AI to check its own answer is not independent verification.

The system may repeat the same error, produce another unsupported explanation, or invent a more convincing citation.

If the problem concerns a factual claim, date, law, statistic, quotation, official rule, or named source, check it outside the conversation.

This may involve:

• reading the original source;

• using an official website;

• consulting a textbook or database;

• using an appropriate calculation tool;

• asking a qualified person.

Detailed source criticism using SIFT will be introduced in Lesson 12.

Restarting

Restart when the original conversation has a faulty foundation.

This may be necessary when:

• the purpose was unclear;

• the RUKF prompt contained conflicting instructions;

• the task was too broad;

• important context was missing;

• the role encouraged an unsuitable response;

• the conversation has drifted away from the goal.

Restarting is not failure. It may be more efficient and responsible than continuing to repair an unsuitable conversation.

7. Compare versions deliberately

The newest version is not automatically the best. A longer version is not automatically deeper, and a more confident version is not automatically more accurate.

Compare the first and revised responses using the same purpose and criteria.

Ask:

• Did the targeted weakness improve?

• Was important meaning preserved?

• Did the response become more useful?

• Did it introduce new assumptions?

• Did it add unsupported claims?

• Is the language suitable for the intended reader?

• Do I still understand and control the work?

If the task involves people, rights, health, safety, assessment, or other important consequences, also consider whether the revision introduced new risks or unfair assumptions.

You may decide to:

• use part of the revised version;

• combine useful parts from different versions;

• revise the material yourself;

• verify a claim externally;

• restart;

• reject both versions.

The final choice belongs to the person using the output.

8. Know when to stop

Follow-up prompting should have an endpoint.

A stop condition is a rule for ending the interaction or moving to another source of help.

Stop when:

• the original purpose has been met;

• the selected weakness has been corrected;

• further changes no longer improve the result;

• evidence is unavailable;

• the conversation is creating confusion;

• the risk is too high;

• a qualified person is needed;

• the learner is becoming dependent on AI instead of thinking independently.

Example:

“If the central claim still lacks a verifiable source after one follow-up, stop prompting and check an external source.”

Stopping protects time, judgement, and responsibility.

9. Record the improvement process

A revision log shows what changed and why.

Version	Main weakness	Next action	What changed?	Decision
First response
Follow-up 1
Follow-up 2 or external check

The log does not need to include every word of a long conversation. It should record the parts needed to explain your decisions.

A useful log can show:

• the original purpose;

• the first prompt and response;

• the weakness you selected;

• the follow-up you wrote;

• any external check;

• what improved;

• what remained weak;

• what you accepted, revised, or rejected.

Saving only the final polished text does not show how learning and judgement occurred.

Classroom activity: The improvement cycle

Use the RUKF prompt and first response you saved from Lesson 6.

Step 1: Evaluate the first response

Mark parts of the response as:

• useful;

• unclear;

• unsupported;

• missing;

• inappropriate.

Step 2: Select one priority

Choose the weakness that matters most for your original purpose.

Explain briefly why it has priority.

Step 3: Write one targeted follow-up

Use:

Target + requested change + condition for success

Save the new response.

Step 4: Compare the two versions

Ask:

• Did the selected weakness improve?

• Was important meaning preserved?

• Were new problems introduced?

Step 5: Choose the next action

Choose one:

• write another targeted follow-up;

• verify a claim externally;

• restart with a better RUKF prompt;

• stop because the criteria are met;

• stop because another source or qualified person is needed.

If your task contains an important factual claim, verify at least one claim outside the chatbot.

If your task is creative or language-based, identify what would require external verification in a factual task.

Step 6: Complete the revision log

Record the first response, your follow-up, what changed, and your decision.

Step 7: Write a reflection

In approximately 250–400 words, explain:

• which weakness you selected;

• why it mattered;

• what follow-up or other action you chose;

• whether the result improved;

• what remained your responsibility;

• why you continued, restarted, verified, or stopped.

Reflection questions

1. Why should evaluation come before a follow-up?

2. What is the difference between unclear and unsupported content?

3. Why should you choose one priority at a time?

4. What makes a follow-up targeted?

5. When should you ask for clarification?

6. What is the difference between a longer answer and a deeper answer?

7. Why can an alternative view create a false impression of balance?

8. Why is asking AI to check itself not independent verification?

9. When is restarting better than continuing?

10. What stop condition would be useful in your own AI work?

Key vocabulary

Follow-up:
A new prompt written after evaluating an earlier response.

Targeted change:
One specific and observable improvement requested in a follow-up.

Clarification:
A request for clearer meaning, definition, distinction, or example.

Depth:
A request for more useful explanation of factors, assumptions, evidence, conditions, or consequences.

Challenge:
A request that tests an assumption, conclusion, limitation, or alternative explanation.

Reformatting:
Changing the structure of a response to make it easier to understand, compare, or use.

External verification:
Checking a claim outside the chatbot using appropriate sources, tools, or qualified people.

Restart:
Beginning a new interaction when the original purpose, prompt, context, or conversation is unsuitable.

Stop condition:
A rule for ending the interaction or moving to another source of help.

Revision log:
A record of responses, weaknesses, follow-ups, checks, changes, and human decisions.

Summary

In Week 7, we learned that improvement begins with evaluation.

A useful follow-up identifies one important weakness and requests a visible change. Depending on the problem, the learner may clarify, deepen, challenge, reformat, or turn the response into practice.

But another prompt is not always the right answer. Unsupported factual claims require external verification. A faulty foundation may require a restart. High-risk situations may require a qualified person. When the purpose has been met, it is time to stop.

The goal is not endless prompting.

The goal is stronger judgement: evaluate the response, choose the right next action, compare the result, and explain your final decision.

Lesson 7 Interactive Quiz: Follow-up Questions and Improvement

Choose one answer for each question. Then select Check my answers. You will receive feedback for every question. This practice quiz does not collect names or scores.

1. What should happen before writing a follow-up prompt?

A. Ask AI to produce a longer version. B. Change several parts of the response at once. C. Evaluate the first response and identify its most important weakness. D. Accept the response if it looks polished.

2. An AI response gives a precise statistic but provides no reliable evidence. How should that part be classified?

A. Unsupported B. Unclear C. Missing D. Useful because it is precise

3. Why should a learner usually target one important weakness at a time?

A. Because AI can change only one word in each response. B. Because the remaining weaknesses no longer matter. C. Because one change guarantees that the result will be accurate. D. Because the effect of the change becomes easier to inspect.

4. Which follow-up is the most targeted?

A. “Improve the explanation and make it more interesting.” B. “Explain the distinction in paragraph two using one workplace example. Preserve the difference between correlation and causation.” C. “Rewrite everything, add more detail, and change the conclusion.” D. “Try again using a more confident tone.”

5. An explanation is clear but too superficial. Which follow-up move is most suitable?

A. Reformat it without changing the content. B. Ask for simpler vocabulary only. C. Ask for the key factors, assumptions, and evidence supporting the conclusion. D. Ask for the same answer with more confident wording.

6. Three useful alternatives are described in a confusing paragraph. What is the best first follow-up?

A. “Present the alternatives in a table using the same criteria for each.” B. “Add a fourth alternative even if it is not relevant.” C. “Choose the best alternative without explaining the criteria.” D. “Repeat the paragraph using more technical vocabulary.”

7. AI names a new Swedish regulation that may affect an assignment. What should the learner do?

A. Ask AI whether it is completely certain and accept the answer. B. Ask AI to repeat the regulation in more formal language. C. Accept the claim if AI provides a detailed explanation. D. Check the regulation through the relevant official source.

8. The first prompt asked AI to write a finished essay, but the learner actually wanted practice questions. What is the best next action?

A. Keep revising the essay until it contains questions. B. Restart with a prompt that reflects the real learning purpose. C. Accept the essay because the AI has already completed it. D. Ask AI to make the essay longer before changing direction.

9. How should a learner decide whether a revised response is better?

A. Select the newest version because it includes the latest changes. B. Select the longest version because it contains more information. C. Compare both versions against the same purpose and criteria. D. Select the version with the most confident language.

10. The response now meets the learner’s purpose and agreed criteria. What should happen next?

A. Stop prompting, record the decision, and complete the work using human judgement. B. Continue prompting until the AI produces a completely different version. C. Ask for additional revisions even if they serve no clear purpose. D. Submit the AI response without reviewing or adapting it.

Reflection: Think of one AI response you recently received. Would the best next action be to clarify, deepen, challenge, reformat, verify externally, restart, or stop? Explain why.