This page needs a newer browser to run its tools. You can still read everything here.

Prompting: from a first try to reliable results interview questions

Interview questions on “Prompting: from a first try to reliable results”: warm-ups, then what each harder question tests and how a good answer starts.

12 questions: 4 warm-ups and 8 with model answers · Levels: Fresher to senior

Warm-up questions

Short questions with short answers, each answered in full.

Fresher to mid

What makes a prompt clear?

Answered in the lesson The anatomy of a clear prompt
Fresher to mid

What does grounding an AI answer mean, and why does it make answers easier to check?

Answered in the lesson Giving AI your sources and context
Fresher to mid

What does a few-shot prompt do, and why should its examples share a format?

Answered in the lesson Examples, roles and output formats
Fresher to mid

Why is an AI's review of its own draft not enough as a final check?

Answered in the lesson Iterating, chaining and self-review

Questions with model answers

What each question tests, the first step of a good answer and the follow-ups to expect.

Fresher to mid

A manager says the team's AI drafts always need three rounds of fixes before they can be sent. Looking at the team's prompts, how would you find the cause, and what would you change first?
What it tests

Diagnosing rework from the parts a brief leaves out, replacing adjectives with checkable limits, and writing a test before reading any reply.

First step

Collect five recent prompts and the fixes each draft needed. Sort every fix by the question it answers: the reader, a missing fact, a length or tone, the layout, or what the result had to achieve. The parts that keep needing fixes are the parts the prompts leave out.
Follow-ups
- The prompts are already long. Why are they still unclear?
- How would you show that the change worked?
- Should the team share one template?
From the lesson The anatomy of a clear prompt
Model answer, Pro. Model answers open with any pass; prices are in the notice on this page.
Opening the model answers with your pass…
Model answers open with any MySmartCoPilot pass.
Pro pass · 1 week₹79$2.99
30 days₹179$5.99
1 year₹1,699$45
One payment; a pass never renews. The price shown is the total you pay: no fees or charges are added at checkout.
Passes can’t be bought right now, so the model answers can’t be opened at the moment. If you have a pass, enter your licence key.
Choose a pass I have a licence key
It also opens every other in-depth part in Learn and every Pro tool. Sales are final, except where the law gives you a right to withdraw, for example in the EU and the UK (refund policy).
In the answer:
- Model answer
- Follow-up answers
- Common mistakes
- Rubric
- Variants
- about 3 minutes
Mid to senior

Twenty support agents will use one AI prompt to draft replies to customers whose deliveries are late. How would you write that shared prompt, and how would you know it works before everyone uses it?
What it tests

Turning the six questions into a reusable prompt with slots, setting checkable limits and a test, and trying it on sample cases before rollout.

First step

Write the brief with slots for what changes per case, such as the order number, the new date and the reason, and fixed lines for what never changes: the reader, the tone, the limits, the layout and the test. Then run it on five real cases with the personal details masked, and check every reply against the test.
Follow-ups
- A customer is angry and threatens to cancel. What changes?
- Agents paste customers' phone numbers into the prompt. What do you do?
- The tool's model is updated next month. What then?
From the lesson The anatomy of a clear prompt
Model answer, Pro.
In the answer:
- Model answer
- Follow-up answers
- Common mistakes
- Rubric
- Variants
- about 3 minutes
Mid to senior

Your HR team wants staff to ask an AI tool questions about the leave policy by pasting the policy into the chat. How would you set this up so that the answers can be trusted, and what would you check?
What it tests

Grounding instructions, quote-first answers, questions the policy does not cover, keeping to the current version, and privacy.

First step

Write one grounded prompt for everyone: answer only from the policy inside the tags, start with the clause number and its exact words, and reply "The policy doesn't say" when it has no answer. Then test it with ten questions whose answers you know, two of them on things the policy does not cover.
Follow-ups
- The policy is 40 pages long. What changes?
- Staff paste their medical details into their questions. What do you do?
- The policy is revised. How do you stop answers from the old version?
From the lesson Giving AI your sources and context
Model answer, Pro.
In the answer:
- Model answer
- Follow-up answers
- Common mistakes
- Rubric
- Variants
- about 3 minutes
Fresher to mid

A colleague pastes a supplier's email into an AI assistant that can send emails and asks it to handle the email. The email has a line asking for your price list to be sent to a new address. What is the risk, and how should the team handle pasted material?
What it tests

Recognising prompt injection in pasted material, separating material from instructions, and keeping a person in charge of actions.

First step

Name what happened: a line in the material is trying to act as an instruction, which is prompt injection. Then separate reading from acting: the assistant may summarise the email and list what it asks for, but sending anything, above all to a new address, needs a person's approval.
Follow-ups
- The new address uses the supplier's own domain. Does that change anything?
- Would tags around the email be enough?
- How would you write the team's rule?
From the lesson Giving AI your sources and context
Model answer, Pro.
In the answer:
- Model answer
- Follow-up answers
- Common mistakes
- Rubric
- Variants
- about 3 minutes
Fresher to mid

The product descriptions your marketing team gets from an AI tool come out in a different structure every time. How would you use examples and format instructions to make them consistent, and how would you check 200 of them?
What it tests

Consistent and varied examples, keeping facts apart from the pattern, naming the format, and checking it with a program instead of by eye.

First step

Fix the shape first: write down the exact structure, such as a name line, two sentences and a care line within 45 words, then write two or three examples that have it and differ in content. Put each new product's facts in a separate part, and check every output against the structure with a script.
Follow-ups
- The examples all start with the same word. Does it matter?
- Would a role such as "expert copywriter" help?
- How do you check 200 descriptions without reading them all?
From the lesson Examples, roles and output formats
Model answer, Pro.
In the answer:
- Model answer
- Follow-up answers
- Common mistakes
- Rubric
- Variants
- about 3 minutes
Mid to senior

A team pastes meeting summaries from an AI tool straight into a spreadsheet import, and one day the import breaks. How would you request and check structured output so that this cannot slip through again?
What it tests

Naming formats exactly, validating tables and JSON with code, required keys as well as valid syntax, and rejecting bad replies instead of importing them.

First step

Find what broke by comparing the failing reply with what the import expects. Then name the format exactly in the prompt, such as a JSON list of objects with the keys decision, owner and due, and put a program between the reply and the import that rejects anything invalid or incomplete.
Follow-ups
- The JSON is valid but one item has no owner. Do you import it?
- Could the AI fix its own JSON?
- What do you record when a reply is rejected?
From the lesson Examples, roles and output formats
Model answer, Pro.
In the answer:
- Model answer
- Follow-up answers
- Common mistakes
- Rubric
- Variants
- about 3 minutes
Fresher to mid

An AI tool produced a 40-row event budget, then "double-checked" it and said it was correct, but the total was wrong. How would you set up the work so that this does not happen again?
What it tests

Why self-review misses arithmetic, splitting the task into a chain, and checking numbers outside the model.

First step

Find where the error came from: rebuild a few lines from the original prices and quantities and compare. Then split the work into steps, the table first and the total second, and put a recomputation by a spreadsheet or a program between them, so that no number goes on unchecked.
Follow-ups
- The tool shows its working for every line. Is that enough?
- Next time the budget has 400 rows. What changes?
- When is a self-review still useful?
From the lesson Iterating, chaining and self-review
Model answer, Pro.
In the answer:
- Model answer
- Follow-up answers
- Common mistakes
- Rubric
- Variants
- about 3 minutes
Mid to senior

Your team keeps a shared document of prompts that work. After the AI tool updates, some of them start giving worse results. How would you organise and test the team's prompt library?
What it tests

Recording prompts with their tool, model, test cases and checks, re-testing after updates, versions, and who owns each prompt.

First step

Turn the document into entries: each prompt with its purpose, the tool and model it was tested on, two or three test cases with the result each must give, and when the tests last passed. Then run every entry's tests on the updated tool, and mark the ones that fail.
Follow-ups
- Who should own each prompt?
- A prompt fails one of its three tests. What do you do?
- How often should the tests run?
From the lesson Iterating, chaining and self-review
Model answer, Pro.
In the answer:
- Model answer
- Follow-up answers
- Common mistakes
- Rubric
- Variants
- about 3 minutes

The lessons of this module

The anatomy of a clear prompt Free lesson, Pro answers: 15 minutes
Giving AI your sources and context Free lesson, Pro answers: 15 minutes
Examples, roles and output formats Free lesson, Pro answers: 15 minutes
Iterating, chaining and self-review Free lesson, Pro answers: 20 minutes

All AI Literacy & Prompting interview questions · The AI Literacy & Prompting track

Warm-up questions

What makes a prompt clear?

What does grounding an AI answer mean, and why does it make answers easier to check?

What does a few-shot prompt do, and why should its examples share a format?

Why is an AI's review of its own draft not enough as a final check?

Questions with model answers

A manager says the team's AI drafts always need three rounds of fixes before they can be sent. Looking at the team's prompts, how would you find the cause, and what would you change first?

Twenty support agents will use one AI prompt to draft replies to customers whose deliveries are late. How would you write that shared prompt, and how would you know it works before everyone uses it?

Your HR team wants staff to ask an AI tool questions about the leave policy by pasting the policy into the chat. How would you set this up so that the answers can be trusted, and what would you check?

A colleague pastes a supplier's email into an AI assistant that can send emails and asks it to handle the email. The email has a line asking for your price list to be sent to a new address. What is the risk, and how should the team handle pasted material?

The product descriptions your marketing team gets from an AI tool come out in a different structure every time. How would you use examples and format instructions to make them consistent, and how would you check 200 of them?

A team pastes meeting summaries from an AI tool straight into a spreadsheet import, and one day the import breaks. How would you request and check structured output so that this cannot slip through again?

An AI tool produced a 40-row event budget, then "double-checked" it and said it was correct, but the total was wrong. How would you set up the work so that this does not happen again?

Your team keeps a shared document of prompts that work. After the AI tool updates, some of them start giving worse results. How would you organise and test the team's prompt library?