4 Getting Started: Your First AI Conversation

Begin with a task you can inspect and evidence you can check.

4.1 Choose a safe route first

Complete the first-use data gate before opening a tool. Use only the fictional pilot materials or your own non-sensitive invented example. Institutional approval must cover the account, features and intended use; an enterprise label alone is not permission to upload course or student data.

The exercise works in a browser-based approved service if available. Otherwise, compare the supplied responses below on paper. You do not need a paid account to learn how context and criteria change the usefulness of a request.

4.2 Your First Prompt: The Weak Version

If you have an approved tool, try this request. Otherwise use the authored excerpt below:

Write a case study about business.

The response may be generic, such as a story about a marketing campaign or financial decision. The request supplies little basis for meeting your specific teaching needs. “Business” covers everything. “A case study” could be 100 words or 5,000. The AI had to guess what you wanted.

4.3 Your Second Prompt: A More Specific Version

Now try this instead:

You are an expert marketing lecturer at a
university level.

I need you to create a case study for my
undergraduate students that will help them
practice strategic market analysis and
competitive positioning.

Here are the requirements:
- The scenario should involve a product launch
  decision where the company faces competitive
  threats
- Include enough detail that students need to
  identify which market analysis principles apply
- The scenario should be 300-400 words
- End with three discussion questions requiring
  students to analyse the market strategy and
  positioning

Begin.

The second request gives a clearer task. It may produce more relevant material, but still needs checking. Here are two authored illustrative excerpts, not outputs from a measured comparison:

Generic excerpt: “A business wants to increase sales. It launches a new product. Discuss the marketing strategy.”

More specific excerpt: “A small drink producer can fund only one launch channel. Retail buyers want discounts; direct sales need fulfilment capacity. The producer has no reliable repeat-purchase estimate. Which evidence would you request before choosing a channel?”

The second creates a decision and missing information. It still needs enough supplied data and a criterion for evaluating the student’s response. Do not mistake a more detailed story for a complete classroom activity.

4.4 The CRAFT Framework

What made that second prompt more specific? It included five elements that you can remember with the acronym CRAFT:

Letter	Element	What it does
C	Context	Sets the situation: your course, your students, their level
R	Role	Defines the AI’s perspective: expert lecturer, consultant, practitioner
A	Action	Specifies the task: create, analyse, critique, explain
F	Format	Describes the structure: word count, number of questions, table format
T	Tone/Target	Sets the style and level: professional, introductory, sophisticated

You do not need every element every time. But when a prompt gives you something vague or generic, check which CRAFT elements are missing. The answer is usually context or format.

CRAFT is not the only prompting framework. The companion book Conversation, Not Delegation (available free at books.borck.education) compares several frameworks side by side. Use CRAFT to think intentionally about what you are asking the AI to do, rather than as a rigid formula.

4.5 Conversation, Not One-Shot

CRAFT is your first message, not your only message.

An initial response may miss your teaching context or contain unsupported assumptions. Check it before deciding whether another exchange would help.

The solution is to have a conversation:

PROMPT 1 (CRAFT):
"You are an experienced marketing director.
Create a market analysis case study for my
second-year students. 500 words with 4
discussion questions."

[AI produces a case study]

PROMPT 2 (Refine):
"Good. Make the discussion questions more
specific — ask students to identify which
of Porter's Five Forces apply and what data
they would need to support their recommendation."

PROMPT 3 (Expand):
"Add a section comparing three analytical
frameworks and how each leads to different
conclusions."

PROMPT 4 (Repurpose):
"Turn this into a 10-minute classroom activity
with a 2-minute intro, 5-minute group task,
and 3-minute discussion."

You had a conversation in which each follow-up built on what was working, which is what conversation, not delegation means in practice.

4.6 Discover Your Strategy: Using AI to Help You Use AI

A model can suggest ways to use a tool, but its advice about current features or policy may be wrong. Check those claims against the product’s documentation and your institution’s rules.

Instead of reading generic advice, you can have a personalised consultation where AI interviews you about your teaching context and then recommends both obvious and unexpected applications. Copy this prompt and try it now:

You're an AI expert consultant for business
educators. I'd love your help discovering
how I could better use AI in my teaching.

Please ask me one question at a time. I'll
answer, then you'll ask the next question.
Ask at most five questions using fictional or
non-sensitive context. Stop if I say STOP.
Focus on:

- My discipline and the students I teach
- My main teaching challenges
- My key teaching goals
- My constraints (time, resources, skills)
- What I've already tried with AI

Once you have enough context, provide TWO
types of recommendations:

1. OBVIOUS OPPORTUNITIES: Clear, straightforward
   AI applications most educators think of

2. NON-OBVIOUS OPPORTUNITIES: Unexpected or
   creative uses specific to my discipline
   that I might not have considered

Format your recommendations clearly with
implementation tips for each.

Please start by asking your first question.

Spend 10-15 minutes answering honestly. The recommendations are only part of the value; you also learn about your own teaching by having to explain it. Writing down those constraints can clarify the task whether you discuss them with a model, a colleague or on paper. Treat the model’s recommendations as proposals to evaluate.

Try this (15 minutes)

Run the consultation prompt above. Answer the questions honestly, and don’t polish your responses. Choose one proposal to evaluate against the outcome, data gate, access requirements and staff time. Trying it in class is a separate decision. You can carry out the same planning exercise on paper.

4.7 Activity: inspect the proposed task

Allow ten minutes; no AI required. Compare the two supplied excerpts. Name one decision the learner must make, one missing input and one criterion you would use to assess the response. Then adapt the stronger request to your own discipline without adding real personal or confidential information.

The pilot pack supplies the next step: fixed inputs and worked checks. A usable prompt is only the beginning of a usable learning activity.

# Getting Started: Your First AI Conversation > Begin with a task you can inspect and evidence you can check. ## Choose a safe route first Complete the [first-use data gate](ethics-integrity.qmd) before opening a tool. Use only the fictional pilot materials or your own non-sensitive invented example. Institutional approval must cover the account, features and intended use; an enterprise label alone is not permission to upload course or student data. The exercise works in a browser-based approved service if available. Otherwise, compare the supplied responses below on paper. You do not need a paid account to learn how context and criteria change the usefulness of a request. ## Your First Prompt: The Weak Version If you have an approved tool, try this request. Otherwise use the authored excerpt below: ``` Write a case study about business. ``` The response may be generic, such as a story about a marketing campaign or financial decision. The request supplies little basis for meeting your specific teaching needs. "Business" covers everything. "A case study" could be 100 words or 5,000. The AI had to guess what you wanted. ## Your Second Prompt: A More Specific Version Now try this instead: ```text You are an expert marketing lecturer at a university level. I need you to create a case study for my undergraduate students that will help them practice strategic market analysis and competitive positioning. Here are the requirements: - The scenario should involve a product launch decision where the company faces competitive threats - Include enough detail that students need to identify which market analysis principles apply - The scenario should be 300-400 words - End with three discussion questions requiring students to analyse the market strategy and positioning Begin. ``` The second request gives a clearer task. It may produce more relevant material, but still needs checking. Here are two **authored illustrative excerpts**, not outputs from a measured comparison: **Generic excerpt:** “A business wants to increase sales. It launches a new product. Discuss the marketing strategy.” **More specific excerpt:** “A small drink producer can fund only one launch channel. Retail buyers want discounts; direct sales need fulfilment capacity. The producer has no reliable repeat-purchase estimate. Which evidence would you request before choosing a channel?” The second creates a decision and missing information. It still needs enough supplied data and a criterion for evaluating the student's response. Do not mistake a more detailed story for a complete classroom activity. ## The CRAFT Framework What made that second prompt more specific? It included five elements that you can remember with the acronym **CRAFT**: | Letter | Element | What it does | |--------|---------|--------------| | **C** | **Context** | Sets the situation: your course, your students, their level | | **R** | **Role** | Defines the AI's perspective: expert lecturer, consultant, practitioner | | **A** | **Action** | Specifies the task: create, analyse, critique, explain | | **F** | **Format** | Describes the structure: word count, number of questions, table format | | **T** | **Tone/Target** | Sets the style and level: professional, introductory, sophisticated | You do not need every element every time. But when a prompt gives you something vague or generic, check which CRAFT elements are missing. The answer is usually context or format. **CRAFT is not the only prompting framework.** The companion book *Conversation, Not Delegation* (available free at books.borck.education) compares several frameworks side by side. Use CRAFT to think intentionally about what you are asking the AI to do, rather than as a rigid formula. ## Conversation, Not One-Shot **CRAFT is your first message, not your only message.** An initial response may miss your teaching context or contain unsupported assumptions. Check it before deciding whether another exchange would help. The solution is to have a conversation: ``` PROMPT 1 (CRAFT): "You are an experienced marketing director. Create a market analysis case study for my second-year students. 500 words with 4 discussion questions." [AI produces a case study] PROMPT 2 (Refine): "Good. Make the discussion questions more specific — ask students to identify which of Porter's Five Forces apply and what data they would need to support their recommendation." PROMPT 3 (Expand): "Add a section comparing three analytical frameworks and how each leads to different conclusions." PROMPT 4 (Repurpose): "Turn this into a 10-minute classroom activity with a 2-minute intro, 5-minute group task, and 3-minute discussion." ``` You had a conversation in which each follow-up built on what was working, which is what *conversation, not delegation* means in practice. ## Discover Your Strategy: Using AI to Help You Use AI A model can suggest ways to use a tool, but its advice about current features or policy may be wrong. Check those claims against the product's documentation and your institution's rules. Instead of reading generic advice, you can have a personalised consultation where AI interviews you about your teaching context and then recommends both obvious and unexpected applications. Copy this prompt and try it now: ``` You're an AI expert consultant for business educators. I'd love your help discovering how I could better use AI in my teaching. Please ask me one question at a time. I'll answer, then you'll ask the next question. Ask at most five questions using fictional or non-sensitive context. Stop if I say STOP. Focus on: - My discipline and the students I teach - My main teaching challenges - My key teaching goals - My constraints (time, resources, skills) - What I've already tried with AI Once you have enough context, provide TWO types of recommendations: 1. OBVIOUS OPPORTUNITIES: Clear, straightforward AI applications most educators think of 2. NON-OBVIOUS OPPORTUNITIES: Unexpected or creative uses specific to my discipline that I might not have considered Format your recommendations clearly with implementation tips for each. Please start by asking your first question. ``` Spend 10-15 minutes answering honestly. The recommendations are only part of the value; you also learn about your own teaching by having to explain it. Writing down those constraints can clarify the task whether you discuss them with a model, a colleague or on paper. Treat the model's recommendations as proposals to evaluate. ::: {.callout-tip title="Try this (15 minutes)"} Run the consultation prompt above. Answer the questions honestly, and don't polish your responses. Choose one proposal to evaluate against the outcome, data gate, access requirements and staff time. Trying it in class is a separate decision. You can carry out the same planning exercise on paper. ::: ## Activity: inspect the proposed task **Allow ten minutes; no AI required.** Compare the two supplied excerpts. Name one decision the learner must make, one missing input and one criterion you would use to assess the response. Then adapt the stronger request to your own discipline without adding real personal or confidential information. The [pilot pack](../appendices/pilot-pack.qmd) supplies the next step: fixed inputs and worked checks. A usable prompt is only the beginning of a usable learning activity.