Getting Started with AI: Your First Prompt and Two Workshops

What you will learn

This course needs no LearnIA account. You will write a prompt, review an answer, request a small JSON object and complete two workshops. Copy the instructions into an authorised assistant. Site activities support practice and self-assessment; they do not constitute a qualification.

An assistant can provide a useful draft and a false claim in the same answer. Start with fictional situations and a reference you can check. File and search features depend on the application and account.

Five components of a useful prompt

Our teaching checklist separates five elements. It is a practical way to make a request verifiable, rather than a standard imposed on every model.

“Write a good email” leaves many choices unstated. Supplying the audience, available facts and output format makes errors easier to identify. A role can guide tone; it does not establish verified expertise.

Workshop 1: write a professional email

Fictional situation: Alex is preparing a 30-minute team meeting. November 19 at 10 am is proposed but not confirmed. The email should request agreement and say an agenda will follow.

Help me write a professional email to my team.
Context: a 30-minute meeting, proposed for November 19 at 10 am.
Task: request confirmation and say an agenda will be sent later.
Format: a subject line followed by a message of at most 100 words.
Constraints: do not present the slot as confirmed. Invent no venue,
meeting link or participant. Flag missing information.

Review four points: the date remains a proposal, duration is correct, no link is invented and tone suits the team. If one fails, request a targeted correction naming the fact to preserve. Then review the complete message again.

Structure output as JSON

JSON represents data as keys and values. It helps software read a response, but text resembling JSON can be invalid. Valid JSON can also contain incorrect facts.

For this situation, a reference output is:

{
  "proposed_date": "November 19",
  "time": "10:00",
  "duration_minutes": 30,
  "confirmed": false,
  "venue": null
}

Ask for those fields without adding missing facts. Check keys, types and values. Duration is a number, confirmation is a Boolean and the unknown venue is null. Do not invent a year: none was supplied.

For software integration, use an appropriate parser and schema. OpenAI structured outputs illustrate an API feature distinct from merely asking for JSON. Schema compliance still does not replace factual verification.

Workshop 2: summarise, then propose next steps

Use these fictional notes: “Support requests an FAQ. Alex proposes a draft on Friday. The manager has not approved the deadline. Budget was not discussed.”

First request a table with fact, named person and confirmed or unconfirmed status. Verify that Friday remains a proposal and no budget appears. Then request actions based only on the checked table, listing information needed before a decision.

Separating steps makes review easier but does not guarantee accuracy. An error in the first table can propagate. Keep a review point between steps and allow the intermediate source to be corrected.

Verify, revise and keep a record

Separate supplied facts, model suggestions and missing information. Open sources when external claims matter. Recalculate important numbers independently and compare results with the original inputs.

Apply your organisation's rules before sharing confidential information. Fictional examples are sufficient for this exercise. ChatGPT's official guide and Claude's file documentation explain product features, not guarantees of correct results.

Download the first prompt worksheet. Record the initial instruction, observed error, correction and retained version. This is more informative than collecting prompts without checked outcomes.

Continue learning

The LLM foundations guide explores context and generation. The structured outputs guide develops validation. To choose tools, start with our current model comparison and identical test inputs.

Help me write a professional email prompt using fictional facts. Ask for audience, context, task, format and constraints. Do not complete missing information. Then provide a review checklist for the draft.