From instructions to results

How do you formulate a request to AI to obtain a usable text and check its quality?

1250 € per session, excl. VAT

In short

This half-day session of 3 hours introduces how generative AI responds to an instruction. Using a fictional case, participants build a prompt with an objective, context, data, constraints and format, then compare two results to produce a practical message that has been reviewed and corrected.

Why this training?

A request such as ‘write a welcome message’ leaves the AI to choose the recipient, information and length. Making these elements explicit reduces ambiguity and helps explain why a response can seem convincing yet remain unusable. A structured prompt does not remove the need for review. The training therefore links each instruction to an observable check: fidelity to the notes, compliance with the format, no added factual information and inclusion of the information the recipient needs. Case used throughout: The fictional association Avelune Partage is preparing to welcome new volunteers to its repair workshop. Using the notes provided, participants write a welcome message of 120 to 150 words containing practical information and a request for confirmation. The finish time is deliberately missing: this must be flagged, never invented. Application at work: In the following week, reuse the template for a simple request: rephrasing an instruction or preparing an internal message. Use only authorised data, retain both versions of the prompt and apply the assessment grid before sharing anything.

Objectives of this training

  • Identify ambiguities in an overly brief request.
  • Distinguish between instructions, source data and the expected format.
  • Write a structured prompt with five sections.
  • Compare two responses using a quality assessment grid.
  • Correct an instruction and review the resulting message.

Who is it for and what are the prerequisites?

Roles and tasks concerned

  • Employees new to AI who wish to prepare professional messages using generative AI.
  • Managers and support staff who wish to understand how to frame an initial request.

Prior skills

  • No prior AI knowledge required.
  • Ability to enter text and use a browser.

What you will work on

  1. Examining an ambiguous request

    Objective: Identify what the AI has to guess when details are missing. The principle of a generated response, variation in wording and the difference between plausible text and verified information.

  2. Assembling the five sections

    Objective: Turn the notes into an explicit request. Objective, context, source data, constraints and format; a visible separation between instructions and notes.

  3. Testing and then adjusting an instruction

    Objective: Link a flaw in a response to a specific change. Fact-checking, measurable constraints and targeted rephrasing; the distinction between correcting the prompt and editing the text.

  4. Assessing and finalising the message

    Objective: Decide whether the result can be used after review. Grid with four dimensions: fidelity, completeness, format and tone; each dimension is scored from 0 to 2 with a justification.

What you take away

  • Five-section prompt template with completion guidance.
  • Initial prompt and corrected version, accompanied by the responses obtained.
  • Reviewed welcome message, ready for use in the fictional case.
  • Completed comparison grid and an outstanding point to confirm.

Method and assessment

  • Brief explanations followed by immediate practice using the case explored throughout.
  • Individual work followed by work in pairs, with participants reviewing each other’s outputs.
  • Guided group discussion: what works, what needs correcting and what remains a human decision.
  • Preparation: Arrange access to a generative AI tool authorised by the organisation, in accordance with its usage rules. Read the fictional case file provided; any personal example proposed must be anonymised and authorised.

Observable criteria: The prompt specifies the objective, context, data, constraints and format. ; The message meets the length requirement of 120 to 150 words and requests confirmation. ; Every factual detail in the message is present in the source notes. ; The missing finish time is flagged outside the message, without inventing a value. ; The comparison assigns a score and justifies at least one correction based on an observed discrepancy..

Frequently asked questions

Do you need to know technical terminology?

No. A prompt is a written instruction; the five sections are explained and then completed progressively.

Which tool is used?

A generative AI tool authorised by the organisation. The template does not depend on a particular brand.

Can you use internal documents?

The exercises are based on the fictional case file. Any other example must be anonymised and comply with the organisation’s rules.

Is a precise prompt enough to obtain an accurate text?

No. You must always check the response against the sources and address missing information before sharing it.

Last updated: