Prompts ready for reuse

How can you turn a one-off instruction into a prompt template that can be used in several situations without compromising quality?

1250 € per session, excl. VAT

In short

In a half-day session of 3 hours participants turn common requests into prompt templates with adjustable parameters. They distinguish fixed instructions from variables, test several inputs, resolve ambiguities and document the conditions of use. They leave with two templates and a completed checklist.

Why this training?

An instruction that works once is not necessarily reusable. Implicit references, overly specific examples and missing information make responses unreliable as soon as the context changes. Separating stable instructions from variable data allows a prompt to be reused without rewriting everything. Checking several inputs helps identify recurring flaws rather than focusing solely on one convincing response. Running case study: The fictitious organisation Avelune Signalétique wants to standardise its meeting summaries and follow-up messages. Participants have three sets of fictitious notes: one complete, one ambiguous and one incomplete. The prompts must work without inventing any decision, person responsible or deadline. Transfer to the workplace: In the following week, adapt a template to an everyday task using anonymised data and an approved tool. Assess three results using the same assessment grid, then record a correction and its rationale in the prompt's version history.

Objectives of this training

  • Identify the fixed and variable elements of a recurring request.
  • Write two templates with explicit parameters and a defined response format.
  • Test a template on complete, ambiguous and incomplete inputs.
  • Measure response quality using a shared assessment grid.
  • Document a prompt's limitations and rules for reuse.

Who is it for and what are the prerequisites?

Roles and tasks concerned

  • Staff who use AI to summarise notes and write recurring messages.
  • Assistants, coordinators and staff in support roles who wish to share documented instructions.

Prior skills

  • Have already formulated simple requests in a generative AI assistant.
  • Be able to describe a recurring writing task and recognise a usable response.

What you will work on

  1. Distinguishing an instruction from a template

    Objective: Identify what must remain stable from one use to the next. Anatomy of a prompt: task, relevant context, source data, constraints and output. Difference between a variable and an instruction.

  2. Building two self-contained templates

    Objective: Make the instructions usable without additional verbal explanation. Defining source boundaries, mandatory fields, expected format and handling missing information. Example parameter: [RECIPIENT].

  3. Testing variations and limitations

    Objective: Compare responses against explicit criteria. A set of contrasting tests, shared scoring and response variability. Changing one instruction at a time to interpret differences.

  4. Organising reuse

    Objective: Hand over a prompt together with its conditions of use. Name, version, variables, known limitations and human oversight. A shared prompt must not depend on an invisible conversation history.

What you take away

  • Two templates with adjustable parameters: meeting summary and follow-up message.
  • Variable dictionary with example values.
  • Completed comparative test grid.
  • User guide and change history.

Method and assessment

  • Short teaching inputs followed by immediate practice on the running case study.
  • Individual work followed by work in pairs, with reciprocal review of the outputs.
  • Guided group discussion: what works, what needs correcting and what remains a human decision.
  • Preparation: Have access to a generative AI tool approved by the organisation. Prepare a routinely used instruction and a fictitious or anonymised example, without confidential information.

Observable criteria: Variables are identifiable, explained and separated from fixed instructions. ; Both templates specify the task, sources, constraints and expected format. ; The main template is tested on three different inputs, including one incomplete input. ; Each response is assessed on four dimensions scored from 0 to 2: fidelity, completeness, format and handling of unknowns. ; A revision is justified by the discrepancies observed; any invented information remains flagged as a blocking defect..

Frequently asked questions

Do you need to know how to code?

No. The templates are written as text and their variables are replaced manually.

Does a template work with all assistants?

Not necessarily. Any change of tool or model requires the relevant tests to be repeated.

Can you use internal documents?

The exercises use only fictitious or anonymised cases, within tools approved by the organisation.

Does a good prompt remove the need for review?

No. It makes checking easier, but facts, omissions and commitments must always be verified.

Last updated: