Finaudax
← All chaptersChapter 1 of 8

From prompt to professional workflow

You turn a single interaction into a clearly defined task that can be checked and repeated.

After this chapterFor a complex task, you can design a process with a clear result, appropriate context, intermediate products, checks, and an explicit final decision.
Your route through this level

use one fictional task for your workflow, prompt tests and final project. Complete the lessons, knowledge checks and two guided practical workshops; build the Linde work step by step only during chapter 5. You do not need your own business environment. Applying the method in a second field, an actual peer review and available extra tools are extensions. The final project requires demonstrable work against the rubric; a simulated business trial is labelled a simulation and does not authorise real deployment.

Your progress0 of 40 lessons
1.1

A workflow starts with the result

First define what needs to be done and who will use the result.

An advanced workflow does not start with a long prompt, but with an observable outcome. Describe the decision or usage moment, the target audience, and the characteristics the end product must meet. This prevents an impressive text from turning out to be unusable.

Distinguish between the result and the activity. 'Analyze customer feedback' is an activity; 'deliver a decision memo with the three most substantiated improvement priorities, evidence, and open questions' is a result. Only add process steps when they improve control, safety, or quality.

Before you begin: you can already write a clear prompt, refine an answer and check an important claim. Practise here with invented data. The examples provide all the input you need; you do not need a paid subscription or a connected business environment to practise the method on paper or in an ordinary chat. Some product features do depend on your account.

  • Result before activity
  • Audience and intended use
  • Two to five success criteria
  • Explicit stop or approval moment

Terms in plain language

Workflow
An agreed sequence of steps from input to a checked result.
Acceptance criterion
An observable requirement used to assess whether the result is usable.
How you can use this

Work: a management memo must identify the decisions to be made. Study: a study sheet must reveal misconceptions. Personal use: a comparison must stay within your budget and preferences.

Try this prompt
Result: [final product]. User: [audience]. It is useful if it meets: [criteria]. Ask no more than three questions about missing essential information, then propose a workflow. Do not create the final product yet.
Knowledge check

Your study group must choose three topics for its next practice session. Which deliverable makes that choice easiest to assess?

Your practical exercise

Rewrite one recurring task as a result, specifying the audience, intended use and three success criteria.

1.2

Context as managed input

Treat context as selected source information, not as an unlimited dump of information.

Context has value when it can demonstrably change the answer. Label source facts, internal agreements, preferences, examples, and yet unconfirmed assumptions separately. This reduces the risk that a hypothesis later appears as established fact.

For long-term work, use a project when available features allow it. A project can keep related chats, files, instructions, and connected resources together. Within that project, start a separate chat for each clear result. Regularly check which resources are actually available, up-to-date, and allowed.

  • Separate source facts and assumptions
  • Only relevant context
  • Version and date of sources
  • Separate chat per result

Terms in plain language

Context register
A list of the sources, versions, agreements and unconfirmed assumptions being used.
How you can use this

A policy analysis uses the current document as a source, an older version for comparison, and individual staff notes only as signals to be verified.

Try this prompt
Inventory the necessary context for [task]. Make four categories: reliable sources, internal rules, preferences, and unconfirmed assumptions. Report conflicts and missing core information before you analyze.
Knowledge check

An approved schedule says 19:00. A newer note says 19:30 but does not mention approval. You are preparing the invitation. How do you establish the correct time?

Your practical exercise

Create a context register for a non-sensitive task of your own or the fictional study group in the chapter example. Put error counts under source facts, missing question counts under unknowns, and the single practice evening under agreements. Done: every retained item has a source or clear status, and you explain which decision it could change.

Source for this lesson

OpenAI – Projects
Projects organise context; check the available sources and settings.
Checked: 2026-09-07

1.3

Divide tasks into phases you can check

Break down complex work around decisions and evidence, not to produce as many steps as possible.

Decomposition makes a task manageable. For example, use: exploring, selecting, elaborating, checking, and finishing. Each phase produces an intermediate product that you can evaluate before the next phase begins.

Make dependencies visible. A recommendation can only follow after facts and assumptions have been separated; a publication can only follow after claims and rights have been checked. Let ChatGPT first propose a plan, but assess yourself whether the sequence is logical and whether an early error could multiply later.

  • Phases with their own goal
  • Intermediate product per phase
  • Make dependencies visible
  • Stop if there is insufficient evidence

Terms in plain language

Decomposition
Breaking a complex task into smaller tasks with their own checkpoints.
How you can use this

For research: question definition, source plan, source extraction, comparison, claim check and only then conclusion.

Try this prompt
Divide [complex task] into a maximum of six phases. For each phase, provide: goal, required input, intermediate product, quality control, responsible party, and stopping condition. Mark dependencies.
Knowledge check

You are comparing three courses and must then present one choice. Their entry requirements are still unknown. Which sequence limits work on unsuitable options?

Your practical exercise

Plan the phases of a task that is usually carried out with one long prompt.

1.4

Checkpoints and acceptance criteria

A checkpoint states what is being assessed, by whom, and what happens in case of a deviation.

Vague quality terms such as professional, complete or correct are difficult to apply consistently. Translate them into criteria a reviewer can observe: every claim has a source, every action item has an owner, or amounts match the source table.

Combine substantive, formal, and risk criteria. Automate simple checks where possible, but do not treat model self-check as independent evidence. For high impact, a qualified human assessment is required. Also determine the remedial action: supplement, return to a previous phase, escalate, or stop.

  • Observable criteria
  • Reviewer and evidence
  • Recovery action
  • Stricter control with higher impact
How you can use this

A report review checks source coverage, the accuracy of copied figures, logical inferences, suitability for the audience and confidentiality separately.

Try this prompt
Create an acceptance table for [intermediate product] with criterion, test method, evidence, limit value, owner, and corrective action. Do not use criteria that only express style preference.
Knowledge check

An action list must be usable by someone taking over the work. Which acceptance rule lets a reviewer identify missing responsibility and have it resolved?

Your practical exercise

Write five testable criteria for a proposal from the fictional study group: correct counts, correct source, unknown denominators disclosed, no unsupported ranking and a feasible next step. If possible, ask someone to interpret them. Working alone: write one text that meets the criteria and a copy that only conceals the unknown denominators. Assess both in a separate round. Done: the affected criterion fails and you point to the text that proves this; this remains self-assessment.

1.5

Choose the right working method

Use Chat for dialogue and short iteration; use a more extensive working mode when resources, tools, and a reviewable file come together.

Not every task requires the same working method. A quick rewrite or brainstorming fits in a regular chat. A task with multiple sources, different steps, files, or a larger deliverable fits better in ChatGPT Work when that is available. A project helps when chats need to share the same controlled sources and instructions.

Availability depends on environment and settings. Therefore, design based on the required result and the allowed data, not on a single button. Specify in advance which actions may only be prepared as drafts and where explicit approval is required.

  • Chat for quick dialogue
  • Work for larger deliverables
  • Project for shared context
  • Plugins only with appropriate permission

Terms in plain language

Deliverable
The concrete result you hand over, such as a checked report or file.
How you can use this

A team update from Drive and Slack sources requires a different working method than shortening a single paragraph.

Try this prompt
Advise the appropriate working method for [task]. Compare regular chat, project, and ChatGPT Work on sources, steps, files, review, privacy, and required approval. Take into account possible availability limitations.
Knowledge check

You are creating four separate learning resources from the same regularly updated course sources. Projects are available. Which setup best supports shared source versions and focused review?

Your practical exercise

Classify five of your own tasks and justify your chosen way of working for each one.

Source for this lesson

OpenAI – Prompting
Task, context, output and boundaries; the choice of working method depends on the task and availability.
Checked: 2026-09-07

Worked example

A study dossier that helps you decide what to practise

Fictional practice material; incorrect answers have been created deliberately for this exercise.

An entirely fictional study case. Three practice chapters are called A, B and C. Your study group has one evening to practise. The available exercise results record two errors for A, six for B and one for C; the number of questions completed in each chapter is missing.

Input

Make a proposal for our practice evening. Provide no more than three priorities, with the data used, a limitation and a next step. Use only the exercise results provided. Plan the approach first: take stock, compare, propose, check.

First practice answer

B is clearly the most difficult chapter, followed by A and C. Spend almost all the time on B. This deliberately flawed answer treats error counts as if they were comparable error rates.

Check

The numbers of completed questions are missing. B may have contained many more questions. Put that missing input in the context register. Acceptance criterion: no ranking by error rate without a denominator. Revision: keep the three error counts, remove the categorical ranking and work out what we can already do.

Improved version

Proposal: first collect the number of completed questions for each chapter, then examine the types of errors. For now, bring one example of an error from each chapter to the practice evening. B has the largest absolute number of recorded errors, but relative difficulty is still unknown. The group will decide how to allocate the remaining practice time only after the missing information has been added.

Try it yourself

New fictional input: A has 2 errors out of 10 questions, B has 6 out of 60 and C has 1 out of 10. Calculate the percentages and state a priority with appropriate limitations.

View the model answer

A: 20%; B: 10%; C: 10%. A deserves attention on this measure, but question difficulty and error types may differ. Check those differences before allocating all the practice time. A correct percentage is an intermediate result, not complete study advice.

Chapter assignment

Bring everything together

Design a workflow canvas for one complex task with result, context register, phases, intermediate products, acceptance criteria, stop conditions, and final approval.

Maximum 10,000 characters per note.

Progress and notes are stored only in this browser on this device. Do not enter sensitive data. Download your notes regularly. This course sets no automatic expiry date. You can delete the data through your browser’s site-data settings; export anything you wish to keep first. Browser settings or cleanup may erase it earlier. These local notes are not sent to Finaudax.