You want to use generative AI to draft something. A good starting point is a clearly scoped task: What preparation is needed, which inputs are approved for it, and how will the result be reviewed? A clear request alone does not guarantee a useful result.
Your goal for this article
You write a small preparatory task with a clear endpoint and explicitly leave open any missing information.
Understand: Put the work question in the right context
The BSI describes generative AI as models that can create new content, such as text or images, based on patterns in training data. Being able to create a draft is not the same as having approval to use it without review.
The BSI also points to IT security risks associated with generative AI and the importance of systematic risk assessment. For business use, approved tools, suitable inputs and responsible reviewers are therefore needed. The steps below are a practical template, not a security certification.
A worked example: Preparing a fictional agenda
You want a short agenda drafted for a fictional internal exercise. You use only uncritical sample data you created yourself as input. The draft should organize the existing topics without adding participant names or binding commitments.
One possible task is: “Arrange these three fictional topics into a short agenda. Use only the information provided. Mark missing details as open.” Then check for yourself whether the order and scope suit the purpose and whether anything not provided has been added.
Decide: Is the task properly prepared?
The purpose and permitted inputs are clear
Limit the output format and scope. Use only a framework approved for this use.
Necessary information is missing
Clarify them or leave them explicitly open. An answer that sounds complete does not make up for missing source material.
You cannot adequately review the result
Do not assign it a binding follow-up task. Bring in the necessary expert review or choose a different process.
A work brief instead of an open-ended generation round
Record the task, permitted source material and desired output format. Specify what must not be added or changed. Do not include access credentials or confidential information that has not been approved for this use.
Decide before generation who will review which parts and when the draft is good enough. Publishing, sharing or taking any other follow-up action is a separate step that requires appropriate authorization, not an automatic continuation of drafting.
Your AI quick briefing
Purpose: What limited preparation is needed?
Basis: Which inputs are approved for this purpose?
Limit: What remains open, and what must not be added?
Review: Who checks the result before it is used?
Next: Review the agreement in everyday practice
A small, reviewable task can be one way to get started. If the inputs or review process have not been clarified, pausing the process is the appropriate decision. You don't have to start every task with AI.
Limitations and support
This article offers general guidance for work, not a complete risk analysis or workplace approval. Use only tools and inputs approved for this purpose. Keep access credentials and confidential information that has not been approved outside the input. Review, use, and any potential follow-up actions have their own responsibilities. If the prerequisites are unclear, pause use for now and clarify with the responsible teams.
Frequently asked questions
Is good wording enough for a safe result?
No. The tool, inputs, review, and intended use are also part of the process.
Does AI need to fill in missing information?
Not with this template. Missing information should remain visible and be clarified appropriately.
Can I have the draft sent automatically?
This template does not provide for automatic sharing. Review and approval by the responsible person remain separate steps.
Expert source and context
BSI: Artificial intelligence - Generative AI section →
accessed on October 16, 2026. German-language source.
The BSI overview page provides context on generative models and their IT security risks. The short briefing and fictional example are independent work templates, not an implementation process reviewed by the BSI.
The practical example and template are practical interpretations, not a scientifically validated program.
Read more on the topic
AI at work - understand the opportunities and use it mindfullyChoose the right way to start with AI based on the task, approval, ability to review, and overall effort
Read the article →
Use AI sensibly in everyday work - without adding complexityLimit one AI step in the existing process with a review and a stopping point
Read the article →Your next step
Choose a specific open work question and clarify it with the relevant people. Find more articles on structure, focus, and mindful digital work in the FitStart Work Guide.
