Creating a digital manual with AI: how to avoid mistakes
Creating a digital manual with AI seems easy, but a quick first draft is not always a usable work instruction. Without clear source information, review and agreements, your team will actually end up with more questions and mistakes.

Where do things often go wrong when gathering source material?
The first mistake is often made before you even start using AI. Information is scattered across old documents, separate emails, notes and the knowledge in experienced colleagues’ heads. If those sources contradict each other, AI will not turn them into a reliable manual. It mainly summarises what you give it.
So start by defining one single process. Decide who the instruction is for, what result someone needs to achieve and which steps are involved. Have an employee who does the work every day check the source information. This prevents you from describing a way of working that has already changed in practice.
A second mistake is entering confidential information without thinking. The Dutch Data Protection Authority has received reports of data breaches caused by employees entering personal data into an AI chatbot. So work with anonymised examples and remove names, addresses, customer numbers and other sensitive data from your source material.
How do you give AI a clear instruction?
A vague prompt produces a vague manual. If you only ask for a work instruction, the text may be too general, miss important exceptions or use words your team doesn’t know. So always provide context about the target audience, their level of knowledge and the structure you want.
Also specify which sections are required. Think of a short preparation section, numbered steps, checks during the work and solutions to common problems. Have AI propose a table of contents first. Review it with a colleague before you have the full text written out. This saves you a lot of time on rewriting.
Use these checkpoints when writing your prompt for AI:
- Is the purpose of the task clear?
- Does the AI know the right target audience?
- Are exceptions and choices described?
- Does every step need to be verifiable?
A good digital manual is not only complete, but also practical to follow. So ask for short sentences, active wording and one action per step. Where needed, add screenshots, examples or a short video. AI can help with the structure of the text, but your organisation decides what someone actually needs to do.
How do you check whether the work instruction is correct?
The biggest mistake is publishing AI-generated text without checking it. AI can produce plausible-sounding information that is incorrect or doesn’t match the way you work. So have the manual tested by someone who doesn’t yet know the process well. That person will quickly see which steps are missing, too abstract or in the wrong order.
Also check whether the text is precise enough. An instruction such as “process the application carefully” doesn’t help anyone. Instead, describe which form someone uses, what information is needed, who carries out the check and what happens if information is missing. This also makes the manual usable when a colleague is absent.
When processing personal data, the GDPR applies. Among other things, organisations must have a valid legal basis, be transparent, use no more data than necessary and take appropriate security measures [B2]. So don’t include personal data in practice examples or screenshots unless it is necessary. Also record who is allowed to edit the manual and when the content will be reviewed again.

How do you keep your manual secure and up to date?
A manual is not a document you create once and then forget about. Processes, forms and responsibilities change. So schedule fixed moments to review the content, for example after a process change or when employees keep asking the same question. Keep track of who made changes and why.
Make clear agreements within your team about using AI. Employees need to know which information they may and may not enter. The Dutch Data Protection Authority advises organisations to make such agreements clear and, where possible, to arrange that entered data is not stored [B1]. This prevents speed from becoming more important than privacy.
In addition, AI literacy has been a point of attention since February 2025 for organisations that use AI systems [B3]. That doesn’t mean everyone has to become a technical specialist. But your team does need to understand what AI can do, where mistakes can arise and when human review is needed. Want to use AI more broadly and practically? Read more about AI consultancy.
Frequently asked questions about creating a digital manual with AI
Can AI write a complete work instruction?
AI can write a good first draft based on your source material and a clear prompt. Always have a subject-matter expert check the steps, exceptions and examples before you share the manual.
Do I need to check every AI text manually?
Yes, especially if the manual affects customers, safety, privacy or financial processes. An independent test by a colleague shows whether the instruction is truly understandable and practical to follow.
Can I use customer information as source material?
Preferably not. The GDPR requires data minimisation, which means you only process personal data that is demonstrably necessary. So anonymise examples beforehand.
A clear digital manual saves time, prevents unnecessary questions and helps new colleagues get started faster. Want to explore where AI can support your processes? We are happy to think it through with you, with both feet on the ground.


