Visionary Ventures

Guide · Last updated: 08.11.2026

Learn Prompts: How to Work Safely with AI Instructions

If you want to learn prompts, the programme ‘Safe AI Prompting’ gives you a learning path in five steps. You understand how an AI forms its answers, write clear instructions, set the shape of the answer and check the result. It includes 20 videos (32 minutes), five PDF documents and three audios. Access is included in the membership.

A loose swarm of glass fragments flows into a glowing glass funnel and leaves it on the right as an orderly grid of bright tiles.

At a glance

Learning path
5 steps
Videos
20 (about 32 minutes)
PDF materials
5 with 152 pages in total (length of the English originals)
Audio
3
Language
all PDFs in English and German; videos in English with German and English subtitles; audio in English with a transcript in English and German

Go to the programme «Working with AI Prompts with Confidence»

What do you learn when you learn prompts?

A prompt is the instruction you give to an AI. In the programme you learn to build such instructions so that the answers become useful and checkable: with a clear task, the background that is needed, the shape you want and fitting examples. You also find out why an AI makes mistakes, how to spot them and how to reuse proven instructions as templates.

The thread running through it is an ebook of 78 pages. It first explains how an AI assistant works: it breaks text into small units, called tokens, and guesses the most likely continuation piece by piece. A lot of what surprises people in daily use follows from this. Even small changes in wording lead to a different answer, and tasks such as comparing numbers or counting letters are weak spots.

Then come the tools: show examples instead of only listing rules, let the AI reason step by step, break big tasks into small parts, keep texts apart from instructions and set the format of the answer. The final part covers tests, settings such as temperature (it decides whether answers are predictable or varied) and the differences between AI models.

The videos show the practice on a chat assistant: collecting ideas, sharpening a list of follow-up questions, rewriting texts for another purpose, simplifying difficult texts and checking statements against sources.

Why are AI answers wrong, and how do you check them?

An AI words things convincingly without knowing whether they are true. It calculates which word is likely to come next. So answers can sound plausible and still be made up. The programme teaches a fixed checking routine: ask for sources, open those sources yourself, test statements with a second request and run important instructions several times before you rely on them.

The ebook begins with a real case: an airline’s chatbot gave a customer a refund rule that did not exist. A court still ordered the airline to pay. The lesson runs through the whole programme: if you use AI at work, you remain responsible for the result.

In the videos you see what checking looks like in practice. You first read which limits the provider of the assistant names itself, including answers that sound plausible but are wrong and the tendency to make assumptions instead of asking. Then you ask for sources and web addresses, check whether they exist and ask for statistical evidence. If an answer does not convince you, you reword the question and compare.

With texts, a second look is added. You have the content checked by a spelling and style checker and a plagiarism checker, and you set the tone to fit the purpose, for example formal or easy to understand.

How is the programme structured?

The programme has five steps that build on each other. You start by understanding how an AI forms its answers and finish with the question of how to transfer proven instructions to other AI models. In each step you work with videos, a chapter of the ebook or another PDF and, in some steps, an audio episode.

The order follows one idea: first understand, then shape, then safeguard.

  1. Understand how AI forms answers: You set up a chat assistant, read the limits that the provider itself names and try out sample instructions. This comes with the ebook chapter on tokens and probability, the email course and the first audio episode.
  2. Let it think step by step: You learn to work with examples instead of rules, to set out the steps of reasoning and to break tasks apart. In the videos you collect and rate ideas and ask for sources. This comes with a guide of 25 pages and the second audio episode.
  3. Bring answers into a fixed form: You separate inputs from instructions, set structure and length and ban filler sentences. The videos show small HTML examples, business problems, the checking of statistics and the rewriting of texts. This comes with the list of seven mistakes.
  4. Spot warning signs early: You build a small test set with everyday and difficult cases and compare versions of an instruction. The videos cover checking and simplifying texts and reusing earlier requests. This comes with a checklist of seven pages.
  5. Switch models and keep going: You get to know system instructions, settings and the quirks of different AI models. The videos show how to narrow the frame of a request, solve a business process and derive a method from expert knowledge. This comes with the last ebook chapters and the third audio episode.

Which typical prompting mistakes does the programme cover?

The most common mistakes are unspectacular: introductions that are too long, forgotten format requirements, a single test run and assumptions that exist only in your head. A separate document in the programme lists seven such mistakes with remedies. You do not have to change everything at once; you can fix one mistake after another.

  • Novels instead of instructions: If you put three paragraphs of background before the actual task, you water it down. The advice is to name the task first and add only the context that really changes the answer.
  • No format given: Without details on length, structure or bullet points, the AI has to guess, and you format by hand afterwards.
  • Tested only once: The same instruction can turn out differently next time. That is why you are advised to run an instruction several times before you trust it.
  • Ignoring the quirks of AI models: What works well in one assistant can turn out differently in another, even with the same wording.
  • Important points hidden: Hard requirements such as a word limit belong at the start, not at the end of a long text.
  • Shared knowledge assumed: A new chat knows neither your project nor your audience nor earlier conversations. Every instruction must make sense on its own.
  • Starting from scratch every time: Recurring tasks belong in a template with placeholders that you only fill in.

How do you test an instruction before you trust it?

The programme advises against using an instruction after a single successful try. Instead, you build a small test set: one normal case and several problem cases. Only when the instruction gives a usable answer in every case do you use it in daily work. It costs little time and stops mistakes from being noticed only by customers or managers.

The smallest test set in the materials has five cases: a typical input, an empty input, a long text with a hidden piece of information, an attempt to override the instruction and meaningless characters. Later you mix them, with about seven in ten cases being everyday ones and three being problem cases. A checklist with six areas is added, from the structure of the instruction through logic and format to settings and monitoring.

Another note concerns newer AI models that already reason step by step internally. For them, the request ‘think step by step’ can get in the way. It is better to describe what the result you want should look like.

Who is the programme for, and who is it not for?

The programme is aimed at working people, freelancers and teams who use AI or will soon use it and need more reliable results. It suits you less if you only want to ask a quick question once, or if you are looking for a finished tool that does the work for you. It is a learning programme with materials, not software.

No prior knowledge is needed. The videos start with setting up an account and with simple instructions. To be honest, though: the ebook and several PDFs are written from the viewpoint of people who build AI into software and fixed processes. Terms such as JSON, system instruction or interface appear there. If you do not program, read such examples as an illustration of how precisely an answer can be specified. The basic ideas need no program code: name the task clearly, set the format, test several times, check sources.

The programme is not meant for programming AI applications. Visionary Ventures is a learning portal; if you want processes implemented for you, contact Design Planet.

What materials come with the programme?

The programme includes 20 videos totalling 32 minutes, five PDF documents with 152 pages in all and three audio episodes. The videos are spoken in English with English and German subtitles and a transcript. All PDFs are available in English and German, and the audios are spoken in English with a transcript in English and German. The materials are for members’ personal use.

The five PDFs at a glance: the ebook ‘AI Prompt Engineering’ with 78 pages as the main work, an email course in seven lessons with 17 pages, a guide to error-proof prompts with 25 pages, the list ‘7 Mistakes Costing You Hours with AI’ with 25 pages and a checklist of seven pages on whether an instruction is ready for everyday use. The page counts refer to the English originals; the German PDFs are shorter because of a different layout.

The videos are short, from under one minute to almost three, and each shows one action on screen, using the chat assistant ChatGPT as the example. The three audio episodes are about unreliable AI and what helps against it, about error-proof prompting as a basic skill and about avoiding mistakes on a larger scale.

How do you work with the programme, and what does access cost?

You log in to the member area, open the programme and work through the five steps in order. Access is included in the membership: CHF 39 per month, CHF 299 per year or a one-off CHF 799 (lifetime). Monthly and annual subscriptions can be cancelled at any time in your account, and within 14 days the amount is refunded on request.

There is no fixed pace. For each step you open the videos, read the matching chapter, listen to the audio episode if you like and tick off finished units; the programme shows what comes next. As the videos last only 32 minutes, reading and practising decide how long it takes. Try each method straight away on a task of your own.

The membership covers all Visionary Ventures programmes. Payment is handled by Stripe with card, Apple Pay, Google Pay or Link, and the lifetime membership can also be paid with TWINT. You log in with a password and an authenticator app. Teams with several seats are possible.

Frequently asked questions

Do I need prior knowledge to learn prompts?

No. The videos start with setting up an account and with simple instructions. Some PDF chapters are more technical, for example when they deal with fixed data formats. If you do not program, you can read these examples as an illustration. The basic ideas, which are a clear task, a set format, several tests and checked sources, work without program code.

How long does the programme take?

The 20 videos last 32 minutes in total. Added to this are PDFs with 152 pages (the length of the English originals) and three audio episodes. How much time you need in all depends on how thoroughly you read and how much you try out yourself. There is no fixed timetable; you work at your own pace and can tick off finished units.

Do I need an AI tool, and what does it cost?

Yes, you need access to an AI chat assistant to try out the instructions. The videos show how to set up an account and use ChatGPT as the example. Whether the provider charges anything depends on its offer. These costs are not part of the Visionary Ventures membership.

What language are the materials in?

The programme description and learning path are in English. All PDF documents are available in English and German, and you open them in the language you choose in the member area. The videos are spoken in English, with English and German subtitles and a transcript. The audio episodes are also spoken in English, with a transcript in English and German. The German transcript is generated automatically, so individual words may be inaccurate. You do not need to speak German.

Does what I learn apply only to one particular AI assistant?

The videos show ChatGPT, but the principles apply more widely: name the task clearly, set the format, test several times, check sources. The programme itself points out that AI models differ in style. The fifth step is about adapting proven instructions to another model.

Can I try the programme before I commit?

Yes. The monthly subscription can be cancelled at any time in your account, and within 14 days the amount is refunded on request. So you can look at the videos and PDFs and see whether the programme fits your work. The monthly subscription costs CHF 39, the annual subscription CHF 299.

Keep reading

Ready to learn?

With a membership you get access to all programmes: CHF 39 per month, CHF 299 per year or CHF 799 once.

View membership View programme