Somewhere on your hard drive is a file with good prompts. A few screenshots from LinkedIn, a template from a colleague, three things that worked surprisingly well last time. And yet, you start from scratch with every new task.  

This is not a discipline problem. It's because a prompt collection only helps as long as your new task resembles the last one. As soon as it doesn't, you have a collection – but not a skill.  

The myth of the perfect prompt is persistent: Somewhere out there is that one magical phrasing that solves your vague task in one go. It doesn't exist. For anything complex and vague, quality doesn't come from the perfect question, but from the dialogue.  

And that is exactly what can be learned. In this workshop, you will train in a process we call LLM-Prompting: You develop your prompt together with the model, instead of guessing it beforehand. Sounds trivial, but it isn't – and it is the difference between "I have a good prompt" and "I can build one for any task".  

We practice this on real UX research tasks. The skill itself is completely independent of the discipline – it works just as well for your next concept, your next report, your next whatever.  

Why You Should Attend

• You stop starting from scratch: No more copying other people's prompts. You have an approach that works for every new task – even the one that doesn't exist yet.  

• You practice it instead of just seeing it: Having a process demonstrated once is not enough. You will apply it multiple times in the workshop – on tasks that intentionally look different. That is exactly how you notice that it holds up everywhere.  

• You can trust your results again: You learn to check outputs systematically instead of hoping. Including the question of where an LLM is reliable – and where it structurally is not.  

• You lose your reverence: AI models are astonishingly good and astonishingly stubborn. Both at the same time. Those who can categorize this soberly work faster and get annoyed less.  

• You take away something that isn't outdated in three months: Tool tips are consumables. A process is not.  

What Awaits You

How AI really works – and why that is practically relevant for you

• What a language model actually does when it answers (Spoiler: it knows nothing, it calculates)  

• Why hallucinations are not an oversight, but embedded in the architecture – and what that means for your work  

• The difference between a model that wants to please you and one that is right  

Frameworks: the frozen result of someone else

• How proven prompt frameworks are structured – and what they show you, but what they don't do for you  

• Why a ready-made framework only helps as long as your task fits into it  

• Live demo: How such a framework is created in the first place  

LLM-Prompting: the core process

• Talking to the model about the task before letting it do the task  

• Meta-prompting: letting the LLM develop the prompt you will use afterwards  

• Iterating, testing, cross-checking in a fresh chat – and knowing when you are done  

• Multiple runs on deliberately different types of tasks, with increasing independence  

Breaking down complex tasks

• Prompt chains for analyses that are too big for a single prompt  

• Critical rules that prevent the model from nicely hallucinating the gaps: source restriction, quotation requirement, interpretation ban, honest "the data is not sufficient for this"  

• The quality manager step: letting the model check its own work  

Providing context – blessing and curse

• Why your own data makes the results significantly better  

• And why the same mechanics carry a wrong input just as convincingly as a right one  

• Source hygiene: a skill that you as a UX researcher essentially already have – just directed at a new object  

Categorization and takeaways

• Creating vs. Analyzing: where the respective risks lie  

• Which task type needs which approach  

• The everyday checklist you will continue with tomorrow  

How We Work

Two-thirds of the workshop is practical exercise. You will work in small groups on a continuous case: a fictional startup that wants to improve the customer journey around dentist visits. The case is deliberately neutral – this way, everyone can work with the LLM of their choice without confidential data ending up in a chat window that no one controls.  

In-house with a secure environment? Then we can work with your real data and your real tasks instead. Just talk to us – that can be arranged.  

Who Is the Seminar For?

For everyone who no longer just wants to try out AI, but wants to use it reliably. The workshop is aimed at UX people with very different prior knowledge: Those who have done little so far will get a solid entry point here. Those who have been prompting for a while will get what is missing in most AI trainings – not more techniques, but the process behind it.  

The examples come from UX research. You take the skill with you into any other field.