Prompt engineering refers to the deliberate design, structuring and optimisation of prompts for AI systems. These prompts are known as Prompts referred to as.
The aim is to explain to an artificial intelligence as clearly as possible what task it is to perform, what information it should take into account, and in what form it should output the result.
A good prompt might, for example, specify:
- what role AI is expected to play
- what the aim is
- what information is relevant
- which target audience is being addressed
- what style and tone the reply should have
- what format is expected
- what requirements or restrictions apply
A simple structure for prompts is:
| component | Example |
| Task | Carry out a market analysis. |
| Context | The company develops software for medium-sized businesses. |
| Target audience | The analysis is aimed at potential investors. |
| Requirements | Use plain language and highlight both the opportunities and the risks. |
| Output format | Present the results in a table. |
Prompt engineering can be used, amongst other things, to:
- Writing and revising texts
- Summarising information
- Developing ideas and concepts
- Analysing data
- to write programme code
- Responding to customer enquiries
- to support internal processes
- Integrating AI features into digital products
Typical prompt engineering methods include:
- Clear instructions: The task is described in concrete and unambiguous terms.
- Background information: The AI receives all the information that is relevant to the task.
- Examples: Samples show what the desired result should look like.
- Step-by-step tasks: Complex requirements are broken down into smaller steps.
- Formatting guidelines: The length, structure or format are specified.
- Iterations: The prompt is tested and improved based on the results.
A good prompt should be precise, but it does not have to be as long as possible. The key thing is that it contains all the relevant information and does not contain any contradictory instructions.
Common mistakes in prompt engineering include:
- unclear or overly general task descriptions
- lack of context
- several conflicting objectives
- No details regarding the desired outcome
- Unverified reproduction of the AI output
- Disclosure of confidential or personal data
- the expectation that AI will reliably fill in missing information
Even with well-formulated prompts, AI systems can generate incorrect, incomplete or fabricated content. Results should therefore always be checked by knowledgeable staff. Prompt engineering is no substitute for specialist knowledge or careful quality control.
Prompt engineering can be particularly relevant for start-ups when developing AI applications or seeking to make existing work processes more efficient. innoWerft supports founders in identifying meaningful use cases for artificial intelligence, further developing AI-based ideas and facilitating dialogue with experts and potential business partners.