A founder opens an AI tool and types: “Create a marketing plan for my startup.” The answer arrives in seconds. It is fluent, well organised, and, almost useless. It assumes the wrong customer, invents a budget, recommends channels the team cannot manage and presents guesses as facts.
This is not simply a technology problem. It is a task-definition problem. When a prompt leaves out the business objective, the audience, the evidence and the constraints, the AI has to fill the gaps. The result may sound confident while being poorly matched to the decision the entrepreneur actually needs to make.
Prompting is therefore not about discovering “magic words”. It is the practical skill of briefing an AI system as carefully as you would brief a colleague, freelancer or consultant. That skill sits directly within the EYE4AI mission: helping young entrepreneurs use AI purposefully, responsibly and in real business situations.
The BRIEF framework
Before submitting a prompt, build a BRIEF. The five parts below are simple enough for daily use, but specific enough to improve the usefulness of the output.
B — Business goal
State the decision, task or outcome the output must support.
R — Relevant context
Explain the venture, target customer, stage, market and available resources.
I — Inputs and evidence
Provide the notes, data, interview findings or source material the AI may use.
E — Expected output
Specify format, length, tone, audience and any constraints.
F — Fact-check and follow-up
Ask the AI to separate evidence from assumptions, flag uncertainty and suggest what must be verified.
“GIVE ME IDEAS FOR LAUNCHING MY APP IN GERMANY.” |
The request does not describe the app, the customer, the evidence already collected or the type of decision to be made. The AI is forced to guess. |
BUSINESS TASK |
Act as a market-entry research assistant. I am testing a subscription app that helps independent cafés reduce food waste. Our first users are ten cafés in Bulgaria. Based only on the interview notes and product information pasted below, identify three hypotheses we should test before entering Germany. For each hypothesis, provide: (1) why it matters, (2) evidence from my notes, (3) missing information, and (4) one low-cost validation experiment. Use a table and keep the answer below 700 words. Do not invent statistics or regulations. Mark every unsupported statement as an assumption and finish with five questions for a German sector expert. |
The stronger prompt does not guarantee a correct answer. It does something more valuable: it makes the output inspectable. The founder can see which statements come from supplied evidence, which are assumptions and which questions still require field research.
Use AI in three passes, not one
Produce. Ask for an initial analysis or draft using the BRIEF framework. Treat it as working material, not a finished decision.
Challenge. Ask the AI to identify contradictions, weak assumptions, missing stakeholders and reasons the proposal could fail. A second prompt should test the first output, not merely rewrite it.
Verify. Check important claims against primary sources, real customer evidence and qualified human advice. Record what was confirmed, changed or rejected.
This three-pass method reduces a common risk: accepting a polished first answer because it looks complete. For decisions involving money, legal obligations, people or safety, the verification step is not optional.
A reusable prompt template
|
Protect the information behind the prompt
A useful prompt often needs context, but more context is not always better. Do not paste personal data, customer records, confidential contracts, credentials, unpublished intellectual property or sensitive financial information into a public AI service. Remove names and identifying details, use synthetic or aggregated examples where possible, and check the tool’s data and retention settings before using business material.
Good prompting also requires fair framing. If the task affects recruitment, pricing, access to services or another high-impact decision, ask whose perspective is missing and whether the evidence represents the people who may be affected. Human accountability remains with the entrepreneur.
Turn good prompts into a small business asset
When a prompt repeatedly produces a useful result, save more than the prompt itself. Store the purpose, required inputs, an example of a good output, the verification checklist and the name of the person responsible for final approval. This creates a lightweight prompt playbook that a growing team can use consistently.
The EYE4AI consortium developing a Prompting Booklet with guidelines, tips and 25 real-world examples contributed by the five project partners. The practical lesson is already clear: a strong prompt is not a shortcut around thinking. It is a structured way to make thinking visible, testable and easier to improve.
Continue learning with EYE4AI
For more information and practical examples on this topic, explore these (EYE4AI) resources:
15 Perspectives on the Future of AI Entrepreneurship
AI Basics: Understanding Machine Learning, NLP, and Automation
How AI Can Help You Write Copy That Converts
Breaking Down the AI Hype: Practical Applications for Startups
English