Topic 6: Know when to trust GenAI and when not to
Generative AI can produce answers that are clear, detailed and convincing. But a confident-sounding answer is not necessarily a correct answer. When you use GenAI in professional work, evaluating the output is just as important as generating it. Before using an AI-generated recommendation, analysis, document or decision input, take a few moments to check where the information came from, what evidence supports it, and what might be missing.
Evaluate GenAI Outputs before you use them.
Use these seven checks before relying on an important GenAI output
1. Check the source: Ask where the information came from. Separate information you provided from information GenAI generated.
2. Look for evidence: Check whether important claims, recommendations and conclusions are actually supported by evidence.
3. Spot assumptions: Identify places where GenAI may have filled in missing information or introduced assumptions that weren’t part of your original context.
4. Find the gaps: Ask what might be missing, incomplete or overlooked. A useful answer can still leave out something important.
5. Verify important information: Independently validate important facts, numbers, sources and other critical information using trusted sources or your own data.
6. Challenge the output: Don’t only ask GenAI to improve its answer. Ask it to identify weak evidence, alternative interpretations, risks and reasons its answer could be wrong.
7. Apply human judgement: Treat GenAI output as input to your thinking, not the final decision. You remain responsible for deciding what is appropriate to use.
When should you be more cautious?
The level of validation should increase with the importance and potential impact of the output. Be especially careful when GenAI is being used to support decisions involving customers, business performance, financial information, sensitive data, compliance, governance or other high-impact situations.
For lower-risk activities such as brainstorming ideas, exploring alternatives or creating an early draft, you may need less verification, but you should still recognise what is generated rather than known.
A useful habit: Ask GenAI to challenge itself
After receiving an important response, try follow-up questions such as:
- What assumptions did you make?
- Which claims in this response need verification?
- What information is missing?
- What evidence supports this conclusion?
- What are the strongest arguments against this recommendation?
- Where could this analyis be wrong?
These questions don’t make the answer automatically trustworthy, but they can help you identify areas that deserve closer human review
DSM Tip
The goal isn’t to trust GenAI or distrust GenAI by default. The goal is to know when an output is sufficiently supported for the way you intend to use it and when you need to investigate further. GenAI can generate. GenAI can assist. But you validate and decide.
