Director, Learning Design & Development| PMIAsheville, NC, United States
Validating and checking outputs is critical when working with AI systems like Generative AI. Such validation approaches may include establishing clear criteria, implementing strong testing protocols, and continuous refinement.
In your experience with AI, what are some best practices for ensuring the results you receive are accurate, relevant, and aligned with your original goals?
When you talk to someone and you want to pass the message you have in mind, you take the responsibility of the communication: if the other person cannot understand what you say you provide more details, you explain with an example, you use a different wording, etc... until the other person can understand.
We can apply the same approach with AI, we take the responsability to provide the useful information needed by AI to provide the result we need. Saving Changes...
It is good to have an understanding of the subject or be ready to comb other sources for relevant information to compare with AI output. You can only catch an error, if you know it to be an error. Hence, having a good knowledge of the subject is paramount to optimizing AI. In my experience, I have had to call out wrong output, but, surprisingly, AI would maintain it is accurate, and I have had to push back a couple of times to set the record straight. Saving Changes...
I think the best way to be precise with your prompt, ask for sources, and putting timing around it, when applicable. And don't be afraid to review, question and prompt again. The initial response may generate questions or thoughts you previously did not have. The key is to remember the "A" in AI is artificial. The tool is only as smart as the data it has learned. Saving Changes...
Anonymous
I like to provide very specific detail to generate a response that I can review and determine if it is accurate or not. I then will continue to tweak my response until I am confident in the output. Saving Changes...
Efrain AylasProject Manager| INGETEC C&EMagdalena Del Mar, Lima, Peru
That’s a very relevant and thoughtful question. From my experience, ensuring that AI-generated results are accurate, relevant, and aligned with your goals starts with understanding your own level of familiarity with the subject matter. The more knowledgeable you are, the better you’ll be at evaluating the quality and reliability of the output.
One key practice is to always ground the prompts in verifiable facts or data. The more factual context you provide, the better the AI will perform. If the prompt lacks sufficient structure or clarity, the results may drift away from your original intent.
It’s also true that AI outputs don’t always come with clear indicators of accuracy, especially in fast-evolving or ambiguous domains. For this reason, I tend to apply AI mostly in areas where I already have a strong foundation, such as electrical calculations. This allows me to cross-check results quickly and confidently.
When dealing with new or high-risk content, I recommend consulting with subject matter experts and always validating the results with independent sources. Over time, as you become more familiar with how AI behaves, you develop better instincts to judge what’s useful and what needs refinement.
Ultimately, AI is a tool—its value depends on how thoughtfully it is applied and reviewed.
I’ve found that ensuring alignment with project goals requires a disciplined and intentional approach. Some best practices I follow include:
Clarifying the Objective Up Front: Before engaging with any AI system, I ensure the prompt or input clearly reflects the desired outcome and business context. Ambiguity leads to misalignment.
Establishing Evaluation Criteria: I define what “accurate” and “relevant” mean for the specific task—whether it’s drafting a risk register, generating stakeholder communications, or analyzing trends.
Validating with Subject Matter Experts (SMEs): Even when outputs seem correct, I review them with the project team or domain experts to confirm their validity, especially when decisions are involved.
Using Iterative Feedback Loops: I treat AI interaction like agile development—test, review, refine. This helps narrow in on quality outputs while maintaining alignment with evolving project needs.
Monitoring for Bias or Gaps: I’m mindful that AI tools may reflect data limitations or unconscious bias. Regular checks help ensure outputs support ethical and inclusive decision-making.
Documenting Assumptions: I keep a record of prompts, context, and responses, especially when using AI for planning or documentation. This supports transparency and auditability.
For me, I feel you need to fine-tune using the iteration process all the time before you finally arrive at the desired result. This may certainly not be a one-off thing. Saving Changes...
When using AI systems, I always start by clearly defining the objective and providing well-structured, context-rich prompts. This helps in getting results that are aligned with the original goal. I also validate outputs by cross-checking them against trusted sources or internal benchmarks, especially when accuracy is critical. It's important to review the AI’s response critically—treating it as a starting point, not a final answer. Lastly, involving subject matter experts to review or refine the results ensures both relevance and reliability before implementation. Saving Changes...
PMI Team Member
