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?
Jessica GidwaniProgram Management| US ArmyNorthern Virginia, United States
Trust but verify the output, ensure the LLM has not hallucinated or made assumptions based on gaps in the prompted information.
Saving Changes...
Mohammed ElrasheedStrategic Advisor | Business Development and Digital Transformation Consultant| Consulting ServicesRiyadh, Saudi Arabia
Mixing the CREATE Formula with the Patterns such as Chain of Thought, Chain of Feedback, ReAct, Question Refinement, Flipped Interaction, Risk Assessment Matrix, and Cognitive Verifier to have very well context, specified structure and clear guidance consolidated by given materials controlled by the validation techniques and continuous fact check to the outputs against the trusted and known primary sources; might lead up to 95% and above of receiving an accurate, relevant, and aligned result with your original goals expected.
Saving Changes...
Mohammed ElrasheedStrategic Advisor | Business Development and Digital Transformation Consultant| Consulting ServicesRiyadh, Saudi Arabia
Mixing the CREATE Formula with the Patterns such as Chain of Thought, Chain of Feedback, ReAct, Question Refinement, Flipped Interaction, Risk Assessment Matrix, and Cognitive Verifier to have very well context, specified structure and clear guidance consolidated by given materials controlled by the validation techniques and continuous fact check to the outputs against the trusted and known primary sources; might lead up to 95% and above of receiving an accurate, relevant, and aligned result with your original goals expected.
A thorough review of the initial result from the prompting is required, with a mandate to check the results with already established works. It also requires continuous validation. Breaking the units into subtasks will also be required.
Just like conversations between humans, AI prompt engineering needs to be flexible enough to go through iterations of refinement. This can ensure the prompts provide refined enough (not too much!) questions for Ai to respond to with enough specificity for the prompter's needs. Saving Changes...
have a well structured prompt, understand Project injection, drifting, leaking and AI Hallucination. Here are some common elements of well structure prompt.
●Instruction - a specific task or instruction you want the model to perform
●Context - external information, Persona or additional context that can steer the model to better responses
●Input Data - the input or question that we are interested to find a response for
●Output Indicator - the type or format of the output
●Response Tone – Tone of the response
I think reinject the output many times, refining each one.
Ensuring AI accuracy and relevance requires a disciplined approach that many teams skip in their enthusiasm for new technology. Here are best practices that consistently deliver reliable results.
Define clear success criteria before using AI. Know what a good output looks like so you can objectively evaluate what the AI produces. Without predefined criteria, you risk anchoring to whatever the AI generates simply because it sounds plausible.
Use structured prompts with explicit context. Provide the AI with relevant background, specify the format you need, define the audience, and state any constraints. The more precise your input, the more aligned the output will be with your goals.
Implement a verification workflow. Never use AI output as final without human review. For critical deliverables, establish a two-person review process: one person evaluates factual accuracy and another evaluates strategic alignment. This catches both hallucinations and misalignment with organizational context.
Cross-reference with authoritative sources. AI can generate convincing but incorrect information. For any data points, statistics, or claims that will influence decisions, verify against primary sources before incorporating them into project artifacts.
Iterate and refine. Treat initial AI outputs as rough drafts, not finished products. Use follow-up prompts to refine, challenge, and improve the output. Ask the AI to critique its own response or provide alternative perspectives.
Track accuracy over time. Keep a log of AI outputs versus actual outcomes. This builds organizational knowledge about where AI is reliable in your specific context and where it consistently falls short, enabling you to calibrate your trust appropriately. Saving Changes...
I find that AI provides useful results, tailored to my work if I treat the first output as a draft. I refine the results by asking the AI to focus on specific constraints, sometimes adjusting the context, or prompting it to explain its own reasoning before finalizing the output. Saving Changes...
Anonymous
I start with RTF prompts to create a baseline and have a better understanding and generate ideas for proceeding further to add details.
In my experience, the best practices are to clearly define the objective, provide precise instructions, validate the results against reliable sources, test different scenarios, and iteratively refine the prompts.
"When one door closes another door opens; but we often look so long and so regretfully upon the closed door that we do not see the ones which open for us."
PMI Team Member
