The challenge with using AI is having the right prompts to ensure you receive the best information available, how would you ensure as PM you don't try to influence the information when you don't like
Luis BrancoCEO| Business Insight, Consultores de Gestão, LdªCarcavelos, Lisboa, Portugal
Excellent question.
One practice I use is to deliberately ask for disconfirming evidence rather than additional supporting evidence.
If AI provides a recommendation that aligns with my preferred view, I will often ask: • What assumptions could make this conclusion wrong? • What evidence points in the opposite direction? • What are the strongest counterarguments? • Under what conditions would an alternative option be preferable?
This helps reduce the temptation to keep refining prompts until I obtain the answer I want.
In my view, the challenge is not simply prompt engineering. It is maintaining intellectual discipline throughout the decision process.
AI can improve access to information, but it cannot prevent us from seeking validation for our existing beliefs. That responsibility remains with the project manager.
Perhaps the real test is not whether AI provides the right answer, but whether we are willing to seriously consider answers that challenge our preferred conclusions. Saving Changes...
To ensure you don’t bias or manipulate AI outputs when you dislike the initial results, apply a structured “Objective Prompter” approach:
Audit for Emotional Language: Check your follow-up prompts to ensure you aren’t using leading or frustrated language (e.g., instead of asking "Why is this wrong? Show me why it won't work," use "Provide an objective counter-argument or friction points for this scenario").
Enforce Role-Based Guardrails: Explicitly assign the AI a neutral, critical persona within your prompt. Use instructions like: "Act as an objective, risk-averse PMO auditor. Critique this approach without bias."
Mandate Multi-Perspective Outputs: Never ask for a single answer. Structure your prompt to demand alternative viewpoints by default: "Provide three distinct approaches to this problem: one optimistic, one conservative, and one disruptive."
Verify with Raw Ground Truth: Don’t let the AI guess or validate your biases; feed it unvarnished project data, industry standards (like PMI or IEEE guidelines), or hard metrics, and ask it to analyze the gaps objectively.
Saving Changes...
Sergio Luis ConteHelping to create solutions for everyone| Worldwide based OrganizationsBuenos Aires, Argentina
AI has been moved from prompt based to context or harness based. Including it from human in the loop to human in the lead. So, prompts have low impact today when you work with AI. Just to comment AI is a boarder term then I am assuming we are talking about generative AI. Saving Changes...
Program Manager| HARPER SRLSanto Domingo / Distrito Nacional, Dominican Republic
One approach is to ask AI to present multiple perspectives, identify assumptions, and explain the reasoning behind its recommendations instead of asking for a specific answer.
I also find it useful to challenge the output with follow-up questions and compare it with other sources. As PMs, we need to be aware of our own confirmation bias and avoid using AI only to validate conclusions we've already reached. Saving Changes...
Prompt engineering is indeed a critical skill for PMs leveraging AI, and the risk of garbage in, garbage out is very real. Here are practical strategies to ensure you get reliable, useful outputs from AI tools.
First, be specific about context. Instead of asking AI to create a risk register, specify the project type, industry, phase, team size, and known constraints. The more context you provide, the more relevant the output. Think of it as writing a thorough project brief rather than a vague request.
Second, always validate AI outputs against your professional judgment and domain expertise. AI can generate plausible-sounding but incorrect information. Cross-reference AI recommendations with your experience, consult subject matter experts, and verify critical data points independently.
Third, use iterative prompting. Start broad to explore the problem space, then refine with increasingly specific follow-up prompts. This approach often produces better results than trying to craft the perfect prompt on the first attempt.
Fourth, establish verification protocols for your team. Before anyone acts on AI-generated analysis, it should pass a human review that checks for factual accuracy, relevance to your specific context, and alignment with organizational constraints the AI may not know about.
Fifth, maintain a prompt library. When you find prompts that consistently produce useful results for common PM tasks like stakeholder analysis, risk assessment, or status reporting, document and share them with your team. This creates organizational knowledge that compounds over time.
The PM who treats AI as an infallible oracle will make poor decisions. The PM who treats AI as a capable but fallible assistant will make better decisions faster. Saving Changes...
Community Champion
