Product Managers: Beware GPT PRDs

Think it’s a good idea to use an LLM to generate your PRD? It could do more harm than good. When we use LLMs to create Product Requirement Documents (PRDs), we risk amplifying all the classic pitfalls of poorly crafted PRDs — turning them from strategic tools into ticking time bombs of waste and frustration.

There’s been a proliferation of tips online about how to use LLMs to generate your PRDs, along with promotions of thin LLM wrapper tools that promise to, “save you valuable time and effort” by generating a PRD “with only minimal input.” While I have written before about my belief that AI will allow product managers to usher in a new age of abundance, I worry that the thoughtless use of AI to automate core responsibilities could instead lead to the worst kind of performative product management.

AI can be used effectively to make you a more impactful and productive product manager, but it's crucial to be aware of the risks and challenges to avoid exacerbating common pitfalls.

The Perils of PRDs

The PRD is a foundational tool in product development. A well-crafted lightweight and living PRD helps set context, encourages alignment on goals and priorities, and stimulates collaboration and exploration. At its best it is both a conduit for communication and a sandbox for ideas.

At its worst, it is a drag on productivity and a stain on the reputation of product management. Despite their importance, PRDs have often fallen short of their potential due to a variety of entrenched issues.

• Rigid and complex PRDs reinforce outdated waterfall methodologies, discouraging agile and collaborative approaches that are more suited to modern product development. 
• The attempt to automate and de-risk product development has led some to insert multiple layers of stakeholder review, stuffing documents with consensus-bloat requirements.
• In an effort to cover every possible angle, PRDs can become excessively long and detailed, which paradoxically makes them less likely to be read and adhered to by the team.
• A focus on documentation and obsessive refinement of the asset itself delays exploration, prototyping, and discovery work.
• Treating PRD creation as a form-filling exercise (with every field required) obscures what is not yet known behind placeholder content. 
• Some organizations treat PRDs merely as a “proof of work” for product management, creating poor incentives to generate documentation rather than create value. 

These issues have plagued PRD practices for years, often turning what should be a dynamic and strategic document into a static, ignored, and misunderstood formality. This fraught history led the best PMs to move to a more thoughtful approach to PRD creation—one that values insight over form and collaboration over procedural documentation. 

LLMs Gone Wild

We’ve all read about the problems of LLM hallucination–the tendency to generate plausible but factually incorrect content–or their cliché and grandiose writing style, but that’s not my main concern. I’m worried that the careless use of text-generating machines will only exacerbate the problems that we’ve worked so hard to remove from these processes.

Product managers that use LLMs to automate PRD creation, whether as a reaction to organizational incentives or as a result of inadequate training, will make the historical challenges plaguing the PRD process worse. Here are just a few of the pitfalls:

• Reinforcing rigidity: LLMs can produce prescriptive, overly elaborate documents that solidify waterfall methodologies, stifling agile and iterative development.
• Increasing consensus-bloat: AI can act like a group of overbearing reviewers, overloading PRDs with extraneous content and over-specification.
• Generating unwieldy documents: Capable of vast output, LLMs may create overly detailed PRDs that are cumbersome and less likely to be effectively used by the team.
• Inward focus: Over-reliance on a presumed all-knowing AI can shift attention away from essential first hand research, product discovery, and iterative prototyping.
• Masking unknowns: LLMs filling every PRD field can obscure critical gaps in knowledge with generic content, hindering genuine investigative development efforts.
• Proof of work instead of value: Quick and detailed document generation by LLMs may promote PRDs as mere bureaucratic artifacts, prioritizing proof of effort over genuine value creation.

Accepting LLM outputs without sufficient scrutiny or adaptation, leads to disengagement on the part of product managers. Treating PRD creation as a clerical task to be automated away undermines its core purpose—to push product managers to think hard about their products as they communicate context, goals, motivation, and value aimed at solving market problems.

Well Crafted PRDs

Despite these challenges, I remain optimistic about the role of AI as a tool for product managers, even in the context of creating PRDs. To understand how we can turn AI into a PM superpower, let’s start by outlining what makes for a well crafted PRD.

The goal of a PRD is to provide context to orient the team around the problem to be solved, the target audience, and the key characteristics of a successful solution. They are lightweight, clear, visual, and approachable documents. They are the starting point (not a substitute) for a collaboration between product, design, engineering, and even product marketing.

• Clear Product Vision and Positioning. Articulate a compelling vision for the product, defining who the customer/segment is, what the customer values, and what sets this solution apart.
• The Business Case. Provide context of the market conditions, competitive landscape, and consumer trends that justify the product's development.
• Communicate Customer Problems. Describe the user's journey, pain points, and how the solution will improve their experience or solve their problems.
• Strategic Alignment. Link the product’s objectives with the company’s strategic goals, showing how the product supports broader business objectives.
• Benefits and Differentiation. Describe concretely the benefits the users will get from the solution, tying that back to pain points and user stories described above.
• Highlight Key Features. Outline key features, again in the context of customer pain points and user stories, that must exist for the solution to be viable.
• Risks and Open Questions. Discuss potential risks and their impact on the product's success, along with mitigation strategies. Call out what critical questions still need answering.
• Success Metrics. Define what success looks like for the product, including specific, measurable outcomes.
• Wireframes, Mocks, Concepts. Work with your team on early designs and prototypes that make concrete what you’ve described in words.

This list is illustrative and should be adapted for the needs of your project. Use your judgment to add or remove elements and don’t feel the need to have everything fleshed out before sharing with others. Use the PRD as a framework to think through the problem, clarifying what is known and what is not, while you continue doing discovery, seeking feedback, and iterating.

Collaborating With AI

I admit it. I use AI daily for all sorts of work, including helping with PM duties. So how do I avoid the traps I laid out above? I stay in the driver’s seat,  treating the AI as an assistant that enhances my capabilities, not as a replacement for my own judgment. 

Here’s how you can effectively use AI to support the creation and refinement of PRDs:

• To get you unstuck. Feeling blocked or stuck in your own head? Feed your notes into an LLM, along with a description of the key elements of a PRD, and ask it to organize the information into a draft. You’ll immediately jump into editor mode to fix the results, but it will get you going.
• To give you feedback. Use LLMs to scan your PRDs for inconsistencies and suggest improvements. This can help enhance the clarity and coherence of your documents and eliminate a first round of basic feedback.
• For technical research. Deploy LLMs to quickly gather essential technical details on public APIs or open-source software. Not to substitute for a conversation with engineers, but to make that future conversation more informed and productive.
• For quick prototyping. Use AI-based UX prototyping to visualize your ideas for yourself and your team. This can significantly speed up the prototype testing phase, giving you faster feedback and more iterative cycles.
• For generating ideas. Ask LLMs to generate permutations on your ideas, whether it's coming up with product names or identifying related customer segments. This can help stimulate your own creativity and uncover new opportunities.

By strategically using AI as a collaborative tool in your process, you can maintain control over your PRDs while tapping into the efficiency and capabilities of AI. You aren’t automating a rote and thoughtless task, instead you are enriching a critical process.

Conclusion

As you integrate AI into the product management process, it’s crucial to avoid repeating the historical mistakes that led to ineffective PRDs in the past. This ensures that PRDs continue to act as effective guides in the product development process, rather than becoming mere formalities.

By leveraging AI as an assistant rather than a crutch, you empower yourself to create PRDs that are not just documents, but dynamic tools that drive innovation and clarity across your teams. Remember, the true value of a PRD lies not in its creation but in its daily use as a living document that guides your team towards delivering exceptional products.

Let’s learn to collaborate with AI intelligently, ensuring it amplifies our skill and judgment rather than automates and encourages bureaucracy. In doing so, we safeguard the craft of product management and elevate our ability to deliver products that truly matter.