AI Model Wiki

    Skip to main content
    15 min
    Lesson 6 of 6

    Common Pitfalls & How to Avoid Them

    From the course AI for Product Managers

    The most expensive AI mistakes PMs make — and how to sidestep each one.

    Pitfall 1: Trusting AI Outputs Without Verification

    What happens: A PM asks Claude to summarize competitive positioning, gets a confident, well-formatted answer, and includes it in a strategy deck. Three of the "facts" turn out to be hallucinated or outdated.

    Why it happens: AI models are trained to sound confident. They don't flag uncertainty the way a human analyst would.

    How to avoid it:

    • Treat every AI factual claim as a first-draft hypothesis, not a citation
    • For market data, pricing, or competitor features: verify against primary sources (the competitor's own site, official announcements)
    • Use Perplexity AI over ChatGPT for research — it cites sources inline, making verification faster
    • Build a "verify before decking" habit: one sentence of human confirmation per AI-generated stat

    Pitfall 2: Using AI to Avoid Talking to Users

    What happens: A PM generates AI-synthesized user personas and skips actual user interviews because "we have enough data."

    Why it happens: AI synthesis feels like insight, and interviews are time-consuming.

    How to avoid it:

    • AI is excellent at synthesizing existing user feedback — it cannot generate net-new user insight
    • Use AI to prepare smarter interview questions, not to replace the interview
    • A rule of thumb: AI handles synthesis of what users have already said; humans discover what they haven't said yet

    Pitfall 3: Generating Low-Quality PRDs at Speed

    What happens: PMs use AI to write PRDs so fast that they ship under-specified features. Engineers make assumptions, the feature launches incomplete.

    Why it happens: AI makes writing fast; the bottleneck shifts from writing to thinking, but the thinking step gets skipped.

    How to avoid it:

    • Spend more time on the input than the output — the quality of your prompt brief determines the quality of the PRD
    • Always add: edge cases, explicit non-goals, and success metrics before handing a PRD to engineering
    • Use AI to stress-test your PRD: "What questions would an engineer ask about this spec?" Then answer them

    Pitfall 4: Creating AI Dependency Without Skills Transfer

    What happens: The PM who learns AI tools becomes a bottleneck — teammates don't know how to use them, everything routes through one person.

    How to avoid it:

    • Document your workflows in a shared prompt library from Day 1
    • Run monthly "AI office hours" for the team
    • Pair on tasks rather than doing them for colleagues — teach the process, not just the output

    Pitfall 5: Ignoring Data Privacy

    What happens: A PM pastes customer PII, unpublished roadmap details, or internal financials into a public AI tool. Data ends up in training sets or is visible to the vendor.

    How to avoid it:

    • Check your company's AI usage policy before using any tool with company data
    • Use enterprise versions of AI tools (ChatGPT Enterprise, Claude for Enterprise) — these opt out of training
    • Never paste raw customer data; anonymize or aggregate before analysis

    Key Takeaways

    • •Never include AI-generated facts in strategy decks without verifying against primary sources
    • •AI synthesizes existing user feedback; it cannot replace the discovery of net-new user insights
    • •Spend more time on the PRD input brief than the output — quality in, quality out
    • •Always anonymize customer data and use enterprise AI tools to avoid privacy violations
    0 of 6 lessons marked complete

    Progress is kept in this browser only.