AI makes it easy to generate a plausible-looking answer fast — which is exactly why it's important to be deliberate about where it belongs in discovery and where it doesn't. Here are four areas where boundaries are better than shortcuts, and what to do in each case instead.
1. Don't let AI originate your personas or JTBD statements
It's tempting to ask AI to generate a persona from scratch — it's fast, and the output looks polished. It will even add images!
Instead: Use AI to draft a testable hypothesis from what you already know, then treat it as a starting question for real research, not a finished artifact. The value is in the shortcut to a first draft. Never skip the validation that follows it.
2. Don't justify a roadmap decision with "AI said users want this"
A synthesized answer with no traceable source is easy to lean on when you need a decision fast.
Instead: Use AI to help synthesize the user signals you've already collected — then cite the real users or data points behind the synthesis, so the reasoning holds up when someone asks "how do we know?"
3. Don't present an AI-generated finding as a standalone fact
An untraceable claim is hard to catch once it's in a deck.
Instead: keep every finding traceable back to a real participant, transcript, or data point, and treat anything that can't be traced as a hypothesis worth testing, not a conclusion worth building on.
4. Don't let AI smooth over the messy middle of your research (This might be the most important one.)
Synthetic summaries tend to look cleaner and more confident than real research usually is, which makes them easy to consume, and extremely convenient. I’m not gonna lie, some of the gobs of enterprise research findings made my eyes glaze over when our team researcher was presenting them. But that’s just a sign that it’s overwhelming, not that it’s not necessary.
Instead: Let AI help organize the messy data — clustering, tagging, transcribing — but sit with the contradictions yourself before drawing conclusions. That friction is often where the actual insight is hiding.
None of this is an argument against using AI in discovery — it's a guide to using it in the places it genuinely helps, while keeping a human checkpoint on the places that decide what you actually build.
If your team wants a research-process audit that puts guardrails like these in place before they're needed, that's exactly the work Strategic Creations does.