5 · Using AI to synthesize research — without inventing findings
Once you've collected real interviews and survey answers, you have a pile of messy, raw notes. This is where AI earns its keep: it's excellent at clustering responses into themes and surfacing the most common pains — if you keep it on a tight leash. The danger is that a model will smooth over, exaggerate, or outright fabricate a finding that sounds plausible.
How to synthesize honestly with AI:
- Feed it only the real data. Paste your actual (anonymized) responses and ask it to group them into themes and count how often each pain shows up. The output must be grounded in what people actually said.
- Demand traceability. Ask it to tie each theme back to specific quotes. If it can't point to a real response, the theme isn't real.
- Watch for invented numbers. If your notes never mentioned a percentage and the summary now has one, the model made it up. Delete it. A made-up statistic is a hallucination, and shipping it is the opposite of research.
- Don't let it round up your hopes. "A couple of people mentioned X" must not become "users overwhelmingly want X." Keep the summary as honest as the raw data.
The hard rule — no synthetic data as evidence. Asking AI to "imagine what 50 customers would say" or to "role-play my ideal customer" can help you rehearse, but its answers are fiction. Never present a synthetic persona's opinion, or AI-generated "responses," as real market validation. This is the bright line between a rehearsal and lying to yourself. NIST frames trustworthy AI as valid and reliable — output you can actually stand behind — which a fabricated finding never is (NIST, 2023).
Trust DNA: AI may organize the truth you collected. It may never invent truth you didn't.
Check yourself. You ask AI to summarize 12 interviews and it reports "73% of users are frustrated by X," but you never recorded any percentages. What do you do, and why?
Sources
- National Institute of Standards and Technology. (2023). AI Risk Management Framework (AI RMF 1.0) — trustworthy AI is "valid and reliable"; outputs you can stand behind, not fabricated. https://www.nist.gov/itl/ai-risk-management-framework
- U.S. Small Business Administration. (n.d.). Market research and competitive analysis — base findings on your real direct-research data, not invented inputs. https://www.sba.gov/business-guide/plan-your-business/market-research-competitive-analysis