The Buyer's Guide to Synthetic Research · Article 3
A three-part series for research and insight buyers · 2025
Ten questions that separate the credible from the confident
You don't need a data science degree to buy synthetic research safely. You need ten questions and the nerve to sit in silence while they're answered badly.
You don't need a data science degree to buy synthetic research safely. You need ten questions and the nerve to sit in silence while they're answered badly.
Procurement has a superpower the industry underrates: the awkward question, asked plainly, with a pen hovering over the scoring sheet. Synthetic research is new enough that most suppliers haven't rehearsed their answers. This is your window. Use it.
The ten questions
One: what real data underpins your synthetic outputs, and can you evidence its provenance and consent basis? Two: how do you validate synthetic outputs against real respondents (which specific tests, and what pass thresholds)? Three: what proportion of synthetic outputs do you benchmark against real humans, and how often? Four: how do you assess and correct bias, particularly for non-Western populations that AI training data under-represents? Five: which model and version generated this data, and can you reproduce the same outputs on request?
Six: what re-identification risk assessment have you done? "It's synthetic so it's anonymous" is an assumption, not an answer. Seven: will synthetic elements be explicitly labelled in every deliverable, including verbatims and charts? Eight: what is this data not suitable for? Nine: how do you comply with MRS and ESOMAR requirements, and the EU AI Act's transparency obligations for AI-generated content? Ten: if we audited your methodology tomorrow, what documentation would you show us?
How to read the answers
Credible suppliers answer questions two, three and five fast, specifically, and without adjectives. They volunteer limitations before you ask (question eight should produce a list, not a pause). They treat labelling as obvious professional hygiene rather than an unusual demand. And they can name the humans in their loop: who reviews, who signs off, who is accountable.
The confident-but-not-credible have tells. They pivot every hard question back to speed and cost. They say "proprietary" when they mean "undocumented". They promise sample sizes that would be impossible with humans and see no problem worth mentioning in that. And they bristle, subtly, professionally, at the idea of an audit.
Making it stick
One question set changes nothing unless it enters the machinery. Put the ten questions into your standard RFP template so every supplier answers them in writing, scored. Add the disclosure requirement to your master services agreement, not just the statement of work. And ask for the calibration evidence, not a summary of it, as a contract deliverable. Suppliers raise their game remarkably quickly when documentation is a payment condition rather than a promise.
The opportunity hiding in the interrogation
Here's the quietly optimistic ending. Buyers who learn to ask these questions don't just avoid bad synthetic research. They develop the organisational muscle to buy the good kind early and confidently, while competitors are still frozen between hype and fear. In a market where research budgets are flat and decision windows are shrinking, the ability to use synthetic acceleration safely is becoming a genuine competitive asset. The nervous buyer's endgame isn't avoidance. It's discernment. And discernment, unlike fieldwork, compounds.