Synthetic customer research uses LLM-generated personas and digital twins to pre-test journeys, pricing and messaging, but it risks mistaking statistically plausible answers for actual voice of customer.
Synthetic customer research replaces or supplements human panels with AI-generated respondents — personas built from large language models, trained on demographic, behavioural and attitudinal data, that answer surveys, react to concepts, or simulate journey walk-throughs at a fraction of the cost and time of fieldwork.
Teams are using it early: pressure-testing messaging before a campaign launch, stress-testing a journey redesign against dozens of synthetic segments overnight, or generating a first-pass read on pricing sensitivity. The appeal is speed and volume — thousands of simulated responses where a real panel would yield dozens.
The unresolved question is fidelity. A synthetic persona reproduces patterns in its training data, not the lived friction, mood or context of an actual customer on an actual day. Used as a filter before real research, it accelerates the funnel. Used as a substitute for it, it launders assumption as evidence.
Why we think it'll come up
Digital twins enter the research stack
Research and CX vendors have moved from AI-assisted analysis of human data to AI-generated respondents themselves — personas that answer concept tests, journey walk-throughs and pricing questions without a real customer in the loop, marketed as a way to run hundreds of iterations before fieldwork begins.
Industry bodies press buyers to ask harder questions
ESOMAR's 2023 guidance, '20 Questions to Help Buyers of AI-Based Services', urges agencies and brands to interrogate suppliers on how AI and synthetic data inform their findings — a prompt, not a mandate, but an early sign that the practice is widespread enough to need scrutiny.
Speed is the pitch, not accuracy
Vendors selling synthetic panels lead with turnaround — same-day results versus weeks of recruitment and fieldwork — rather than claims of superior insight, tacitly acknowledging that synthetic data is a proxy, not a replacement, for lived customer response.
What it changes for customer experience
For customers
Products and messaging shaped partly by simulated versions of people like them, not always tested against their own actual reactions before launch.
For business
Faster, cheaper pre-testing cycles — but a growing risk of shipping decisions built on plausible-sounding synthetic consensus rather than real friction.
For CX & operations
Research teams need a clear protocol for where synthetic panels end and human validation begins, plus disclosure discipline when synthetic findings inform live decisions.
Industries on the front line
The Panel That Was Never in the Room
Customer research has always run on a trade-off between speed and truth. Recruit a real panel and you wait weeks for fieldwork, pay for incentives, and manage no-shows. Skip the panel and you move fast, but on assumption. Synthetic customer research promises to collapse that trade-off: generate a persona from an LLM, brief it with demographic and behavioural parameters, and ask it to react to a concept, a price point, or a redesigned journey — in minutes, at scale, without recruiting anyone.
The mechanics are straightforward. A model is prompted to adopt the profile of, say, a price-sensitive urban renter aged 28–34, then asked to walk through a checkout flow or respond to a campaign line. Run that prompt across hundreds of parameter combinations and a team gets what looks like a full segmentation study by the following morning. For CX teams under pressure to test more, faster, with flatter research budgets, the appeal is obvious.
A synthetic persona reproduces patterns in its training data. It does not reproduce the customer's Tuesday.
Why This Matters Now
In 2023, ESOMAR published '20 Questions to Help Buyers of AI-Based Services' — guidance aimed squarely at research buyers, urging them to ask suppliers precisely how AI and synthetic data are used in producing findings. It is not a rule and it does not require disclosure by law or code; it is a prompt for due diligence. But its existence signals something clearer than any mandate could: synthetic panels had moved from experiment to mainstream practice fast enough that the industry's own standards body felt buyers needed a checklist to interrogate the methodology behind the numbers they were being sold.
The instinct behind that guidance is sound, even if it stops short of compulsion. Synthetic respondents are statistically plausible, not empirically grounded. They answer the way training data suggests someone like them would answer — smoothing out the contradictions, the bad moods, the irrational loyalty, the context collapse that makes real customers messy and real insight valuable. A synthetic panel will rarely tell a brand something it didn't already, in some diffuse way, already know.
The Renascence Read: A Filter, Not a Substitute
Used correctly, synthetic research is a triage tool. It lets a team discard the weakest three of ten messaging options before spending real panel budget on the finalists. It lets a journey redesign get a rough directional read overnight instead of waiting three weeks for a proper study. That is genuine value — it makes human research more targeted, not less necessary.
The danger is organisational, not technical: the moment a synthetic finding is treated as equivalent to voice of customer, decisions inherit the model's blind spots as fact. A synthetic panel cannot surface the complaint a real customer makes about a product that doesn't exist yet in the training data. It cannot register the specific, local frustration of a regulatory change last month. It cannot be angry in a way that surprises the researcher — and surprise is often the most valuable output of real research. The absence of a hard disclosure requirement makes this discipline a matter of internal culture, not compliance — which is precisely why it is so easy to erode under deadline pressure.
What Good Practice Looks Like
- Use synthetic panels for early-stage filtering and directional hypotheses, never for final go/no-go decisions on price, product, or major journey change.
- Ask suppliers the questions ESOMAR's 2023 guidance sets out — how AI and synthetic data shape the findings — before commissioning any study, and expect a straight answer.
- Validate synthetic findings against a small live sample before committing budget or roadmap decisions to them.
- Treat contradictions between synthetic and human results as the most interesting finding in the study, not noise to be averaged away.
Synthetic customer research is not a fad and not a shortcut to the truth. It is a new instrument with a specific, narrow use: speeding up the early filter before real customers are asked to spend their time and honesty on something worth their attention. The guidance now emerging from bodies like ESOMAR does not settle the question of fidelity — it simply insists that buyers keep asking it.
Watch closely and pilot with strict guardrails: use synthetic panels only for early-stage filtering, never as a substitute for final human validation, and disclose their use internally and externally.
Trends Radar
Other trends
Build for what's next
Turn this trend into a measurable experience advantage.