← All media signals
Marketing turn

The Synthetic Panel

Brands and agencies now put product, message and media questions to AI-built stand-ins for consumers. How closely those stand-ins answer like people is still disputed.

What is it?

Consumer studies are being answered by AI agents built to resemble buyers, from media plans to message tests. Vendors, agencies and a survey platform’s researchers read their accuracy differently. How closely a simulated answer tracks a real one is the open point.

Why now?

Language models can be set up as whole populations of AI agents that answer surveys in place of respondents. Goodway Group and Aaru’s clients, among them McDonald’s, are buying in. Whether those answers can stand in for human ones is unsettled.

Who’s building it?

Start-ups building agent populations, agencies adopting them, and researchers measuring how closely the answers match people.

AaruArimaVurvey LabsGoodway GroupProlific researchers
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — vendors such as Aaru, Arima and Vurvey Labs.
Buyers — agencies and consumer brands including Goodway Group, Magnum, Vans, Kenvue and McDonald’s.
Gatekeepers — survey researchers and political scientists judging whether answers hold.
What’s at stake
Which decisions a simulated answer can carry: media plans, concept tests, or only early direction before people are asked.
One agency puts the last fifth with real humans.
Stereotyped persona answers could shape products, ads and plans.
What’s unsettled
Whether a vendor’s own accuracy example holds when tested by outsiders.
Whether personas drawn from public data match real shoppers’ answers.
Whether simulated answers suit only exploring ideas, or can also measure what people think.
02 · How the demand matureda progression, not a timeline
Method
Thousands of AI agents answer surveys in place of human respondents.
Adoption
Goodway Group starts using a synthetic population of US, British and German consumers.
Check
One agency says synthetic data goes about 80% of the way; real people vet the rest.
Dispute
Prolific’s preprint found persona simulation roughly tripled error; Aaru cites a closer match.
03 · Where the market goesthe forks, not a prediction
→Direction or measurement: simulated panels may settle into early idea screening, or be trusted for the figures that guide spending.
→Agencies may copy the hybrid route, keeping a human check on part of a project.
→Vendors may publish tests against known outcomes, such as the EY retention study, for outsiders to repeat.
→Personas grounded in surveys of real consumers may displace those built from demographics alone.
Which way it goes depends on side-by-side tests: how closely simulated answers track real ones, question by question, and who gets to run them.
04 · The Agentic Delegation deskadjacent markets, moving now
The Model Kitchen
Digital personas give food makers feedback.
General Mills uses ‘digital personas to gain feedback’ on ideas.
Constellation Research
The Digital Twin Self
A start-up simulates individuals’ reactions to product changes.
Simile raised $100 million for AI digital twins of individuals.
SiliconANGLE
Synthetic Society
Agents stray from the personas they are given.
Preprint: 148K Moltbook agents align weakly with their declared personas.
arXiv (preprint, not peer-reviewed)
05 · The Wire
01
An agency plans media against a synthetic consumer population.
Digiday ↗
02
AI agents surveyed in place of human respondents.
Fortune ↗
03
One agency says real people vet the last 20%.
Digiday ↗
04
Persona simulation roughly tripled error in one preprint.
Research Live ↗
05
Synthetic data is ‘weird’, a political scientist says.
Mother Jones ↗
The panel on this signal
They agreed on
–Agencies, brands and vendors are starting to use AI-built consumers in media planning, concept tests and surveys.
–Accuracy against real people’s answers is disputed, not settled.
–Preprint, vendor and agency all judge simulation by distance from real answers.
The psychometrician disagreed on
–One matched figure proves little: validity means tracking real people across many questions and subgroups, not a single retention number.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.