Synthetic Persona Focus Group
Simulate diverse buyers, collect individual reactions, then synthesize the verdict
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 6
- Confidence
- 98%
The Synthetic Persona Focus Group turns a general-purpose language model into a panel of simulated buyers. First, the operator supplies enough product and category context for the model to identify plausible customers. The model then creates several demographic or professional personas and asks each one to assess the same question critically from the perspective of that persona's experience. Keeping those responses separate produces variation that a direct request for one answer would suppress. Finally, the model combines the individual reactions into a joint anonymous conclusion, including the reasons behind the preference. The result is a fast directional signal that can expose overlooked audiences, sharpen positioning, and prioritize what should receive genuine interviews, surveys, or experiments.
Origin
Mike Taylor demonstrated the method on Marketing Against the Grain by asking simulated HubSpot buyers to compare two landing-page headlines and then aggregate their reactions.
Core principles
- 01Generate diverse buyer perspectives before seeking consensus
- 02Make each persona respond from a specific background and experience
- 03Preserve individual reactions before aggregating conclusions
- 04Use synthetic feedback for direction rather than final proof
- 05Add product-category context when the brand is unfamiliar
How to run it
- 1
Set the product context
Name the product, explain its category, and provide relevant positioning details. Include more context when the product is niche or poorly represented online.
Pro tip Describe who currently buys the category even if the brand itself is unknown.
Watch out A brand name alone may produce weak personas when little public information exists.
- 2
Generate buyer personas
Ask the model for five to ten distinct demographic or professional personas who could plausibly buy the product.
Pro tip Review the list for overlooked audiences that may reveal new targeting opportunities.
Watch out Do not let every persona collapse into the same generic ideal customer.
- 3
Pose one consistent question
Give every persona the same concrete choice or research question and require a critical response based on their background.
Pro tip Comparisons between two messages or concepts tend to produce actionable reasons.
Watch out Vague questions produce generic reactions that are difficult to use.
- 4
Preserve individual responses
Have each persona state its preference and reasoning independently before requesting an aggregate answer.
Pro tip Look for disagreements that reveal meaningful audience segments.
Watch out Requesting consensus too early removes the variation that makes the method useful.
- 5
Synthesize the panel
Ask the model to combine the responses into a single paragraph written as a joint anonymous answer. Capture the majority direction and the reasons recurring across personas.
Pro tip Retain minority objections as hypotheses for segmentation or follow-up research.
Watch out Do not interpret the combined answer as a statistically valid survey result.
- 6
Validate the direction
Use the synthetic verdict to choose what to test through interviews, advertising, landing-page experiments, or other behavioral evidence.
Pro tip Spend real research resources on the questions where the simulated panel reveals the most uncertainty.
Watch out Never replace required legal, safety, or high-stakes human research with synthetic personas alone.
In the wild
Taylor generated plausible HubSpot buyers, asked each to choose between “grow better with HubSpot” and “grow without the guesswork,” and collected their individual reasoning. The simulated startup founder and marketing manager preferred different options, while the aggregate response favored the latter message because it promised clarity and actionable insight.
→ The process produced a directional headline preference and explained which audiences responded to each message.
A small-business software team could generate personas such as a freelancer, agency owner, retailer, and nonprofit director. Each persona would compare an automated cash-flow alert against a new reporting dashboard, explain the most urgent benefit, and identify objections before the team selects an experiment.
→ The team identifies the concept and audience worth validating with real customers.
Common mistakes
Asking for the average answer immediately
A direct question tends to return a safe, generic summary. Generate and preserve distinct persona responses before synthesis.
Treating simulation as representative data
The model reflects patterns in its training data, not a recruited statistical sample. Use the result to direct validation rather than claim market proof.
Omitting niche product context
Sparse public information weakens the model's understanding of likely customers. Supply category, customer, and product details before generating personas.
Is it for you?
Best for
It is best for marketers, founders, and product teams comparing messages, concepts, audiences, or positioning options.
Not ideal for
It is not ideal for high-stakes decisions requiring statistically representative evidence or verified customer behavior.
From the transcript
“if you ask it to think of a bunch of personas to roleplay as first then you know you get the individual responses from those…”
“first you want to set the scene and say give me 10 demographic personas just like regular people who would be buyers of uh your…”
“as if these people had collaborated in writing a joint Anonymous answer”
From the episode
This AI Prompt Gets You Customer Insights in 5 Minutes (Free Tool)