Synthetic profiles in Meduzia: what they are and how they're measured.

1. What a synthetic profile is

A synthetic profile is a description of a person, created with AI. With that description, the model answers in the first person, the way someone like that would: with their desires, their barriers and their way of deciding.

The description says who they are, how they decide, what they want and what holds them back. It's created for a category and a Latin American country by researching data on your brand's consumers in that market, not from a real person. The value shows up when many respond: how the answers are distributed and which option wins in each segment.

2. What each profile is made of

Each profile has eight fields and is created by researching data on your brand's consumers in each Latin American country: the socioeconomic level, the city and the way of speaking are those of that country.

Name, age, city

Basic data

E.g.: Lucía, 22, Medellín

SEL

Socioeconomic level, on its country's scale

E.g.: Estrato 3 (Colombia)

Occupation

What they do

E.g.: Design student

Decision mindset

Meduzia's own typology: impulsive, selective, social or critical

E.g.: Impulsive

Habit

How they shop

E.g.: Online, driven by discounts and videos

Desires

What they're looking for

E.g.: To feel unique in clothes she doesn't see everywhere

Barriers

What makes them hesitate

E.g.: Not knowing her real size and not being clear on the return policy

Personality traits

Five traits (Big Five), from 0 to 1

E.g.: Openness 0.8; conscientiousness 0.4; extraversion 0.7; agreeableness 0.6; neuroticism 0.6

3. How a question becomes an answer

You pick a category and write a question. The profiles in that category answer in parallel, each with its own description and the same rules.

The rules each profile receives:

• You are a synthetic profile generated by a model, and you say so if asked.
• Answer in the first person, in your country's language and with its expressions, in one or two sentences.
• If something doesn't matter to you or you don't know it, say so. Only talk about what you know.
• Your barriers and your way of deciding weigh as much as your desires.
• If the proposal clashes with what holds you back, hesitate or reject it: a real person is almost never 100% in favor.

Each profile returns its answer and a stance: wants it, hesitates or rejects it. That way you see not only what it chooses, but why.

4. From answers to results

Answers are counted by stance and by segment: how many want it, how many hesitate and how many reject it, with each one's reasons. The numbers are calculated by code, not by the model.

The result is a read on where your market is heading: which option sparks the most desire, among whom, what holds it back and in what words your consumer puts it.

5. How we ensure answer quality

• Profiles with a story. The more story a profile has, the better it answers: that's why each one carries habits, desires, barriers and traits.
• Answers in their own words. Intent measured in free text is closer to people's than a score is.
• Rules that ask for honesty. Each profile can hesitate or say no, just like a person.
• Topics where it performs best. Products, messages, packaging and shopping experiences: the decisions brands make.

6. What the research says

The method builds on the most cited work on simulating people. The figures come from each study, with its own system and in its own country.

7. How it's validated

We take real studies conducted in Latin American countries, with people's answers held by a third party. We ask the profiles the same questions without seeing those answers, and compare against science's benchmark: how much the same people change when the survey is repeated with them.

1. Choose 3 to 6 real studies per category, with closed-ended questions and stored answers.
2. Write down and date what will be measured, before running.
3. Run blind: whoever runs Meduzia doesn't see the human answers.
4. Measure the distance question by question, and whether it gets the winning option right.
5. Compare against the same survey repeated and against a simple baseline prediction for the category.
6. Publish everything: the number, how many questions and studies, the dates, the model and the version.

8. Where it's growing

What we're adding to the engine:

• Studies with thousands of profiles, each evaluated with four ways of deciding.
• Profiles that update themselves with every new review, trend or competitor move.
• Image evaluation: packaging and creatives.
• A prediction log with date, version and real outcome.
• Every result with its confidence level, measured against previous replications.

9. Standards we follow

We follow the ICC/ESOMAR Code 2025: every synthetic result is declared as such, right next to the data. We also follow AAPOR's 2026 recommendations for using AI in research.

These answers were generated by an AI model using synthetic profiles for the category. They're useful for exploring and for deciding what to test.

If a brand publishes a Meduzia result, it presents it as what it is, a synthetic-profile result, and checks the wording with us before publishing it.