Audience Research
What if you could ask your audience anything?
There is a new way to do audience research. With AI Avatars, we can give you conversational access to thousands of data points from more than 480,000 U.S. consumers. This means you can have a conversation with an AI-powered avatar that represents your target audience and ask it anything at any time. From their opinions on different products to their thoughts on marketing campaigns, answers are backed by human data and provide a real-time read on consumers.
Try it
Talk to a ‘Millennial Mom’
Tap a question or write your own. You’ll get an answer informed by real women just like her.
An AI audience is a chat-based persona representing a specific consumer segment. It’s grounded in thousands of audience data points — demographics, attitudes, and behaviors. It’s given its own bio and a point of view based on the self-reported information provided by real humans, plus instructions so you can ask it questions or have it react to content you bring.
Jess · Millennial Mom
39, married, two kids, rents in Columbus OH. Full-time office manager, part-time logistics coordinator for everyone else. Does her shopping from her phone — usually after bedtime. Finds the deal before you do.
This is a demo, and does not give you access to the actual tool or the data it contains.
The data behind our Millennial Mom audience
Some significant insights from 30,285 data points
Relationship
Married
49%
Target
38%
Nat Rep
130 Index
How likely this audience is to give this answer compared with the U.S. population. 100 is average; 130 means 30% more likely.
41,100 Base
How many survey respondents this figure is based on.
Household size
Four people
28%
Target
15%
Nat Rep
179 Index
How likely this audience is to give this answer compared with the U.S. population. 100 is average; 130 means 30% more likely.
41,100 Base
How many survey respondents this figure is based on.
Housing
Rents
40%
Target
32%
Nat Rep
124 Index
How likely this audience is to give this answer compared with the U.S. population. 100 is average; 130 means 30% more likely.
3,666 Base
How many survey respondents this figure is based on.
Employment
Taking care of home/family full-time
21%
Target
5%
Nat Rep
433 Index
How likely this audience is to give this answer compared with the U.S. population. 100 is average; 130 means 30% more likely.
41,100 Base
How many survey respondents this figure is based on.
“We usually sit down together for a meal at home”
Definitely agree
32%
Target
24%
Nat Rep
137 Index
How likely this audience is to give this answer compared with the U.S. population. 100 is average; 130 means 30% more likely.
1,805 Base
How many survey respondents this figure is based on.
“I worry I’ll never be able to save for a rainy day”
Definitely agree
30%
Target
20%
Nat Rep
149 Index
How likely this audience is to give this answer compared with the U.S. population. 100 is average; 130 means 30% more likely.
1,274 Base
How many survey respondents this figure is based on.
FAQ
The questions everyone asks first
No. Each audience is grounded in thousands of data points from real, self-reported human research — licensed panel data plus first-party client data. The model does the talking; the research decides what it says.
Every audience is built from significance-tested survey data: only traits that clear a statistical threshold against the general population make it into the profile, and any answer can be traced back to the numbers behind it. In engagements where we've run AI testing alongside traditional qual and quant, the audiences surfaced the same core findings — at a fraction of the time and cost.
Where it breaks down is the edge of the data. Ask about a niche product the audience was never surveyed on, or a hypothetical with no behavioral precedent, and it will reason from adjacent signals rather than evidence. Treat those answers as informed hypotheses, not findings — and when a project needs more than that, we can run additional qual and quant research to complement it.
No — it changes what you spend real research on. Audiences do the early, high-volume work: pressure-testing a brief, comparing messaging territories, killing weak concepts before a focus group ever sees them, getting a first read in an afternoon instead of three weeks.
Yes. Every audience starts from a definition — who they are, what they do, how they buy — and we assemble the research profile to match from our licensed panel data. If the category has a real population in the panel, we can build for it.
Yes. Licensed panel data gives every audience its foundation; your data makes the signals even stronger. Segmentation studies, brand trackers, CRM data, past qual — we can layer your data on top of what we have, so the audience reflects your customers as you actually know them, not a category proxy.
Your conversations stay inside the platform, scoped to your organization, and are used only to generate the responses you see. They are not used to train the underlying model, and nothing you share is visible to other clients.
The panel data underneath every audience is collected daily and refreshed weekly, drawn from a continuous conversation with over 30 million panel members worldwide. First-party data can be updated whenever you send us new inputs. And for what happened this morning, audiences have live web access: they can react to a news cycle or a competitor's launch and tell you how it changes their thinking.