Research · Evidence graded · Updated when the studies are
Audience Personas
Last updated 2026-09-09 · 3 studies · 45 verified evidence rows · 1 persona
Most personas state a person's motivations with the same confidence they state their age, when only one of those came from a study. These do not. Every trait carries a grade and links to the research behind it, and the traits with no evidence are shown rather than hidden, because hiding them is how a persona becomes fiction.
Build anyone, and see where they actually are
Click or drag traits into the profile. A woman, 33, Asian, earning $80,000. Every platform row answers, and every row says how it knows: measured means a published figure exists and it is cited, estimated means it was combined from published figures and the arithmetic is printed when you tap the row. Ask for something nothing measures, and it says so instead of inventing it.
All US adults
National baseline · Published figures for all US adults.
Build anyone
Gender
Age
Race
Salary
Education
Lives in
Leans
Also ask
50 of 100 · use Instagram
Out of 100 US adults.
48 we can say do·4 could go either way·48 we can say do not
Show me
The dashed ring is the margin of error, drawn to scale. Add traits and watch it widen: that is the cost of being specific, and every other tool hides it.
Where they are
How they get online · every cell published
| All | |
|---|---|
| Uses the internet | 96% |
| Has home broadband | 78% |
| Owns a smartphone | 91% |
| Owns a cellphone of any kind | 98% |
| Owns a cellphone but not a smartphone | 7% |
| Smartphone-only: owns a smartphone but has no home broadband | 16% |
One column per trait, not combined: these sit near the ceiling where multiplying odds adds error without adding information. The smartphone-only row is the one that changes what you build.
Where they get news
AI chatbots are in this survey for the first time: 9% of US adults, 19% of Asian adults, the widest spread of any channel here.
Things nothing measures
Drag one of the dashed chips in for immigration status, children, a life event or a city, and this says what is missing instead of inventing it.
Open this and drag in the sources a buyer would actually use. It names the buying pattern, or refuses to until the mix is strong enough.
The country they live in
$81 median household income
$63 average debt owed per american
65.2% of households own their home homeownership rate
about $1 median rent, including utilities
35.9 million people living in poverty
4.1% unemployment rate
Measured figures are published survey results, cited. Anything labelled estimated is this site's own combination of them and carries no guarantee. Provided as is, for research and planning, with no warranty and no liability for decisions taken on it. Method and sources.
The estimator works in odds, not percentages. Each trait multiplies the national odds by the ratio its own published cell implies, and the result converts back to a percentage, so it cannot run past 100 the way multiplied percentages do. It assumes the traits shift the odds independently of each other, which is rarely exactly true, so the confidence label drops as traits are added rather than rising. Sampling error is combined in quadrature and shown per row; model error from the independence assumption sits on top of that and is not quantified. The underlying cells are all on the data page.
Every question, answered on its own page
The studio above answers any combination. These answer one question each, with every published cut charted, its margin of error, and what the pages currently ranking for it actually measured.
Which platforms they use2025 data
How many people use them2025 data
Where they get news2025 data
Measured
A study reports this for this population
Derived
Real data, but a wider population than this persona
Inferred
No source. An assumption to test, not a finding
The personas
The studies behind them
Each one states who was actually surveyed and what it cannot tell you. A persona is only ever as good as the population line.
US adults 18+ · n=5,022 · fielded 2025-02-05 to 2025-06-18
Cannot tell you: Reports whether someone ever uses a platform, not time spent, frequency, or what they do there. It publishes each demographic dimension separately, so there is no published figure for any INTERSECTION of two dimensions.
Global internet users, broad age bands · fielded 2026
Cannot tell you: Panel-based and self-reported, global rather than US, and reported for all users rather than any specific age and gender cell. Use for order of magnitude, never as a per-persona figure.
US consumers who bought a new or used vehicle in the prior 12 months · n=2,300 · fielded Fall 2025
Cannot tell you: Covers vehicle buyers as a whole. It is not cut by parental status, child age, or specific model, so it cannot describe a dad shopping a 3-row SUV specifically.
Why this is built the way it is
The behaviour that makes this section unusual, losing confidence as a persona narrows, is not a stylistic choice. It is the finding of a measured result.
Chapman, Love, Milham, ElRif & Alford · Proceedings of the Human Factors and Ergonomics Society Annual Meeting · 2008
Across six real survey datasets (N=268 to N=10,307) and two simulated ones, the authors generated 10,000 random persona-like descriptions per dataset and measured how many real people each matched. Prevalence fell rapidly with every attribute added. In their words, the 99th percentile description "fails to match anyone when 9 or more attributes are combined in 5 out of 6 survey datasets, including all consumer datasets", and only one dataset saw descriptions with 7 or more attributes match more than 0.3% of respondents. Their conclusion: a persona with more than a few attributes cannot be assumed to describe many actual people.
Applied here: This is why the composer gets LESS confident as you narrow, why two selected lenses are printed separately instead of merged, and why the do-not-assert list bans multiplying marginals. A tool that grew more confident with each attribute would be contradicting a measured result.
Chapman & Milham · Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 50(5), 634-636 · 2006
The original methodological critique: it is difficult to know how many real users a persona represents, personas cannot be verified or falsified and so have no demonstrable validity, and they tend to settle questions politically rather than with data.
Applied here: Answered directly by grading every trait and linking it to a study. A trait marked measured is falsifiable: open the source and check it. A trait marked inferred is labelled as an assumption to test rather than presented as a finding.
Salminen, Guan, Jung & Jansen · International Journal of Human-Computer Interaction · 2021
A systematic review of 77 data-driven persona articles from 2005 to 2020, identifying three eras: quantification (2005-2008), diversification (2009-2014) and digitalisation (2015 onward). The constructive counterpart to the Chapman critique: personas can be built from data, provided the data and method are stated.
Applied here: The reason this section exists at all. Personas are not abandoned here, they are made auditable: every figure carries its study, population, sample size, field dates and margin of error.
The data underneath
Every cell the studio reads from is printed in full on its own page, with sample sizes and margins of error: platform reach by gender, age, income, education and race, daily use as a separate survey, the US baseline from USAFacts, and which source answers which layer at what cost.
Open the data pageSources
Every figure the studio produces traces to one of the following. Each entry states what the study actually measured, its sample and field dates, and when it was last checked.
- [1]Americans' Social Media Use 2025 (full report, PDF)
Pew Research Center · accessed 2026-09-09
The primary source for every reach and daily-use figure on this page. Published November 20, 2025. Reach data: survey of 5,022 US adults, February 5 to June 18, 2025, margin of error ±1.9pp. Daily-use data: a separate survey of 5,123 US adults, February 24 to March 2, 2025. Read directly from the PDF after the interactive fact sheet produced figures that would not reconcile.
- [2]Social Media Fact Sheet
Pew Research Center · accessed 2026-09-09
The full crosstabs for the same survey: 11 platforms by age, gender, race, income, education, community type and party, plus the 2012-2025 trend. Read through Pew's machine-readable .md endpoint, which its robots.txt publishes for AI clients. All 136 cells that overlapped the PDF extraction matched exactly, which is what licensed the rest of the grid being taken from here.
- [3]Internet, Broadband Fact Sheet
Pew Research Center · accessed 2026-09-09
Internet use and home broadband subscription, each by age, race, gender, income, education and community type, with trend lines back to 2000. Published November 20, 2025 from the same NPORS survey. The source for the access layer: 96% of US adults use the internet, 78% subscribe to home broadband.
- [4]Mobile Fact Sheet
Pew Research Center · accessed 2026-09-09
Cellphone and smartphone ownership by seven dimensions, and smartphone dependency, defined as owning a smartphone without a home broadband subscription. Published November 20, 2025. The source for the smartphone-only figure of 16% nationally, which rises to 34% among adults in households under $30,000.
- [5]Social Media Use 2025: Methodology
Pew Research Center · accessed 2026-09-09
Where every sample size and margin of error on this site comes from. Total sample 5,022 at ±1.9pp; subgroups range from White adults at n=3,304 (±2.3pp) to Asian adults at n=211 (±8.9pp). Pew publishes this table as an image, so it was read from the image rather than parsed.
- [6]National Public Opinion Reference Survey (NPORS) methodology
Pew Research Center · accessed 2026-09-09
Address-based sampling with a web, mail and phone protocol. 2,349 completed online, 2,331 on paper and 342 by phone. The published unweighted sample size and margin of error for every segment on this page comes from its appendix, which is why each column carries its own ± figure.
- [7]Answers (94 questions, harmonised from 70+ federal agencies)
USAFacts · accessed 2026-09-09
Nonpartisan nonprofit founded by Steve Ballmer in 2017, which collects and standardises data from more than 70 federal agencies and names the agency behind each figure. Used here for the national baseline: income, wages, debt, rent, homeownership, poverty, population and household composition. Every row carries the originating agency as well, because the agency is the source of record and USAFacts is the route. Retrieved by direct fetch, permitted by their robots.txt.
- [8]Digital 2026 Global Overview Report
DataReportal, with GWI panel data · accessed 2026-09-09
Source of the time-per-day figures. Panel-based, self-reported, global and all-user, so it is graded derived rather than measured wherever it appears against a US persona.
- [9]Car Buyer Journey Study, 16th annual (summary PDF)
Cox Automotive · accessed 2026-09-09
Source of every vehicle-purchase figure. Surveyed 2,300 US consumers who bought a new or used vehicle in the prior 12 months, fielded autumn 2025, published January 13, 2026. Measures the journey for buyers as a whole and is not cross-cut by platform use, parental status or model.
No paid audience-intelligence tool was used. Everything here is public research, read at the source rather than through an aggregator.