This is the page to read if you want to know what the answers are actually made of. The honest short version: some of what we know about people comes from talking to them, some from large public studies that governments and research institutions have run for decades, some from what people have published themselves, and some from data you bring. They are not equally strong, they are strong in different ways, and the product tells you which is which rather than blending them into an undifferentiated claim of accuracy.
The four kinds
Interviews
People we have talked to, verified as real and consented. The most expensive kind of evidence to gather and the most valuable, because an interview produces something nothing else does: a person's own account, in their own words, of why they do what they do.
Interviews run continuously across ages, incomes and regions. A portion of every interview round is held back and never used to build twins, so that it can be used to test them instead — see Validation.
Strong at: reasons, language, contradictions between what people say and do. Limited by: cost, and therefore scale.
Public statistical data
Large-scale studies of the population, run by statistical agencies and research programmes, and published as anonymised individual records: census and community surveys, expenditure diaries, time-use studies, long-running attitude surveys. Millions of real people who answered real questionnaires, with the weights needed to make a sample represent a country.
This is the backbone of the base population. It is the reason we can say something about a national market rather than only about the people we happened to interview.
Strong at: scale, representativeness, circumstances and behaviour. Limited by: having no voice. A survey record contains no words. It tells you what someone does, never how they would put it.
Because no single study covers everything, records are sometimes combined across studies for people of the same profile — expenditure from one, time use from another. Where that happens it is labelled combined. It is reliable at the level of the pattern and looser at the level of the individual: a combined twin's spending is right for someone like them, and the link between their spending and their attitudes is weaker than it would be in a real person.
What people have published
Public writing, posts and reviews. Where someone has written enough about their own life, that body of writing is evidence about them in a way a follower count never is.
Used two ways. From a large body of writing by many people we can measure what a group actually talks about and how — which becomes part of the environment. From one person's own writing we can build an identity with a genuine voice.
Strong at: language, current concerns, what people volunteer rather than what a questionnaire asked. Limited by: who writes publicly is not who exists. It skews, and we treat it as skewed.
Your own data
What you bring: customer records, purchase history, your own research, your own knowledge of your buyers. See Population for how to use it, and note that the most valuable version is often the easiest — telling us the shape of your customer base, without sending us a single customer record.
Strong at: being about your actual customers rather than the public. Limited by: covering only people who are already yours.
One rule about voice
A person's own words are only ever attached to the person who said them. We never give one person's quote to a different twin because they look similar on paper.
That constraint costs us something real. It means twins built only from statistical records have no verbatims, and read more flatly than twins built from interviews or someone's own writing. We would rather that than a population where you cannot tell which quotes are real.
What we can do with a body of writing from a group is measure it — what people in this group commonly say, how often a concern comes up — and that becomes a fact about the group in the environment, which is a true statement rather than a borrowed one.
Seeing sources
Every twin shows the source behind each line of its identity. Every environment fact shows the source it came from and, where it came from a document, the passage that supports it. If a fact is an assumption rather than something we can point at, it is marked as an assumption.