A persona turns the abstract idea of "our users" into one concrete person the whole team can picture the same way. Demographics like name, photo, and job are memory aids — what actually matters is goals, frustrations, and behavior patterns. "Checks review count before price" is useful for design; "woman, 30s" mostly isn't.
The most common way personas fail is being invented from the team's assumptions instead of research. Nielsen Norman Group calls the result a "decorative persona" — a pretty poster nobody actually consults while designing. They earn their keep only when built by clustering patterns that repeat across interviews, surveys, or behavioral data.
They also differ from market segments, which split people by business metrics like purchasing power. A persona splits by behavior and need, built to answer one design question: would this person want this feature?
When to use
Build one when the team is clearly imagining different users, and only once real research data exists to build it from — without data, a persona just manufactures false confidence.