Online research panels are large groups of people who have agreed to answer surveys in return for small payments, and the marketplaces through which researchers buy their answers by the thousand. A survey is only useful if the right people answer it. A panel lets a researcher reach, say, 1,000 people of a chosen type within days.
How is a panel built? A panel company recruits people through advertisements and websites. Each person who joins fills in a profile: age, city, income, job, the products they own and much more. When the person answers a survey, they earn points. They exchange the points for shopping vouchers, mobile recharges or cash. A panel recruited and run by the panel company itself is called a first-party panel.
No single panel has enough people of every type. So researchers often buy through a marketplace. This is an exchange that connects one researcher with hundreds of panels at once. It works like an ad exchange, where advertisements are bought and sold automatically. The researcher states who is needed, how many and at what price. The marketplace sends the survey link to matching people on many panels, and stops when the quota is filled. A quota is the number of answers needed from each group.
Consider a research agency in Mumbai working for a scooter company. It needs 1,000 answers from women aged 25 to 40, living in smaller cities, who ride a scooter or plan to buy one. It sets quotas: at least 150 answers from each of five states. Through a marketplace, the survey reaches people on 20 panels. The agency pays about ₹250 for each complete set of answers, so ₹2.5 lakh in all. The full 1,000 answers arrive in four days.
The biggest problem is quality. Because answers earn money, some people answer without reading. Some people use programs, called bots, that fill in surveys automatically. So panels check every set of answers. They remove people who answer far too fast, people who give the same answer to every question, and people who fail a simple check question, such as "Choose 'Agree' for this row". They check whether many answers come from the same device. Bot management software tells real people apart from scripts. In the scooter survey, about 8% of the answers were removed this way, and the marketplace replaced them.
The survey itself is built in survey software. Panels also supply viewers for tests with emotion AI, which reads faces as people watch an advertisement. A newer and cheaper idea is to ask an AI model to answer in place of real people. These are called synthetic respondents, and nobody yet knows how far to trust them.