Polling has a trust problem. We built a firm around fixing the part we can.

Twenty-five years ago, roughly a third of the people a pollster called would pick up and finish the interview. Today the figure is in the single digits. Every poll now depends on a weighting model to stand in for the people who did not answer, and the people who did not answer are different from the people who did in ways that demographic weights do not capture.

The misses followed. By the industry’s own post-mortem, the national polls of 2020 were the least accurate in forty years. The error was analyzed and debated and then, for the most part, the same firms published the next cycle’s polls the same way, with no public accounting of how the last cycle’s had done.

Meanwhile the information a reader would need to judge a poll for themselves went missing. Question wording lives in a PDF, if anywhere. Crosstabs are available to subscribers. Weighting targets are proprietary. Response rates are not mentioned. And the fastest-growing bloc in American politics, the voters who say neither party represents them, appears in most horse-race polls only as “other” or “undecided.”

The result is a market in which a careful poll and a careless one look identical to the reader, and the reader has responded rationally: by trusting neither. In our own demonstration tracker battery, public opinion polls rank below the Supreme Court and barely above the news media.

What we can fix

We cannot raise response rates. Nobody can. What a polling firm can do is make every assumption it is forced to make visible, so that a reader, a journalist or a rival pollster can see where an error would come from if there is one. That is the whole idea. It is not a marketing position. It is the publication checklist, and a release that cannot meet it does not go out.