AIUI they look at how many people who have symptom X also test positive and then make a guess about prevalence based on how many other people reported symptom X.
The whole thing smells distinctly unsciencey to me. But great for generating press releases.
Symptoms.. Self selected group... Need I go on