Nothing in a discovery process produces disconfirming evidence by accident. You choose who to interview, you write the questions, and you decide what was notable — every step filters quietly toward the belief you already hold. Thirty interviews later the record is unanimous and has proven nothing.
The best published example of doing this deliberately is Superhuman's treatment of the people who were not enthusiastic. The obvious move with a segmented survey is to study your fans; they mined the middle group instead:
"we paid very close attention to this group, because our main benefit did resonate. Something — probably something small — held them back."
That is the whole technique. The "very disappointed" users tell you what you already do well. The "somewhat disappointed" users — the ones who got the point and still would not miss it — are the only group whose answers can change the product, because they have located a specific, surmountable obstacle rather than a mismatch of category.
The generalisation: the most informative respondent is the one who nearly converted and did not. Not your advocates, who confirm, and not the people outside your segment, who were never candidates. Three cheap ways to reach them: interview people who evaluated the category and bought nothing; interview people who tried the obvious incumbent and stayed; and write up the interviews where nothing lit up, because those are the ones nobody bothers to record, which is precisely why the file skews.