Recent studies have shown that biased search results can produce substantial shifts in the opinions and voting preferences of undecided voters – a phenomenon called the “search engine manipulation effect” (SEME), one of the most powerful list effects ever discovered. We believe this is so because, unlike other list effects, SEME is supported by a daily regimen of operant conditioning. When people conduct searches for simple facts (86% of searches), the correct answer invariably turns up in the top position, which teaches users to attend to and click on high-ranking search results. As a result, when people are undecided, they tend to formulate opinions based on web pages linked to top search results. We tested this hypothesis in a controlled experiment with 551 US voters. Participants in our High-Trust group conducted routine searches in which the correct answer always appeared in the first search result. In our Low-Trust group, the correct answer could appear in any search position other than the first two. In all, participants had to answer five questions during this pre-training, and we focused our analysis on people who answered all the questions correctly (n = 355) – in other words, on people who were maximally impacted by the pre-training contingencies. A difference consistent with our hypothesis emerged between the groups when they were subsequently asked to search for information on political candidates. Voting preferences in the High-Trust group shifted toward the favored candidate at a higher rate (34.6%) than voting preferences in the Low-Trust group (17.1%, p = 0.001).