Dr. Robert Epstein on science: what the evidence says · JRE #1768

What the evidence says 01 / RECORD

Robert Epstein and colleagues have published a series of peer-reviewed, randomized controlled experiments since 2015 on what they call the Search Engine Manipulation Effect (SEME) and related effects. The original 2015 PNAS study, using 4,556 undecided voters in the US and India, found that biased search rankings shifted voting preferences by 20% or more on average, with shifts as high as 80% in some demographic subgroups (e.g. Moderate Republicans, VMP = 80.0%), and that masking the bias produced little to no participant awareness of manipulation. A related 2022 study on "answer bots" and voice assistants found shifts of more than 65% in some experimental conditions, with masking reducing awareness of bias to "close to zero." A 2024 replication testing non-political topics found smaller effects, with shifts of 17.8% to 30.9%. Epstein's specific figure of "70% to 90%" and the claim that "not a single person" detected the manipulation do not correspond precisely to any single published result; the peer-reviewed papers report a range of effect sizes (roughly 18% to 80%) depending on topic, demographic group, and experimental design, and describe post-masking awareness as very low or "close to zero" rather than literally zero. The broader body of work is published in reputable journals (PNAS, PLOS ONE) and has not been retracted, but some findings have drawn methodological criticism, including concerns about sample representativeness, self-selected "undecided" participants, and reliance on Epstein's own lab (through his American Institute for Behavioral Research and Technology) for replication.