Dr. Robert Epstein on the Joe Rogan Experience #1768, fact-checked

What the evidence shows: Misleading. Epstein frames the 2011 removal of a large domain as one former Google employee's unilateral quality judgment. Google's own Online Security Blog post (June 17, 2011) shows it was a systemic anti-malware action: Google modified its automated scanners to flag entire bulk-subdomain services being abused, noting cases of more than 50,000 malware domains from a single provider. This is the basis for de-indexing the co.cc free-subdomain service (about 11 million domains). The roughly 11 million scale is accurate, but the cause was automated abuse-prevention policy, not a personal aesthetic decision.

What the evidence shows: Alphabet has at various points held cash and marketable-securities reserves in the vicinity of $100-150 billion, so the scale of the company's cash position is roughly plausible depending on the specific quarter referenced. However, federal law bars corporations, including Alphabet and Google, from using general treasury funds to make contributions to federal political candidates or committees; any political spending must go through a separate-segregated-fund PAC funded by voluntary employee contributions and subject to federal contribution limits, not the company's own cash reserves. The Federal Election Commission states plainly that corporations and labor organizations are prohibited from making contributions and expenditures in connection with federal elections using general treasury funds. Epstein's own research concerns search-engine ranking effects on undecided voters, a distinct claim from direct corporate political donations, and the two should not be conflated. As stated, the claim that Google can and does make "huge donations to political candidates" out of its corporate cash reserves misstates U.S. campaign finance law, even though Alphabet's overall political spending through its PAC and lobbying arms is real and substantial.

What the evidence shows: Federal Election Commission records, searchable directly at fec.gov, show that individuals who self-report Google or Alphabet as their employer have donated to federal candidates and committees overwhelmingly in favor of Democrats across multiple election cycles. Independent analyses of these FEC filings have put the Democratic share of Alphabet/Google-affiliated employee donations at roughly 80-90% in most cycles, with 2016 commonly cited as a peak year at approximately 94%; figures near 96% have also circulated for individual cycles or narrower subsets of Google-specific giving. Epstein's 94-96% figure falls within the range documented in FEC-based analyses, though "all donations out of Google" is an imprecise framing: the underlying data reflects self-reported individual contributions from employees using various Google and Alphabet subsidiary employer names, not a single official corporate donation total, and the exact share fluctuates by year and by which subsidiaries are counted. The claim is broadly consistent with publicly available federal campaign finance data, but the specific percentage should be treated as an approximation drawn from third-party analysis rather than a single precise, official statistic.

What the evidence shows: 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.

What the evidence shows: Epstein and colleague Ronald Robertson published peer-reviewed experimental work in PNAS (2015) showing that biased search rankings can shift undecided voters' preferences in controlled lab settings. The specific 2.6-to-10.4-million-vote figure for the 2016 election, however, comes from a separate, non-peer-reviewed white paper and later congressional testimony, not from a published, peer-reviewed study; it extrapolates from a small 2016 field observation of 95 subjects (21 undecided) over 25 days, using an unpublished computational model. When the same 2.6-to-10.4-million figure was cited (via a distorted "up to 16 million" framing) by then-President Trump in 2019, PolitiFact's review found outside experts identifying major methodological problems: the study lacked a clear definition of "biased" search results, relied on crowdsourced subjective bias ratings, excluded Gmail users based on a speculative theory that Google was intentionally showing them unbiased results, and never publicly disclosed the formula used to extrapolate from a few dozen subjects to a multi-million-vote national estimate. Independent researchers said there was insufficient information in Epstein's materials to verify that his model validly translated observed search-result bias into vote totals, and Google called the claim inaccurate and previously debunked. As of this writing, Epstein's vote-shift estimates for 2016 have not been independently replicated or published in a peer-reviewed venue, and the methodology remains contested.

What the evidence shows: Epstein, a psychologist who coined the term Search Engine Manipulation Effect (SEME), has published one peer-reviewed study on this topic: a 2015 Proceedings of the National Academy of Sciences (PNAS) paper with Ronald Robertson showing in lab experiments that biased search rankings can shift undecided voters' preferences, and estimating that Google bias in the 2016 election could plausibly have shifted 2.6 to 10.4 million votes. The much larger 78-million figure cited in this clip (78.2 million in his own writings) comes from Epstein's 2018 midterm monitoring project, which he described in his 2019 U.S. Senate Judiciary Committee testimony as based on more than 47,000 preserved election-related searches on Google, Bing, and Yahoo, spread across hundreds of local and regional races. That estimate has not been published in a peer-reviewed journal or independently replicated, and the underlying monitoring data and methodology remain proprietary to Epstein's organization. Google has disputed Epstein's conclusions, stating it does not design or adjust its algorithms to influence elections. Because the 78-million-vote figure rests on unpublished, non-peer-reviewed monitoring data extrapolated from a limited, non-random sample rather than on verified election outcomes, it remains an unverified estimate rather than an established finding.