A search across behavioral economics, marketing and human-computer interaction literature turned up no published experiment with these exact parameters, no paper testing a “Recommended for You” label specifically, on two identical products, with a 48-to-65 percentage point shift, or with an awareness measure showing zero moderation. That does not mean nothing like it could exist in an unpublished industry test or a marketing white paper this search did not surface. It means the specific numbers should not be repeated as an established fact, and readers who have seen this statistic in a deck or an article should treat it as unconfirmed until someone produces the underlying paper.
What can be confirmed, and what matters more, is that the general phenomenon the claim is gesturing at, that being told you are being nudged does not reliably stop the nudge from working, is a real and replicated finding in the academic literature. That is the part of the story worth taking seriously.
The Real Experiment on Telling People They’re Being Nudged
The closest documented match is a 2018 paper in the Journal of Economic Psychology by Hendrik Bruns and colleagues. The researchers ran a laboratory experiment with 214 students at Erasmus University Rotterdam in June 2016, using a default nudge for contributions to carbon emission reduction rather than a product recommendation label, but the design directly tests the question at the center of the viral claim: does telling someone the mechanism and purpose of a nudge reduce its power over them.
Participants were split into five groups. One group saw no default at all, and unsurprisingly, none of them ended up at the contribution level a default would have set. A second group saw a plain default, with no explanation, and roughly one in five stuck with it. Three more groups saw the same default, but this time were told plainly what was going on: that it was designed to influence them, or that it existed to promote carbon reduction, or both. Telling people this barely moved the numbers. Those three groups landed in roughly the same range as the plain default group, give or take a few percentage points either way, and the differences between them were too small to count as real. The researchers also checked whether people who felt annoyed at being nudged pushed back against it out of spite, and found no sign that they did either.
In plain terms: disclosing the nudge, its mechanism, and its purpose did not produce a statistically meaningful drop in how many people followed it. That is a real, checkable result, published in a peer-reviewed journal, with a modest but clearly reported sample. It is also a much smaller and more specific claim than “48 percent to 65 percent,” and it concerns a charitable-contribution default rather than a shopping recommendation label. Anyone citing the Rotterdam study for the “awareness doesn’t help” mechanism is on solid ground. Anyone citing it for the specific percentages in the viral claim is not, because those numbers belong to a different, unverified source.
Why the Broader Nudge Literature Deserves Its Own Scrutiny
The credibility problem here runs deeper than one unsourced statistic. In 2022, Stephanie Mertens and colleagues published a meta-analysis in the Proceedings of the National Academy of Sciences, combining results from more than 200 separate nudge studies covering over 2.1 million people. It concluded that nudges, defaults especially, reliably and substantially changed what people chose. That paper was widely cited as proof that nudges work.
It did not hold up under a closer look. A follow-up analysis by Maier, Bartoš and colleagues, also published in PNAS in 2022, checked for a well-known problem in research of this kind: studies that find a weak or unimpressive effect are far less likely to get published than studies that find a strong one, which can make a whole body of research look more convincing than it really is. Once the researchers corrected for that pattern, the apparent power of nudges nearly disappeared. PNAS also issued a formal correction to the original Mertens paper. The headline finding from one of the largest nudge studies ever assembled shrank by roughly 90 percent once the missing weak studies were accounted for.
This is the context in which a specific, dramatic figure like “48 percent to 65 percent” should be read. It is not that such an effect is impossible. Default and framing effects are real and have been replicated many times over. It is that the nudge research field has an active, unresolved dispute about how large real effects actually are once selective publication is accounted for, and a single, precise-sounding percentage circulating without a citation is exactly the kind of claim that dispute should make readers pause over.
What Weaker, Real Evidence on Labels Actually Looks Like
Direct research on label-style nudges, closer in spirit to a “Recommended for You” tag, tends to produce messier results than the viral claim implies. A 2020 study in Frontiers in Psychology tested a nudge showing that most people had rejected a meat-based meal option, alongside nearly identical meal options for 97 participants. Overall, the label barely moved anyone’s choice, roughly three in ten picked the meat option either way, with or without the nudge. It only made a real difference for one narrower group: people who already felt torn about eating meat in the first place. That is a realistic picture of how these nudges actually behave in controlled studies: modest, dependent on who’s looking at them, and easy to overstate, not the clean, universal shift the viral claim describes.
What to Take From This
Two things can be true simultaneously. First, the specific numbers attached to the “Recommended for You” claim, the 48-to-65 shift and the zero-moderation-by-awareness finding presented as settled fact, cannot currently be traced to a real published study and should not be repeated as verified. Second, the general claim that disclosure does not reliably neutralize a nudge is supported by actual peer-reviewed research, most directly the Rotterdam default-transparency experiment, even though its numbers and setting are considerably narrower than the viral version suggests. Readers encountering a nudge statistic this specific should ask for the paper before repeating the number, because the field’s own largest attempt to quantify nudge effects had to be cut by an order of magnitude once bias correction was applied.