The Jam Study, Revisited: What "Choice Overload" Actually Showed and Where It Broke

What the jam study actually found

The experiment people remember is deceptively small. Iyengar and Lepper ran a tasting booth in an upscale grocery store and rotated between a display of 6 jams and one of 24. The larger array drew a crowd: more passersby stopped to sample when there was more to see. But drawing a crowd and closing a sale turned out to be different things. Shoppers who encountered the extensive display were about ten times less likely to purchase a jar than those who met the limited one.

The finding did not stop at the till. In companion studies, including a laboratory version using 6 or 30 chocolates, participants given the larger set reported lower satisfaction with the option they eventually chose, and, when asked to write an optional essay, produced worse work than the limited-choice group. The paper packaged all three results under a single deliberately counterintuitive title, "When Choice Is Demotivating," and it became the founding empirical result behind what the field now calls choice overload.

Read carefully, the original claim was narrow and specific: under these conditions, in these settings, more options depressed both the rate of choosing and satisfaction with the choice. It was never a universal statement that fewer options are always better. The folklore that grew around it supplied the universality the data did not.

The paradox of choice, popularized

Four years later, the psychologist Barry Schwartz gathered the jam study and a wider body of work into a trade book, "The Paradox of Choice: Why More Is Less." The book did more than repeat the experiment. It offered a mechanism for who suffers most, drawing a distinction between two decision styles that has since entered common vocabulary.

A maximizer tries to evaluate every option in pursuit of the objectively best one. A satisficer chooses the first option that clears their own bar of good enough, then stops looking. Schwartz reported that maximizers, precisely because they refuse to stop searching, tend to report more regret and lower satisfaction with their decisions than satisficers, even when their objective outcomes are similar. In the maximizer vs satisficer framing, an overwhelming set of undifferentiated options is not merely inconvenient; it is a trap that a certain kind of buyer walks into voluntarily.

This synthesis is influential and useful, but it inherits the same underlying claim the jam study made, which means it inherits the same vulnerability. If the core effect turns out to be weaker or more conditional than the original experiment suggested, the popular framework built on top of it needs the same qualification.

Where it broke: the meta-analysis

By 2010 there were enough choice-overload experiments to ask the harder question: not "does this one study replicate," but "what does the whole body of evidence say on average." Benjamin Scheibehenne, Rainer Greifeneder, and Peter Todd assembled a meta-analysis of 63 conditions drawn from 50 published and unpublished experiments, a combined sample of 5,036 participants, and computed the mean effect of adding more options.

The mean was, in their words, statistically indistinguishable from zero. Across the literature as a whole, more options did not reliably make people worse off. This is the result that most popular retellings of the jam study leave out.

The subtlety that matters is what sat underneath that zero. The average was not zero because every study found nothing. It was zero because the individual studies pointed in opposite directions: some reproduced a strong overload effect, others found the reverse, a facilitation effect where more options helped. Averaged together, the opposing findings cancelled. A near-zero mean with large variance does not mean "no effect exists." It means the effect is real in both directions and the interesting question is what determines which direction you get.

The choice overload moderators that decide the direction

The value of the Scheibehenne meta-analysis is not the null headline. It is the list of conditions the authors identified as moderators, the variables that determine whether adding options helps or hurts. These are the choice overload moderators, and they are where the precise, usable version of the theory lives.

Real but conditional: not debunked, not universal

The temptation with a story like this is to overcorrect. Once the jam study is described as failing to replicate, a natural reaction is to declare choice overload debunked and move on. That reaction is as wrong as the folklore it replaces.

The disciplined position sits between the two. The original experiment is a real, well-executed study whose specific result stands; it was simply narrower than the movement it inspired. The meta-analysis is a real, well-executed synthesis showing that the effect does not generalize into a universal law. Both can be true at once, because the meta-analysis does not erase the original finding, it locates it. Choice overload is a genuine phenomenon that appears under identifiable conditions and reverses under others.

This is why we treat the general claim as contested rather than established, while treating the individual studies as established. The evidence tier is not a verdict on the researchers. It is a precise label on the strength of the sweeping inference, "more options are worse for buyers," that so much marketing advice quietly rests on.

From the shelf to the star rating

The choice-overload literature is usually taught with jam jars, but the mechanism it describes is exactly the one a local buyer runs when they compare businesses. A buyer choosing a med spa, a contractor, or a law firm is not scanning a jam display; they are scanning a set of options that are hard to tell apart, under some time pressure, often without prior expertise in the category. That is close to the precise combination of moderators the meta-analysis says produces genuine overload.

It also explains why the raw count of competitors is the wrong thing to worry about, and why differentiation is the right thing. Separately, there is good evidence that buyers lean on cues that are weaker than they assume. Across 1,272 products in 120 categories, average user ratings did not converge with independent quality scores, yet shoppers weight the star average heavily against smarter cues like the number of ratings or price. A buyer facing a set of undifferentiated options, leaning on a signal that is noisier than they believe, is not choosing badly by accident. The choice architecture they were handed made it hard to choose well.

Why one AI answer re-anchors the choice instead of removing it

There is a tempting reading of AI search that says it solves choice overload outright: if an engine returns a single recommended business instead of a long list, the buyer no longer has too many options to weigh. On the surface that looks like the limited-choice condition the jam study favored.

The more careful reading is that a single answer does not remove choice architecture, it concentrates it. When an engine names one business first, the classic biases do not disappear; they focus. Tversky and Kahneman showed that an arbitrary first reference point exerts disproportionate pull on a later judgment, even when the anchor is known to be irrelevant. And Bikhchandani, Hirshleifer, and Welch showed that once enough people visibly choose an option, it becomes individually rational for the next person to follow, ignoring their own private signal, which is how informational cascades form. A single named answer is the strongest possible anchor and the ideal seed for a cascade.

We flag this as an emerging extrapolation, not an established finding. The anchoring and cascade experiments were run on lists and sequential choices, not on modern single-answer AI surfaces, and the direct causal work on those surfaces does not yet exist. What can be said is that nothing in the classic literature suggests being the single named answer matters less than being one link among ten. If anything, the mechanisms predict it matters more.

What a careful reader should take away

The jam study is a case study in how a precise scientific result becomes an imprecise business truism. The precise result is worth keeping: under specific conditions, more options depress both choosing and satisfaction. The imprecise truism, that fewer choices are always better, is exactly the part the meta-analysis could not support.

For anyone deciding how to present options to a buyer, the operational lesson is not "cut your options." It is "make them legible, and know which of the moderators your buyer is actually facing." And for anyone reasoning about how buyers behave in AI search, the lesson is to resist the clean story in both directions: a single answer neither rescues the buyer from choice overload nor proves the theory, but it does raise the stakes of being the option the engine anchors on.