The Production Effect

Introduction

Try this. Read the word "elephant" silently. Now say "giraffe" out loud. Come back to this page in a week and see which one you remember. If the research is right, and over fifty years of experiments say it is, you will remember "giraffe." Not because it is a more interesting animal. But because you produced it. You moved your lips, heard your own voice, and felt the word leave your body. That act of production, simple as it sounds, created a memory trace that silent reading never could.

This is the production effect. The term was coined in 2010 by Colin MacLeod and his colleagues at the University of Waterloo [1], but the phenomenon itself was first spotted in a small experiment in 1972 and then promptly forgotten for nearly four decades. Its rediscovery sparked an explosion of research that now includes more than a hundred published papers, two journal special issues, brain imaging studies, developmental investigations in children as young as two, and one of the cleanest demonstrations in all of cognitive science that doing something with what you learn, anything at all, beats passively letting information wash over you.

This article tells the full story. From the experiment nobody noticed in 1972 to EEG recordings in 2025 that can decode whether a word was read aloud or silently based on brain wave patterns alone. Along the way, the story passes through singing students, mouthing participants, children in Brisbane, frigatebirds over the Galapagos (briefly), and a debate between two competing explanations that is still not fully settled. It also answers a practical question: if saying things out loud helps memory, why does almost nobody study that way?

The Experiment Nobody Read

In 1972, Ronald Hopkins and Richard Edwards at the University of Kansas ran a simple experiment. They showed college students lists of words, one at a time. Some words the students pronounced out loud. Others they read silently. Then came a recognition test [2].

The result was clear but puzzling. When some words in a list were pronounced and others were silent, the pronounced words were recognized about 10 percentage points better. But when one group pronounced everything and another group read everything silently, the advantage disappeared. Same words. Same pronunciation. But the benefit only appeared when aloud and silent items were mixed together in the same list.

Hopkins and Edwards published their finding as a two-page note in the Journal of Verbal Learning and Verbal Behavior. It landed with a thud. Nobody cited it. Nobody followed it up. The finding sat in the literature like an unopened letter for almost four decades.

Why? Partly because the result did not fit neatly into any existing theory. Partly because nobody gave it a name. And partly because memory researchers in the 1970s were busy with bigger fish. Fergus Craik and Robert Lockhart had just published their levels-of-processing framework in 1972 [3]. Norman Slamecka and Peter Graf were about to introduce the generation effect in 1978 [4]. Against these headline discoveries, a small pronunciation effect in a two-page note was easy to overlook.

A few researchers did stumble across related findings. Martin Conway and Susan Gathercole published two follow-ups in 1987 and 1988 showing 14 to 25 percent advantages for words read aloud [5], [6]. Patricia MacDonald and Colin MacLeod touched on it again in 1998 [7]. But none of these papers connected to each other. The phenomenon kept being discovered and forgotten. Almost as if it were demonstrating the very problem it described: without a distinctive label, without an act of production to make it stand out, a finding can vanish from collective scientific memory.

Naming a Ghost

Colin MacLeod had been studying memory at the University of Waterloo for decades when he noticed something odd. His lab kept finding that words read aloud were better remembered than words read silently, yet nobody seemed to have a systematic account of why. In 2010, he and his colleagues, Nigel Gopie, Kathleen Hourihan, Karen Neary, and Jason Ozubko, decided to pin the phenomenon down [1].

They ran eight experiments. The paper's title was deliberate: "The production effect: Delineation of a phenomenon," a conscious echo of Slamecka and Graf's 1978 paper on the generation effect. MacLeod was drawing a parallel. Just as the generation effect had needed a name and systematic investigation to become a major research topic, so did this neglected cousin.

The eight experiments established several things at once. First, the effect was large. In Experiment 6, words read aloud were recognized 19.7 percentage points better than words read silently [1]. That is not a small effect by any standard. MacLeod himself noted that "the advantage of reading aloud typically is 10 to 20 percent or more, by any standards a fairly large effect of a processing operation."

Second, the effect required item-specific production. In Experiment 4, participants said "yes" to every word they were supposed to produce. Same motor act, same vocalization. But the memory advantage vanished. Production only works when each item gets its own unique vocal signature.

Third, the effect survived even when the produced items were nonwords. Experiment 6 used pronounceable nonsense syllables. Participants read some aloud and some silently. The nonwords read aloud were recognized much better. This mattered because it ruled out a semantic explanation. You do not understand a nonword any deeper by saying it aloud. Something else was going on.

Fourth, and most strikingly, production stacked on top of generation. In Experiment 7, participants both generated and produced items. The combination beat generation alone. Production added a bonus even when deeper processing was already happening.

The paper was published in the Journal of Experimental Psychology: Learning, Memory, and Cognition. It was dedicated to the memory of Norman Slamecka. And it changed the field. Within six years, more than forty papers on the production effect appeared, culminating in a 2016 special issue of the Canadian Journal of Experimental Psychology, guest-edited by Glen Bodner and MacLeod [8].

Why It Works: Two Theories, One Argument

What makes produced items easier to remember? Two explanations have dominated the debate since 2010. Neither has won completely.

The distinctiveness account, championed by MacLeod and most of the Waterloo group, argues that production adds unique sensory and motor information to a memory trace [9]. When you say a word aloud, three things happen simultaneously. Your motor cortex fires an articulatory program. Your auditory cortex processes the sound of your own voice. And your brain tags the experience as self-produced. The word now has extra features that silently read words lack. These features make the produced word stand out from its neighbors, like a red marble in a bag of blue ones. At test time, you can use these distinctive features to confirm: yes, I studied this word. I remember saying it.

This explanation neatly accounts for the mixed-list requirement. In a mixed list, produced items stand out against a backdrop of silent items. But in a pure list where everything is produced, there is no contrast. Every marble is red. Distinctiveness disappears.

The strength account, proposed by Bodner and Taikh in 2012 [10], takes a different angle. Maybe production simply creates a stronger memory trace. More processing, more attention, more engagement. The word gets encoded more deeply, not just more distinctively.

The evidence? Amanda Fawcett's 2013 meta-analysis in Acta Psychologica [11] pooled data from between-subjects studies where one group produced everything and another read silently. Even without the mixed-list contrast, there was still a reliable (though smaller) production advantage: g = 0.37 for hit rates and g = 0.50 for discriminability. If distinctiveness were the whole story, pure-list production should show nothing. It does not show nothing. It shows something smaller.

The modern consensus, articulated by Fawcett and Ozubko in 2016 and by MacLeod and Bodner in their 2017 review [9], is that both mechanisms contribute. In mixed lists, distinctiveness dominates and drives the large effect through recollection, the vivid "I remember saying that" experience. In pure lists, strength provides a smaller boost through familiarity, the vaguer "this seems more familiar" feeling.

Think of it this way. Saying a word aloud is like highlighting a sentence in a textbook. If you highlight a few key sentences, they stand out. That is distinctiveness. But highlighting also makes the ink slightly bolder, slightly more visible. That is strength. Highlight everything and nothing stands out anymore, but the ink is still a bit bolder everywhere.

The Brain During Production

For years, the production effect was studied entirely through behavior. Recognition rates. Recall percentages. Hit rates and false alarm rates. But MacLeod and Bodner's 2017 review explicitly called for neuroimaging studies [9]. The brain had to weigh in.

The first EEG evidence came from Hassall, Quinlan, Turk, Taylor, and Krigolson in the 2016 special issue [12]. They recorded electrical brain activity while participants studied words aloud, silently, or by singing. The P300 component, a positive voltage wave peaking around 300 milliseconds after stimulus onset and long associated with attention and distinctive encoding, was significantly larger for aloud and sung words compared to silent ones. The brain was processing produced items differently at the moment of encoding.

The defining brain imaging study arrived in 2021. Heather Bailey and a large team at Dalhousie University used functional magnetic resonance imaging (fMRI) to watch the brain during production [13]. Participants studied words in three conditions: reading aloud, reading silently, or saying "check" while seeing the word (a motor control condition). The fMRI results were strikingly clean. Aloud reading activated primary sensorimotor cortex, the strip of brain tissue running across the top of the head that controls mouth and tongue movements. It also activated auditory cortex, the region on the side of the brain that processes sound, including the sound of one's own voice. The "check" condition activated motor cortex too, but not auditory cortex in the same way, because saying the same word every time does not create a unique auditory signature for each item.

Most importantly, these activations during encoding predicted successful recollection at test. The more motor and auditory cortex fired during study, the more likely the participant was to later say "I remember" rather than just "this seems familiar."

The most recent evidence comes from Zhang, Abdullah, Yan, Hou, Chen, and McLaren, who in 2025 published an EEG study using multivariate pattern analysis (MVPA) [14]. Participants studied words aloud or silently and then took a remember/know/new recognition test. The Late Positive Component (LPC), a brain wave appearing 500 to 800 milliseconds after seeing a test word and associated with recollection, was significantly larger for words originally read aloud. The FN400, an earlier component linked to familiarity, showed no difference. This pattern supports the distinctiveness account: production enhances recollection, the rich "I remember" experience, not just vague familiarity.

Even more striking, MVPA could decode whether a word had been studied aloud or silently from EEG patterns alone, with peak classification accuracy between 760 and 840 milliseconds. The brain carries a detectable neural signature of production that persists from encoding through retrieval.

Not Just Reading Aloud: The Many Faces of Production

Reading aloud is the most studied form of production. But it is not the only one. Researchers have tested a remarkable variety of productive acts, and nearly all of them beat silent reading.

Noah Forrin, MacLeod, and Jason Ozubko published a key study in 2012 titled "Widening the boundaries of the production effect" [15]. They tested spelling aloud, handwriting, and typing. The results were strikingly consistent. Spelling aloud improved recognition by about 11.5 percentage points. Handwriting by about 13 points. Typing by about 11.4 points. The authors called this consistency "remarkable" because the motor acts involved are completely different. Moving your lips, moving a pen, and pressing keys use different muscles, different brain circuits, different feedback loops. Yet the memory benefit was nearly identical.

Quinlan and Taylor pushed the boundaries further in 2013 [16]. They had participants sing words instead of speaking them. Singing produced a bigger effect than speaking. And speaking loudly produced a bigger effect than speaking at normal volume. The pattern suggested a gradient: the more distinctive the production, the better the memory. Singing is more unusual than talking, and talking loudly is more unusual than talking normally. More unusual means more distinctive. More distinctive means more memorable.

Even imagining production works. Jamieson and Spear found that merely imagining typing a word improved memory above silent reading, though less than actually typing it [12].

And then there is drawing. Jeffrey Wammes, Melissa Meade, and Myra Fernandes at the University of Waterloo reported in 2016 that drawing words produced an even larger memory advantage than writing them [17]. In their experiments, participants recalled "more than twice as many drawn than written words." Drawing combines motor production, visual processing, and elaborative encoding into a single act. It is production with extra channels.

The general gradient, compiled across studies, looks something like this:

The takeaway is consistent. Any act that transforms information from a passive visual input into an active motor output creates a more memorable encoding. The richer and more distinctive the transformation, the larger the benefit.

When Production Fails: Boundary Conditions

No memory effect works everywhere. The production effect has clear boundaries, and understanding them matters more than celebrating the successes.

The most important boundary is the mixed-list requirement. As Hopkins and Edwards noticed in 1972 and MacLeod confirmed in 2010, the large production effect depends on mixing produced and silent items in the same study list [1]. In pure lists where everything is produced or everything is silent, the effect shrinks dramatically. Fawcett's meta-analysis found a reliable but moderate between-subjects effect of g = 0.37 [11]. This means studying everything aloud is better than studying everything silently, but only modestly. The big gains come from strategic selection: produce the important stuff, read the rest silently.

Free recall tells a more complicated story. Travis Jonker, Meredith Levene, and MacLeod reported in 2014 that the production effect was absent in pure-list free recall [18]. Their explanation: production helps item memory (recognizing individual words) but disrupts order memory (remembering the sequence of items). Free recall depends heavily on remembering associations between items, not just individual items. In a mixed list, the item benefit outweighs the order cost. In a pure list, they cancel out.

Another limit emerged in 2024. Benjamin Roberts and colleagues, including MacLeod, tested whether production helps readers understand text, not just remember it [19]. They used Nelson-Denny reading passages and gave participants multiple-choice questions. Some questions tested memory (could you remember specific facts from the passage?) and others tested comprehension (could you draw inferences and understand the argument?). Production improved memory-focused scores. It did not improve comprehension scores at all. Zero benefit for understanding.

This is a critical finding. Saying things aloud helps you remember what was said. It does not help you understand what it meant. If you need to memorize drug names for a pharmacology exam, production is your friend. If you need to understand a philosophical argument, you need a different strategy.

Hourihan and Smith found another surprising failure in 2016 [20]. Production did not help participants remember face-name associations. Saying a name aloud while looking at a face did not improve the ability to later match faces with names. The effect seems tied to verbal items, not associative binding.

It Gets Personal: The Self-Reference Gradient

One of the most elegant findings in the production effect literature came from a 2011 paper with a catchy title: "I said, you said: The production effect gets personal" [21].

MacLeod tested whether hearing someone else produce a word was as good as producing it yourself. It was not. Self-production gave a substantially larger memory benefit than hearing another person say the word. Hearing someone else helped, but not as much. The effect was personal.

Forrin and MacLeod deepened this finding in 2018 with a study titled "This time it's personal: the memory benefit of hearing oneself" [22]. They added a clever condition: hearing a recording of your own voice reading the word. This dissociated two components that are normally fused in live production. When you read aloud, you simultaneously produce the motor act and hear yourself. By playing back a recording, the motor component is removed and only the self-referential auditory component remains.

The results revealed a clean gradient. Silent reading was worst. Hearing someone else was better. Hearing a recording of yourself was better still. And reading aloud in real time was best of all. The study used 95 participants and tested them two weeks after the initial exposure. Even after a two-week delay, the gradient held.

What this means is that the production effect draws on two independent sources. One is motor: the physical act of moving your mouth, tongue, and vocal cords. The other is self-referential: hearing your own voice and recognizing it as yours. Both contribute. When you read aloud, you get both. When you hear a recording, you get only the self-referential component. When you hear someone else, you get neither.

This has practical implications. If you cannot read aloud (in a library, on a train), recording yourself reading your notes and then listening to the recording captures part of the benefit. It is not as good as live production. But it is measurably better than silent review alone.

Production in Children and Across the Lifespan

Does the production effect work for children? The question matters because children are the population most likely to benefit from simple, evidence-based study strategies.

Vanessa Pritchard, Bianca Heron-Delaney, Stephanie Malone, and MacLeod tested this directly with children aged 7 to 10 in Brisbane, Australia [23]. In Experiment 1 (n = 41), children studied familiar words in mixed and blocked lists. In Experiment 2 (n = 40), they studied novel nonwords. The production effect was reliable in both experiments and comparable for words and nonwords. Even young children produce a distinctive encoding record when they say things aloud.

A more surprising result came from younger children. Lopez Assef, Desmeules-Trudel, Bernard, and Zamuner studied children aged 2 to 6 in 2021 [24]. They found a developmental reversal. In very young children, the production effect sometimes went in the opposite direction. The cognitive machinery for production-based distinctiveness appears to mature during early childhood.

At the other end of the lifespan, Lin and MacLeod tested older adults in 2012 [25]. The production effect was reduced but not eliminated. This fits a broader pattern: distinctive processing becomes less effective with age, but it does not stop working entirely. Icht, Mama, and Ben-David confirmed in 2022 that production still improved text memory in older adults across both visual and auditory presentation modalities [26].

Clinical populations show benefits too. Mama and Icht tested adults with ADHD in 2018 [27]. Vocalization improved verbal learning regardless of whether participants were on or off methylphenidate. The effect did not interact with medication status. This suggests production taps into encoding mechanisms that are preserved even when attention is impaired.

Production, Generation, and the Family of Active Learning

The production effect does not exist in isolation. It belongs to a family of encoding techniques that share a common thread: doing something active with information is better than passively receiving it.

The closest relative is the generation effect, discovered by Slamecka and Graf in 1978 [4]. In the generation effect, you produce the item itself from a cue. For example, you see "hot, c___" and generate "cold." In the production effect, the item is already given to you; you just read it aloud. Generation requires more cognitive effort and semantic processing. Production requires less. Yet both yield similar-sized benefits of roughly 10 to 20 percent [9].

MacLeod's Experiment 7 in the 2010 paper showed something important: production stacks on top of generation. Items that were both generated and produced were remembered best. The two effects tap into different mechanisms. Generation improves memory through semantic elaboration. Production improves memory through sensory-motor distinctiveness. Combine them and you get both.

The testing effect is a more distant cousin. Retrieval practice improves memory by forcing the brain to reconstruct a memory trace during recall [28]. Production, by contrast, operates during encoding, not retrieval. But the two combine naturally. When you use flashcards and say the answer out loud, you are stacking production on retrieval practice. When you then review those cards on a spaced schedule, you are adding spaced repetition to the mix.

An important clarification from Forrin, Jonker, and MacLeod: production is not simply a levels-of-processing effect [9]. Levels of processing predicts that deeper semantic processing leads to better memory. But production works equally well after shallow and deep processing tasks. It adds a sensory-motor dimension that is orthogonal to depth. You can process a word deeply (thinking about its meaning) and then produce it aloud, and the production adds an additional boost on top of the depth benefit. The two are independent.

What separates production from Bjork's "desirable difficulties" framework is that production is not really difficult. It takes almost no extra time. It requires almost no extra effort. Yet it yields a substantial memory benefit. It may be the easiest effective study technique ever documented.

The Ongoing Debate: What Production Cannot Do

Every good science story needs a section on what we do not know and where the evidence pushes back.

The biggest open question is whether distinctiveness or strength (or some combination) best explains the production effect at a neural level. The 2025 EEG study by Zhang and colleagues [14] found that the LPC, which indexes recollection, was enhanced for produced items. The FN400, which indexes familiarity, was not. This supports distinctiveness. But Fawcett and Ozubko's 2016 dual-process account argues familiarity also contributes, at least in between-subjects designs [29]. The debate is not settled.

Another concern: most production effect research uses word lists. Real students do not memorize word lists. They read textbook chapters, attend lectures, and review notes. Roberts and colleagues' 2024 finding that production benefits memory but not comprehension of text passages is sobering [19]. It means production is a retention tool, not a learning tool. It helps you hold onto what you already encountered. It does not help you understand it in the first place.

Long-term retention data is also limited. Ozubko, Hourihan, and MacLeod showed the effect persists for at least one week [30]. Forrin and MacLeod showed it holds across a two-week delay [22]. But what about a month? A semester? A year? We do not have data beyond a few weeks.

Metacognitive issues add another layer. Castel, Rhodes, and Friedman showed in 2013 that people misuse production as a cue for judgments of learning [31]. After studying a mixed list, participants estimated they would remember silent items better than they actually did. Production improved actual memory but distorted predicted memory. Students might over-rely on feelings of familiarity for the items they did not produce.

A curious finding from 2022 showed that the production effect disappears for unusual voices unless those voices appear frequently during study [32]. If you speak in a strange voice for every produced item, the distinctiveness of production is wiped out by the distinctiveness of the voice itself. The effect needs a "normal" backdrop against which produced items stand out.

And the directed forgetting interaction reported by Spear, Reid, Guitard, and Jamieson in the 2024 special issue [33] showed that production's distinctiveness component is largely resistant to intentional forgetting. You can tell someone to forget a produced word. They will be less able to comply than with a silently read word. Production creates a memory that resists erasure.

What This Means for the Way You Study

The practical implications of fifty years of production effect research can be stated simply. But simple does not mean obvious, and obvious does not mean widely practiced.

First: produce the most important material, not everything. The mixed-list structure is what creates the biggest benefit. If you read everything aloud, the advantage shrinks. Select the terms, definitions, or formulas that matter most and say those aloud. Let the supporting material stay silent. This creates contrast, and contrast is what drives distinctiveness.

Second: make each production unique. Saying "okay" after each item does not work [1]. Each item needs its own vocal fingerprint. Read the actual word, definition, or formula aloud.

Third: combine production with retrieval practice. Make flashcards and say the answer aloud before flipping the card. This stacks production on the testing effect. Both techniques are backed by decades of evidence, and they target different aspects of memory. Production strengthens encoding. Retrieval practice strengthens consolidation.

Fourth: combine with spaced repetition. Spacing your reviews over increasing intervals is among the most effective study techniques ever documented [28]. Add production to each review session and you are attacking memory from two directions simultaneously.

Fifth: do not expect production to replace understanding. Roberts and colleagues' 2024 finding is clear: production helps you remember facts, not understand arguments [19]. For comprehension, use elaboration, self-explanation, or the Feynman technique. For retention, use production.

Sixth: writing and drawing work too. If you cannot speak aloud, write key terms by hand or sketch them. Forrin and colleagues showed handwriting yields approximately 13 percent improvement [15]. Wammes and colleagues showed drawing can more than double free recall [17].

Seventh: your own voice matters. If you are reviewing on the go, listening to recordings of yourself reading your notes captures part of the production benefit [22]. It is not as powerful as live production. But it beats passive listening to someone else.

The Next Fifty Years

The production effect story is far from over. Three research directions are likely to define the next decade.

The first is better neuroimaging. Bailey and colleagues' 2021 fMRI study [13] localized the effect to motor and auditory cortex. But hippocampal contributions remain unclear. The hippocampus is the brain's memory factory, and recollection-based effects like the production effect almost certainly involve it. High-resolution fMRI targeting the hippocampal subfields during production encoding is the obvious next step.

The second is computational modeling. Several teams have already implemented production within formal memory models. Jamieson and colleagues used MINERVA 2. Kelly and colleagues used the Retrieving Effectively from Memory model. Saint-Aubin and colleagues used a revised Feature Model [34], [35]. All successfully capture key findings. But no model yet accounts for the full pattern: the mixed-list advantage, the gradient from mouthing to singing, the comprehension limitation, and the self-reference component. A unified model would be a theoretical milestone.

The third is classroom implementation. Almost all production effect research has happened in laboratories with word lists. Bringing it into real classrooms, with real curricula and real students, is the critical translation step. Early work with children [23] is encouraging. But large-scale randomized controlled trials in K-12 and university settings are needed. How much time does production add to a study session? Does the benefit survive across a full semester? Does it transfer to different types of exams?

The production effect is simple. Say it out loud. Write it down. Draw it. Sing it. Do something with the information instead of letting it pass through your eyes into oblivion. Fifty years of research, from a forgotten two-page note in 1972 to multivariate EEG decoding in 2025, converge on the same conclusion. The act of production transforms passive exposure into active encoding. And active encoding is the only kind that lasts.