Dopamine has the worst public relations of any molecule in the brain.
It’s been called the pleasure chemical, the feel-good hormone, the reward molecule. Every one of these is wrong, and the L&D industry has built a great deal of confident advice on top of them.
Indeed, if you’ve read that gamification works by triggering dopamine hits, or that learners need a dopamine boost to stay engaged, then you’ve read something nobody has properly measured.
The real story is stranger and considerably more useful. Dopamine is not about how good the reward feels. It’s about the gap between what you expected and what you got, and how hard you were willing to work to close it.
Once you understand that, you’ll soon arrive at a set of design decisions that have nothing to do with making training fun. So let’s get started. Here’s what dopamine is actually doing.
What is Dopamine?
Dopamine is a neurotransmitter: a chemical messenger that carries signals between neurons. It’s not a hormone, despite its “feel-good hormone” label. It was first identified as a signalling molecule in its own right by the Swedish pharmacologist Arvid Carlsson in 1957. This work won him a Nobel Prize.
The cells that produce it sit in two small regions of the midbrain, the ventral tegmental area and the substantia nigra. From there they project widely: into the striatum, the prefrontal cortex, and the hippocampus. That reach is why dopamine turns up in accounts of movement, motivation, decision-making, and memory.
It’s also why single word summaries of what dopamine does (read: “pleasure”) are always wrong. The same chemical carries different messages to different places.
What matters for learning is the traffic between the midbrain and the hippocampus, the brain’s memory hub. That is the circuit the rest of this article is about.
What Dopamine Actually Signals
In 1997, Wolfram Schultz, Peter Dayan, and Read Montague published the finding that reorganised the field. Recording from dopamine neurons in monkeys, they found the cells were not responding to reward at all. They were responding to the difference between predicted and actual reward.
The signal has three states, and Schultz still states them the same way:
- Activity increases when a reward is better than predicted.
- Activity is unchanged when the reward is exactly as predicted.
- Activity is briefly depressed when the reward is worse than predicted.
This is what the field calls a reward prediction error. Dopamine encodes the gap between expectation and outcome, not the outcome itself. The bigger the gap, the bigger the signal. Close the gap entirely and there is no signal at all.
Sit with that middle state for a moment. After all, it’s the one that matters most for training design. A fully expected reward produces no dopamine response. None. The badge a learner knew they were getting for completing module four is, neurochemically speaking, nothing at all.
What’s more, when a predicted reward failed to arrive, the neurons fell below their baseline firing rate at exactly the moment the reward should have appeared, with no external cue to prompt it. The brain knew when it had been let down.
Is Dopamine The Pleasure Chemical?
The short answer is no, and it has been no for 25 years. Dopamine drives the pursuit of a reward, not the enjoyment of getting it.
Wanting Versus Liking
Kent Berridge and Terry Robinson spent three decades separating two things everyday language treats as one. Their 2016 summary is blunt: dopamine mediates “wanting” but not “liking” for the same reward.
Deplete dopamine in a rat and it still shows every sign of enjoying a sweet taste. What collapses is the pursuit. It will not work to get more of what it liked.
They put the state of the field even more directly: “It is now rather rare to find an affective neuroscientist studying reward who still asserts that dopamine mediates pleasure.”
Dopamine and the Willingness to Work
John Salamone’s work adds the piece an L&D audience should care about most. Interfering with dopamine transmission produces what his team calls a low-effort bias: animals shift towards options that require less work, while their appetite for the reward itself stays intact.
Raise dopamine transmission and they shift back towards high-effort options. The effect is bidirectional, and it’s about willingness to exert effort rather than enjoyment.
That distinction separates dopamine from the thing L&D usually pairs it with. Enjoyment is not the mechanism. Willingness to work is, which puts dopamine much closer to intrinsic motivation than to entertainment.
Dopamine’s Double Job
Joshua Berke’s review makes the synthesis clean. Dopamine does two jobs that point in opposite directions in time. Motivation looks forward, using predicted value to energise what you do next. Learning looks backwards, updating the value of what you just did.
Popular accounts collapse the two into a single feel-good signal, which is how you end up with advice that just doesn’t work.
A 2024 perspective from seven principal investigators gives the current state of the argument. Their verdict on the prediction error account is that it is “probably too simple” in its original form, though they go on to defend a generalised version of it.
How Does Dopamine Affect Memory?
Anticipation improves encoding, and surprise improves retention. Both effects happen before or around the material rather than inside it.
Dopamine’s link to memory runs through the traffic between the midbrain and the hippocampus. When that circuit is active, the hippocampus encodes more readily.
What’s interesting for learning design is when it activates: not while the learner is studying, but in the moments before the material arrives, and at the point where an outcome fails to match what they expected.
Four studies map this territory.
| Study | What It Found | Sample |
| Adcock et al. (2006) | Scenes preceded by a high-value cue were better remembered a day later. The predictive brain activity occurred before the material appeared. | ~12 |
| Poh et al. (2022) | What predicts memory is not the size of the midbrain signal but whether the hippocampus settles into a particular configuration during anticipation. | 23 |
| Bowen & Madan (2025) | Recall of high-value material ran at 65% against 27% for low-value material. Learners remembered more than twice as much of what they had been told mattered. | 339 |
| Rouhani & Niv (2021) | The magnitude of surprise improved memory for the outcome, whether that outcome was good or bad. | 750+ |
The first two studies are small, but their direction has held up in the larger behavioural work below them. Read together, they say something practical: readiness to learn is a real, momentary brain state, and it is established before the content starts.
In other words: 10 seconds spent telling a learner why a section matters is not preamble. It’s part of the encoding event itself.
The final study is the one that should change how you build your assessments. Rouhani and Niv found that being wrong improves memory for the correct answer. That’s right. Not being rewarded. Being wrong.
That’s the strongest argument in neuroscience for prediction-then-reveal, for confidence-weighted quizzing, and for making learners commit to an answer before they see the right one. The brain learns at the gap and you cannot open a gap without letting someone be wrong inside it first.
This is also the mechanism that underlies retrieval practice and the generation effect, which between them are the best-evidenced techniques in learning science. Both work by forcing a prediction before supplying an answer.
Dopamine and Curiosity: Does Curiosity Improve Learning?
Curiosity and learning have a curious relationship. It does improve learning, but only for the thing the learner is curious about. Not for anything else you put nearby. This is often where most advice on this topic goes awry.
Indeed, this is one of the most popular corners of dopamine research, and it’s often the one that L&D writes about most carelessly. The reason for this is clear. The headline finding is robust and the exciting one next to it is not.
The Robust Finding: Curiosity Improves Retention
Gruber, Gelman, and Ranganath had participants rate trivia questions for curiosity, then tested what they remembered.
| High curiosity | Low curiosity | |
| Recall, same session | 70.6% | 54.1% |
| Recall, one day later | 45.9% | 28.1% |
Curiosity about the material substantially improves retention of that material, and the advantage widens over a day, rather than fading. That much is solid.
The Oversold Finding: The Spillover Effect
In the same study, neutral faces shown during the wait for an answer were also better remembered when curiosity was high. A spillover effect. This produced a decade of spurious advice about opening a curiosity gap so everything in the vicinity sticks.
However, the effect was thin from the start. Nineteen participants, 42.4% against 38.2%. After a one-day delay it only survived for faces the participants were confident about.
The Update
A 2024 study of 239 people, co-authored by Gruber himself, ran the same design with scholastic facts instead of faces. Memory for those facts was consistently worse after high-curiosity questions, across all three versions of the task.
A face is cheap to encode. On the other hand, a fact about history competes for exactly the capacity curiosity has already commandeered. Curiosity is not a general enhancer. It’s a filter, and it filters in favour of the answer the learner is waiting for.
If that sounds familiar, it should. It’s the seductive details effect wearing a different hat: interesting material crowding out the material that matters. This means you should only open a curiosity gap to teach the thing the learner is curious about, not to use the gap as a delivery vehicle for something else.
Can Designing for Dopamine Backfire?
Recall the middle state of the prediction error: a fully expected reward produces no response. That has an uncomfortable consequence.
Award points or badges on a schedule and the first few will land. Then expectations adjust, the gap closes, and the signal goes quiet. To get a response back you have to exceed the new expectation. So you escalate, or you randomise.
- Escalate and you inflate, until the reward that motivated in January is baseline by March.
- Randomise and you have built a slot machine, not a learning system.
The Overjustification Effect
There is a second cost, and it is better evidenced. Deci, Koestner, and Ryan’s meta-analysis of 128 experiments found tangible, expected rewards reduced free-choice intrinsic motivation at d = −0.36. Verbal recognition did the opposite.
This is the overjustification effect. Extrinsic rewards for work people already enjoy is a bad trade. The reward fades as expectation catches up, and you spend some of their interest buying it. This isn’t an argument against ever using them. Compliance content nobody will ever enjoy is where they earn their keep.
None of this means dopamine is harmful or that anyone needs a detox. It means a system built on prediction error will exhaust any predictable reward you feed it. That is a design problem, not a health problem. We cover this wider topic in the dark side of gamification.
5 Dopamine Myths in Learning and Development
Dopamine is one of the most misused words in L&D. Here are five claims that are worth retiring, and what the evidence actually says.
1. Does a Dopamine Detox Work?
No, because there is nothing to reset.
In L&D this often shows up as the case for stripping stimulation out of learning, on the theory that over-stimulated learners need a quieter environment to recover their capacity to focus.
However, as Harvard Health puts it, dopamine “doesn’t actually decrease when you avoid overstimulating activities, so a dopamine ‘fast’ doesn’t actually lower your dopamine levels.” There’s nothing to deplete and therefore nothing to reset.
That said, screen breaks do help people. But they do this by building a different habit, not by topping up a chemical.
2. Do Smartphones Create Dopamine Loops?
Interruption genuinely damages learning, and the evidence is solid enough that we’ve given it its own article. What has never been measured is dopamine during smartphone use. Not once, in any human study.
The study everyone reaches for is Sherman and colleagues, who found that photos with more likes produced greater activity in reward-related brain regions. However, those were fMRI results, which measure blood flow. They tell you whether a region was busy and nothing about which neurotransmitter was involved.
The distinction is practical. If the problem is a dopamine loop, the answer is abstinence. If the problem is interruption, the answer is protected attention and shorter units.
3. Is There a Dopamine Pleasure-Pain Balance?
The intuition here is sound, but the evidence is misplaced. The model, popularised by Dopamine Nation, says pleasure and pain share the same circuitry: every hit tips the balance, the brain tips it back, and repeated stimulation resets the baseline downward into a “dopamine deficit state”.
However, prediction error explains this pattern away without needing a deficit state. Indeed, the deficit evidence comes from PET studies of cocaine and methamphetamine users, not those completing eLearning experiences. Even for drugs the accounting is loose: heroin produces a euphoric high with no measurable rise in striatal dopamine at all.
4. Does Gamification Release Dopamine?
Dopamine release during video game play has been measured exactly once, by Koepp and colleagues in 1998: eight men, a tank game, played for money. The study has never been replicated. The closest anyone has come in a learning context is fMRI evidence of ventral striatal blood flow during a competitive game.
As mentioned above, this measures blood flow, not dopamine production.
What gamification does have is measured effects on learning. Sailer and Homner, across 38 studies, found g = 0.49 on cognitive outcomes, 0.36 on motivation, 0.25 on behaviour. All three effect sizes are small, but they’re worth having. The study’s authors linked them to feedback, autonomy, and social relatedness, not dopamine.
5. Do Variable Rewards Motivate Learners?
Uncertain rewards do sustain effort, but with a condition its popularisers often skip: the effect appears only when people focus on the process of pursuing a reward, and reverses when they focus on the prize.
The industry that has best perfected random rewards is the loot box business, whose spending correlates with problem gambling about as strongly as alcohol dependence does. Borrow the mechanism, borrow the ethics conversation.
How to Apply Dopamine Research to Learning Design
Everything below follows from one sentence: the brain learns best at the gap between prediction and outcome. Close the gap and you get nothing. Open it and you get retention.
- Use Prediction Powers: Ask for a commitment from the learner, then show the answer. A wrong guess opens a large gap, and larger gaps produce better memory for the correction. As such, make sure the prediction is genuinely difficult.
- Audit Your Reward Schedule: Anything awarded on a foreseeable schedule produces no signal. Points for finishing a module merely meet learners’ expectations. Reserve serious rewards for genuine achievements rather than mere attendance.
- Frame Before You Teach: The brain state going in determines what gets kept, and it is set in the seconds before the material appears. That is not a learning objectives slide. It is one sentence on why this matters to this person, or a question they cannot yet answer.
- Use Curiosity Effectively: Open a gap about the target content, then fill it. Do not use the gap to deliver something else, because demanding material arriving mid-curiosity gets crowded out.
- Design for Feedback: If gamification’s measured effects come from feedback, autonomy, and relatedness (see above) then build for those. Points are a wrapper around feedback. The feedback itself is doing the work they get credit for.
- Stop Optimising for Enjoyment: Dopamine tracks willingness to exert effort, not pleasure. Satisfaction scores measure the wrong variable, and the conditions that produce the highest ratings often produce the weakest retention.
Final Words
The pleasure framing of dopamine has survived because it flatters a comfortable theory of engagement: make learning enjoyable and the brain will reward you with retention.
However, the evidence points somewhere less comfortable and more interesting. Dopamine responds to surprise. It also drives the effort you are prepared to spend on a task. Neither of those is the same as enjoyment.
This is the neurogogy position in miniature. Design for the mechanism, not the metaphor, and be willing to retire claims that do not survive a check.
There we have it. Stop trying to make learning feel good and start making it unpredictable.
Ready to build learning that lasts?
- Book a demo to see how the Impact Suite builds prediction, retrieval, and reinforcement into training by default.
- Explore the Impact Suite and the science behind every feature.
- Download the Guide to The Science of Learner Engagement for the full picture of what makes learning stick.
