Classical Conditioning: An S-Tier Behavioral Designer’s Guide to Pavlov

Every product designer has a moment they cannot explain. The phone buzzes on the desk, you feel a small pull of anticipation, and by the time you have checked the screen the buzz is already over. Or you walk past a particular coffee shop and, without deciding to, your body has started budgeting for a latte. Or the Slack sound plays in a movie soundtrack and your pulse jumps before your brain has even finished processing that it isn’t your Slack. These reactions are not products of conscious choice, and most of the behavioral-economics literature that currently dominates product discussion — loss aversion, framing, defaults — cannot explain them. Those models live upstairs, in deliberate judgment. These reactions happen downstairs, in the associative layer that fires before deliberation is even online.

The oldest and still most powerful model of that downstairs layer is classical conditioning, the framework Ivan Pavlov built out of his Nobel-prize-winning digestion research at the turn of the twentieth century. Pavlov discovered that a dog’s body could be trained to treat a neutral sound as if it were food — and once the lab had documented that, the entire theory of learned anticipation fell out of the same apparatus. A century later, the cognitive revolution, the neuroscience of prediction error, and the behavioral-design toolkit of every notification-driven product on your phone all sit on top of Pavlov’s experimental floor.

This post is the designer’s guide to classical conditioning — not as a historical curiosity, but as the pre-rational engine under Core Drive 7 (Unpredictability & Curiosity) and Core Drive 6 (Scarcity & Impatience). We will cover what Pavlov actually demonstrated, why the Rescorla-Wagner and Garcia revisions quietly rewrote the theory without replacing it, where the original model breaks, and how the Octalysis Framework turns the defensible core into engagement systems that work at the same reflex layer your designs are already reaching without realising it.

Speed Run Notes

About the Author

Yu-kai Chou is an S-Tier Behavioral Designer and the creator of the Octalysis Framework, the gamification design system now applied to products and experiences reaching over 1.5 billion users. His book Actionable Gamification is one of the most-cited works in the field, and he has been ranked the #1 Gamification Guru in the World.

He has advised MrBeast, LEGO, Microsoft, Porsche, Tesla, Stanford, Harvard, and governments including Ukraine on turning behavioral psychology into product mechanics that actually change user behavior.

Verify: Wikipedia · Google Scholar · Wikidata · LinkedIn

What Is Classical Conditioning?

Classical conditioning is the psychological process by which a previously neutral cue acquires the power to trigger a response that a biologically significant stimulus already reliably triggers. Put in the tight five-term vocabulary Pavlov bequeathed to every psychology textbook: an unconditioned stimulus (US) naturally elicits an unconditioned response (UR); pair that US repeatedly with a neutral stimulus (NS), and the neutral stimulus itself becomes a conditioned stimulus (CS) that triggers a conditioned response (CR). Food in the mouth is the US; salivation is the UR; the metronome is the NS before training and the CS after. The CR — salivation to the metronome — is the observable proof that the organism has learned something it previously did not know.

The precise framing matters because the popular summary — “ring a bell enough times and the dog will salivate” — misses the two features that make classical conditioning such a durable theory of behavior. The first is that the learning is about anticipation. The CR is not the dog’s reaction to the bell as an object; it is the dog’s preparation for the food the bell predicts. Modern cognitive neuroscience spent decades restating this insight in the language of predictive coding, but Pavlov had it by 1903. The second feature is that the learning happens at the level of the autonomic nervous system, not the deliberative one. You cannot decide to not salivate at the bell, any more than you can decide to not flinch when you step on a thumbtack. Conditioned responses are not opinions; they are installed behaviors.

That installation layer is what makes classical conditioning so consequential for design. Most of the frameworks product teams talk about — nudge theory, framing effects, prospect theory — describe how users make decisions in the slow, reflective System-2 layer Daniel Kahneman popularised. Classical conditioning describes what your users are doing in the fast, automatic layer before they start deciding anything at all. If you have ever wondered why users continue to flinch toward a push notification they know will waste their time, the answer is not in Kahneman. It is in Pavlov.

There are two natural misreads of the framework that are worth flagging immediately. The first is that classical conditioning is the same thing as Skinner’s operant conditioning. It is not. Operant conditioning is how organisms learn that their voluntary behavior produces consequences — pressing a lever causes food to appear. Classical conditioning is how organisms learn that stimuli in the world predict other stimuli — the bell predicts food. The first is about action-consequence contingencies; the second is about stimulus-stimulus contingencies. In almost every real product, both are running simultaneously, and treating them as interchangeable is one of the most common rookie mistakes in behavioral-design practice.

The second misread is that classical conditioning is a primitive behavior-first model that the cognitive revolution has since rendered obsolete. That reading is roughly fifty years out of date. Robert Rescorla’s 1968 contingency experiments and the Rescorla-Wagner model that followed in 1972 turned classical conditioning into one of the earliest computational theories of learning, and the dopaminergic prediction-error framework that dominates systems neuroscience today is often framed as “Rescorla-Wagner in the basal ganglia.” The model has not been replaced; it has been upgraded.

The Core Findings of Pavlov (1901–1927)

Pavlov did not set out to found a theory of learning. He was a digestion physiologist, and his Nobel-prize-winning line of work in 1904 was on the nervous control of gastric secretions. The conditioning experiments emerged as a nuisance variable: Pavlov’s dogs started salivating before the food reached their mouths — in response to the assistant’s footsteps, the sight of the lab coat, the sound of the feeding apparatus. A less curious scientist would have called that noise and filtered it out. Pavlov named it psychic secretion, pointed the whole lab at it, and within twenty years had sketched the architecture of a theory that is still recognizably intact.

The original findings cluster into five claims that every subsequent treatment of classical conditioning has had to account for.

Acquisition. When a neutral stimulus (a metronome click, a bell, a light) is presented shortly before a biologically significant one (food powder on the tongue), the organism’s response to the US progressively transfers onto the NS. In the classic salivation paradigm, the number of drops of saliva elicited by the CS alone rises over trials and plateaus. The curve looks roughly like an inverted exponential, which is not an accident; the Rescorla-Wagner equation produces exactly that shape.

Extinction. If the CS is presented repeatedly without the US, the CR diminishes. Pavlov observed this as clearly as he observed acquisition. Crucially, he did not conclude that extinction was forgetting; the dog learned something new — that the CS no longer predicted the US — on top of what it already knew. That nuance becomes important when we get to the design section, because a lot of product teams assume that “users will habituate” means “the association disappears.” It does not.

Spontaneous recovery. After extinction, if the CS is presented again following a rest period, the CR often reappears — weaker than before, but present. Pavlov took this as evidence that extinction did not erase the original conditioning; it inhibited it. The inhibition is itself time-dependent and fragile, which is why an apparently extinguished response can come back days later with nothing but a rest interval intervening.

Generalization. A CR established to one CS spreads to similar stimuli, with the strength of the spread falling as similarity falls. A dog trained to salivate to a 1,000 Hz tone will also salivate, somewhat less, to an 800 Hz tone. In a human product context, this is why a brand’s sonic signature can carry emotional weight even in covers, parodies, and acoustic variants.

Discrimination. If one CS is reinforced and a similar CS is not, the organism learns to respond only to the reinforced one. Pavlov ran variants in which a tone of one frequency predicted food and a tone of a slightly different frequency did not; the dogs sharpened their discriminations until the differences were at the edge of auditory threshold, and then — strikingly — when pushed past that edge developed behaviors that Pavlov called “experimental neurosis.” Force an organism to discriminate beyond its capacity, and you do not just get confusion; you get behavioral breakdown. This is a live lesson for designers of noisy notification systems, and we will come back to it.

Two additional phenomena from the Pavlovian era are worth naming. Higher-order conditioning is the observation that once a CS has been established, a second neutral stimulus can be paired with the CS alone and acquire some of the CR. You can condition to a bell, then condition a light to the bell, and the light will eventually elicit the response without ever having been paired with food directly. That is how brand associations move from products to logos to typography. Timing matters as well: conditioning is strongest when the CS precedes the US by a short interval (hundreds of milliseconds to a few seconds, depending on the response system), and fails when the CS comes after the US or simultaneously with it. The nervous system is asking, “does this cue predict what comes next?” A cue that arrives after the event cannot predict it, and the organism does not learn.

What Pavlov Got Right

The standard undergraduate line is that Pavlov was a good empiricist but a theoretically limited one — that his S-R reflex framework had to wait for Tolman, Rescorla, and the cognitive revolution to become serious. That reading is a bit unfair. Three elements of the original Pavlovian program have aged well, and are worth listing before we inventory the parts that have not.

The phenomenon replicates. Acquisition, extinction, spontaneous recovery, generalization, and discrimination have been replicated in dogs, rats, pigeons, rabbits, humans, fruit flies, sea slugs, and, at a molecular level, in isolated neural circuits. The dependent variable changes — eyeblink, galvanic skin response, fMRI signal, gill withdrawal — but the structure of the learning survives. Few psychological phenomena from 1900 are still load-bearing in 2020s neuroscience. Classical conditioning is. The Eric Kandel lab won the 2000 Nobel Prize in Physiology or Medicine for, in part, working out the molecular biology of classical-conditioning-like learning in Aplysia. That is a long way from Pavlov’s dog bench, and it is the same phenomenon.

Pavlov was right that the relevant unit is anticipation, not stimulus-response reflex. Even in his most behaviorist-sounding prose, Pavlov keeps coming back to the idea that what the organism has learned is an expectancy. The CR is a preparation: salivation readies the digestive tract for food that is about to arrive. When modern predictive-coding theorists say “the brain is a prediction machine,” they are restating the thesis Pavlov reached empirically a century earlier. The behaviorist schools that followed him sometimes stripped the expectancy language out in the interest of operational cleanliness; the cognitive revolution had to put it back.

Pavlov built the first operational measurement of an internal state. Before the conditioning work, “anticipation” was a philosophical term. After it, anticipation had a unit of measure (drops of saliva per unit time), a dose-response curve (reinforcement schedule), and a decay function (extinction). That move — operationalising an internal state through an externally measurable response — is the methodological move the rest of experimental psychology learned from. Whatever your view of the philosophical commitments of Russian behaviorism, Pavlov handed the discipline a template for turning mental talk into measurement, and most of the subsequent century ran on that template.

If you are a designer, the takeaway from the “what he got right” list is that you are building on a platform that has been stress-tested more thoroughly than almost any other in the behavioral sciences. The questions about classical conditioning that are still unsettled are sophisticated questions at the boundary of learning theory and computational neuroscience. The basic claim — that neutral cues can acquire motivational power through pairing, and that organisms act on the predictions they have learned — is as close to a law as our field has.

Where Classical Conditioning Falls Apart

The version of classical conditioning taught in most introductory psychology courses — “pair a neutral stimulus with a biologically important one enough times and the neutral one becomes a trigger” — is not actually what the scientific literature concluded. Three waves of critique reshaped the theory between 1966 and the early 2000s, and a designer applying the framework today needs the post-critique version, not the 1927 version.

1. The “pair them and it works” reading is wrong — contingency beats contiguity

The most consequential critique came from Robert Rescorla. In a 1968 paper that is now required reading in any graduate learning-science seminar, Rescorla ran experiments in which the temporal pairing of the CS and the US was held constant but their statistical contingency varied. In one condition the CS reliably preceded the US (contingent); in another the CS preceded the US at the same rate but the US also showed up at the same base rate in the absence of the CS (non-contingent). Pure contiguity theory predicted the two conditions should condition equally; the dogs disagreed. Conditioning occurred when the CS informed the organism about the US, and did not occur when the CS was merely correlated in time with it. The relevant variable is not pairing but predictive information.

Rescorla and Wagner published the formal version of this insight in 1972. The Rescorla-Wagner model says the change in associative strength between a CS and a US on any trial is proportional to the difference between the US that was actually delivered and the US the organism was already predicting. Written out informally: ΔV = αβ(λ − V), where V is the current predictive value of the CS, λ is the amount of US delivered, and α and β are rate parameters. The organism learns until its predictions match the world, and then learning stops. This equation predicts the classic acquisition curve, predicts extinction, predicts the blocking effect (if CS-A already predicts the US, adding CS-B to the compound does not condition CS-B), and predicts overshadowing (if CS-A is more salient than CS-B in a compound, it soaks up most of the learning). It is still the single most influential equation in learning theory.

For designers, the Rescorla-Wagner upgrade has one practical consequence that matters more than any other. If your product presents a CS (a notification sound, a scarcity tag, an email subject line) in situations where the predicted outcome does not arrive, the organism is not simply failing to condition — it is learning that your CS is uninformative, and that learning generalises. A user who has been conditioned to treat your notification sound as noise is not a user who can be reclaimed by ringing the same bell louder. Rescorla-Wagner warned us about this decades before growth teams had to live it.

2. Not all stimulus pairings are equal — the Garcia effect and biological preparedness

The second major critique came from John Garcia. In a 1966 study that has become one of the most cited papers in the history of psychology, Garcia and Koelling presented rats with flavored water paired with either a loud noise or with gamma-ray-induced nausea that appeared hours later. A pure contiguity theory predicted that the short-delay, highly-paired noise condition should condition strongly, and the long-delay, low-contiguity nausea condition should not. The opposite happened. Rats learned the taste-nausea association in a single trial with a delay of several hours. They failed to learn the taste-noise association even with many pairings. The result was so counterintuitive that the leading behaviorist journals initially refused to publish it.

The interpretation is that evolution has biased which stimuli can be paired with which outcomes. Rats’ ancestors who could learn to avoid flavors that made them sick in a single experience, even at long delays, survived. Rats’ ancestors who tried to protect themselves from loud noises by never drinking that water again did not have the relevant selection pressure. The principle generalises: conditioning is not a blank-slate associative mechanism that couples any stimulus to any outcome at the same rate. Certain pairings are prepared (Seligman 1971) and condition easily; others are contraprepared and resist conditioning even under massive training.

For designers, preparedness is why certain cues are disproportionately sticky. A sound with perceptual properties close to human alarm calls (a phone buzzer, a doorbell, a siren) is prepared for conditioning to threat- and opportunity-related responses in a way that a visual icon change is not. A countdown timer exploits a preparedness stack around scarcity and loss that a long paragraph of text cannot touch. Part of why modern product design converges on a small set of cue types — buzzes, dots, red numerals, countdown clocks — is that those cues land on prepared associative substrates the organism cannot defend itself against as easily as it can defend itself against explicit argument.

3. Extinction does not erase the original learning — renewal, reinstatement, and the hidden inhibitor

The third critique is about what happens when you stop pairing. The original Pavlovian reading was that extinction trials reduce the CR and, eventually, undo the association. Forty years of follow-up work has shown this is a convenient simplification at best. Mark Bouton’s extinction research — including his work on renewal, reinstatement, and context-dependence — has demonstrated that the original conditioning is not erased by extinction. It is masked by a new, context-specific inhibitory learning that says “in this context, the CS no longer predicts the US.” Change the context, and the original CR returns. Wait long enough, and spontaneous recovery brings it back. Deliver the US once without the CS (reinstatement), and the CR resurges.

The designer-relevant consequence is significant. A product that has accidentally conditioned users to treat its push notifications as spam cannot undo that conditioning by simply sending fewer notifications. The association is still stored in the nervous system, gated by an extinction-specific inhibitor. Ship those notifications in a new context (a new device, an app redesign, a new onboarding flow), and the conditioned aversion re-activates. This is why brand-reset campaigns so often fail: the underlying association is not gone, it is merely suppressed in the old context. Fix requires new CS, not a wash of the old one.

Bringing the three critiques together, the defensible 2020s version of classical conditioning looks like this: organisms learn the predictive contingencies between stimuli, not their pairings; the strength of learning depends on an evolutionarily shaped preparedness structure that biases which pairings can form quickly; and once formed, those associations are permanent enough that extinction is better modelled as inhibitory overlay than as deletion. All three of these updates make the theory more useful to designers, not less, because they specify when and how to use it and when not to bother.

The Brain on Classical Conditioning

The neuroscience of classical conditioning has done something unusual in psychology: it has produced a mechanism that almost everyone in the field accepts. The canonical structure, first sketched in the 1990s and consolidated over the following twenty years, is that midbrain dopamine neurons in the ventral tegmental area and substantia nigra carry a reward prediction error signal — literally, the mathematical quantity (λ − V) in the Rescorla-Wagner equation. Wolfram Schultz’s group demonstrated this in awake monkeys across a sequence of papers in the 1990s, and Peter Dayan and Read Montague formalised it in a 1997 paper that tied the dopaminergic signal to the temporal-difference learning algorithms developed in reinforcement-learning theory.

In the typical experiment, an untrained monkey shown a drop of juice (the US) has a burst of dopamine firing at the juice delivery. After training, when a light (the CS) reliably precedes the juice, the dopamine burst shifts earlier in time — it fires at the CS and not at the juice. If on a given trial the juice is unexpectedly withheld, dopamine shows a dip at the time the juice was expected. The dopamine cells are not signalling reward; they are signalling surprise — reward more or less than predicted. That is exactly the signal Pavlov needed to explain why anticipation shifts to the CS, why the CR is strongest during acquisition, and why extinction produces a measurable negative learning signal rather than simple decay.

The classically-conditioned fear system runs on a parallel circuit, centered on the amygdala. Joseph LeDoux’s work on fear conditioning in rats established that a tone paired with a mild foot-shock produces measurable long-term potentiation in the lateral amygdala; silence those cells and the conditioning does not form; stimulate them and the conditioning can be induced without the tone. The molecular cascade — NMDA-receptor-dependent plasticity, CREB-mediated gene transcription, protein synthesis during consolidation — is the same cascade Kandel’s lab characterised in the sea slug Aplysia. Different animals, different response systems, the same cellular machinery.

The designer-relevant lesson from the neuroscience is that conditioning operates at a level of biological conservation that puts it well below the level most product conversations happen at. When a designer says “users will rationally weigh the cost and benefit of checking the notification,” they are describing System-2. When the user’s phone buzzes, the amygdala-basal ganglia circuit has already budgeted attention, decided a response, and staged motor output; System-2 is being invited to rationalise a decision the organism made a hundred milliseconds earlier. Understanding classical conditioning means understanding that the product decisions you make about sounds, animations, and timing are being installed at a depth the user cannot undo through self-talk.

Classical Conditioning vs. Other Theories

Classical conditioning overlaps with several other behavioral-design frameworks, and understanding the boundaries keeps a design team from conflating tools that look similar on the surface.

vs. Operant conditioning (Skinner). The cleanest distinction. Classical conditioning is about stimulus-stimulus associations learned passively; the organism does not need to do anything for the conditioning to form. Operant conditioning is about action-consequence associations; the organism must emit the behavior and experience its consequence. In most real products, a user’s interaction is operant (you tap a button because tapping produces an outcome), but the stimulus field that surrounds the interaction is classical (the sounds, animations, colors, and timings around the tap have been conditioned over repeated exposures). Separating them lets you debug: if the problem is that users do not engage in the first place, it is usually classical (the cue is not predictive, or has extinguished); if the problem is that they do not persist, it is usually operant (the reinforcement schedule is wrong).

vs. the Hook Model (Nir Eyal). The Hook Model — Trigger, Action, Variable Reward, Investment — is a design-facing synthesis that borrows from both classical and operant conditioning. The “Trigger” phase is pure classical; the internal trigger of boredom that leads you to reach for your phone was conditioned by thousands of prior boredom-phone pairings. The Variable Reward phase is operant and relies on Skinner’s variable-ratio reinforcement schedules. Treat the Hook Model as a layer above conditioning, not a replacement for it; when the Hook Model breaks down in a particular product, it is usually because one of the two underlying conditioning processes is failing.

vs. the Dual Process Theory. The Dual Process framework says the mind has a fast, automatic System-1 and a slow, deliberative System-2. Classical conditioning is one of the main ways System-1 gets built: over time, repeated pairings install automatic responses that System-2 can no longer easily veto. Kahneman’s writing sometimes treats System-1 as a warehouse of cognitive shortcuts; conditioning is the factory where the shortcuts are manufactured.

vs. Bandura’s Social Cognitive Theory. Bandura’s observational-learning work showed that organisms (including humans) do not need direct conditioning to acquire conditioned responses — they can acquire them vicariously from watching others. Seeing a peer flinch at a sound can install the flinch in you without you ever experiencing the pairing yourself. In product terms, this is why social-proof testimonials of “I couldn’t put it down” are such effective priming for conditioned engagement cues.

vs. the BJ Fogg Behavior Model. Fogg’s B = MAP (Behavior = Motivation + Ability + Prompt) treats the prompt as a discrete design variable. Classical conditioning tells you why certain prompts work and others do not. A prompt that has been conditioned — through repeated pairings with positive outcomes — is a different design object from a prompt that is arriving cold. Fogg is the operator model; classical conditioning is the installation model that explains why Fogg’s M and A aren’t constants across users.

Classical Conditioning in the Real World

Classical conditioning is load-bearing across a wider swath of applied domains than any other single construct in behavioral science. The four cases below are the ones I return to most often when teaching designers, because each one isolates a different feature of the framework that matters for product decisions.

Advertising, branding, and the economics of associative warmth

The advertising industry has been running classical-conditioning experiments on the global population for over a hundred years, with mostly unacknowledged methodology. When a soda advertisement pairs the product with images of friends laughing, sunshine, and music, it is performing an acquisition trial. When a luxury-watch advertisement pairs its product with calm cinematography and a baritone voice-over, it is performing a different acquisition trial aimed at a different prepared response. When we see the brand in isolation at the checkout counter, the conditioning pays out: the can of soda triggers a diluted version of the warm affective state; the watch triggers a diluted version of the calm, low-arousal state. This is evaluative conditioning, and the meta-analytic literature (Hofmann, De Houwer, Perugini, Baeyens, and Crombez 2010; De Houwer 2019) shows a small but persistent effect size across hundreds of studies. The effect does not require conscious awareness of the pairings, which is why most consumers believe advertising does not affect them and are, empirically, wrong.

The design implication is that your product is doing evaluative conditioning whether you planned for it or not. Every loading state, onboarding screen, reward animation, and email template is pairing your brand with the emotional context the user is experiencing while interacting with it. If your loading states are slow and your retry flows are frustrating, you are conditioning aversion to your own brand at a rate that no amount of marketing spend will undo. Designers who take classical conditioning seriously audit the in-product emotional context as rigorously as they audit the brand’s out-of-product advertising.

Phobias, anxiety, and the exposure-therapy fix

The clearest applied win of classical conditioning is in the psychotherapy of anxiety disorders. A phobia is, mechanically, a conditioned fear response in which a specific CS (a spider, a plane interior, an elevator) reliably elicits a strong sympathetic-nervous-system CR. Exposure therapy — developed by researchers including Joseph Wolpe and refined by decades of follow-up work including David Barlow’s protocols — uses the extinction logic Pavlov sketched. The patient is exposed to the CS in the absence of the US (the frightening outcome they learned to anticipate), the CR decays, and the extinction learning generalises across contexts.

The clinical take-up rate of exposure-based treatment is probably the best evidence for the practical power of the framework. The procedure is mechanically simple, the effect sizes are large (Norton and Price 2007 meta-analysis), and relapse, where it occurs, is explained by the renewal and reinstatement phenomena Bouton’s lab catalogued. The work of Michelle Craske and Mark Bouton (2008) on maximising extinction durability — massing versus spacing, context variation, retrieval cues — has given clinicians specific techniques to push the extinction learning across contexts instead of leaving it narrowly gated.

The design parallel is that when a product has inadvertently conditioned aversion (users who have been burned by a misleading button, a noisy notification, a broken flow), the fix is not more of the same cue in the hope that “it will get better.” The fix is exposure to the cue in a consistently positive context, spread across multiple settings, until the new learning generalises. That is a slow and expensive process, which is why the cheaper strategy is always to condition correctly the first time.

Food, hunger cues, and the conditioned appetite system

One of the more under-appreciated domains for classical conditioning in the modern consumer economy is the design of conditioned hunger. When a billboard shows a hamburger, your body is not responding to the image; it is responding to the conditioned anticipation of the eating experience the image predicts. Studies by Kent Berridge and colleagues on the distinction between “liking” (the hedonic reaction to a stimulus) and “wanting” (the motivational pull toward the stimulus) have shown that conditioned cues can drive wanting without any increase in liking — and that “wanting” is disproportionately sensitive to the dopaminergic system that encodes reward prediction error.

The upshot is that most of what we experience as hunger is not homeostatic. The stomach is not the sole driver. A substantial fraction of hunger is conditioned anticipation triggered by environmental cues — a smell, a time of day, a familiar location, the appearance of a food-delivery app icon. Products that want to increase snacking behavior have an easy design problem; products that want to help users eat more healthily have a hard one, because the conditioned appetite system is operating with a ten-year head start.

Digital notifications and the cost of a conditioned cue

The single most consequential application of classical conditioning in contemporary product design is in the notification system. Every app with a push-notification privilege is running an informal conditioning experiment on its users. If the notifications reliably predict something the user finds rewarding, the notification tone becomes a CS for anticipatory engagement — users tap it reflexively. If the notifications are noisy, generic, or deceptive, the tone becomes a CS for annoyance or dread — users mute the app, then delete it. The Rescorla-Wagner model predicts exactly this: the associative value of the notification cue tracks the expected value of the payload, updated on every trial.

A lot of growth-team dysfunction traces back to a failure to understand this. When a metrics lift is extracted by sending one extra notification per day, the short-run gain is real. But every uninformative notification is an extinction trial on the notification cue itself. Over weeks and months, the cue becomes less predictive, the CR weakens, and the overall open rate falls below where it started. The product has traded a durable conditioned asset for a one-quarter metric win. Designers who read Rescorla-Wagner correctly do not make that trade.

The Elephant in the Room

The elephant, for a designer learning classical conditioning, is the ethics question. Unlike most of the frameworks in this library, classical conditioning comes with a well-documented history of deliberate abuse. The Little Albert study (Watson and Rayner 1920) conditioned an infant to fear a white rat and generalised the fear to other white furry stimuli; the child was never deconditioned, the experiment violated contemporary research ethics, and the long-running replication attempts have themselves become controversial. Aversion therapy programmes in the mid-20th century paired homosexuality with nausea-inducing drugs in a practice that is now recognised as both scientifically wrong and morally indefensible. The Soviet and American propaganda establishments spent significant resources during the Cold War trying to operationalise Pavlovian principles for political coercion.

This history matters for designers because classical conditioning is unusually effective and unusually hard to undo. You can condition a response the user did not consent to and cannot easily argue with. You can generalise that response across contexts the user never anticipated. You can install it in a child whose metacognitive defences are not yet online. If we take the White-Hat / Black-Hat distinction seriously, classical conditioning is the clearest example of a tool that is exactly as dangerous as it is useful, and that requires a higher bar of design ethics than simpler tools like nudges or framing effects.

The working principle I use with design teams is a version of the classical-conditioning Hippocratic Oath. If we are going to condition a response in a user’s nervous system, the conditioned response must serve the user’s stated goals, and the pairing must be one the user would endorse if they could see it. That rule does not solve every hard case, but it rules out the easy abuses: conditioned scarcity-panic that drives purchases of things the user did not want; conditioned dread around notifications that the user cannot opt out of; conditioned anticipation for Variable Ratio rewards when the expected value of the rewards is zero. If your conditioning programme would fail a consent test if the user understood it, you are on the wrong side of the line.

How to Apply Classical Conditioning with the Octalysis Framework

The Octalysis Framework maps the eight Core Drives that underlie human motivation. Classical conditioning is not a ninth Core Drive; it is the installation mechanism that gives the existing Core Drives their autonomic traction. Every Core Drive can be conditioned, but two of them — CD6 and CD7 — rely on classical conditioning so centrally that designers who do not understand Pavlov cannot design them well.

Core Drive 7: Unpredictability & Curiosity is where classical conditioning does most of its motivational work. The Variable Ratio Reinforcement schedule Skinner made famous is formally an operant schedule, but the anticipatory pull a user feels before the schedule pays out is pure Pavlov. The loot box animation, the pull-to-refresh spinner, the Instagram feed’s slow-reveal of a new post — each of these is a conditioned stimulus the user’s nervous system has been trained to treat as a prediction of a reward. The CR — heightened attention, a small dopamine anticipation, reduced tolerance for delay — is what designers in this space are actually harvesting. If you want to make CD7 work, your conditioning of the CS must be tight (short CS-US interval), contingent (the reward must follow the CS reliably enough that the organism learns the prediction), and prepared (the CS should be the kind of stimulus the nervous system is biologically primed to condition to — sound, motion, sudden change).

Core Drive 6: Scarcity & Impatience is the second major classical-conditioning Core Drive. Countdown timers, limited-edition tags, and stock-low alerts work because they are CSs that have been conditioned, over many prior product interactions, to predict a loss-avoidance opportunity. The reason a countdown timer can move conversion so reliably is not because the user has done the math on the expected value of acting before expiry; it is because the countdown itself has become a conditioned aversive stimulus that the organism wants to resolve. Pavlov would have recognised the mechanism immediately. The White-Hat use of CD6 requires that the threat be real (otherwise the conditioning is non-contingent and will extinguish), and that the user’s resolution of the scarcity actually serves their interests (otherwise the conditioning is serving the product at the user’s expense).

For the other Core Drives, classical conditioning operates more quietly but still matters. CD1 (Epic Meaning & Calling) benefits from evaluative conditioning when the mission is consistently paired with emotionally charged content — ceremony, music, team imagery — so the mission cue carries affective weight even in cold-start moments. CD2 (Development & Accomplishment) benefits from reward sounds and progress animations that have been conditioned against genuine competence wins; the level-up chime becomes a CS for pride over repeated pairings. CD3 (Empowerment of Creativity & Feedback) benefits from conditioning the in-product feedback signals to the completion of creative work, so the feedback itself carries aesthetic satisfaction. CD4 (Ownership & Possession) benefits from conditioning the visual and auditory cues of one’s accumulated inventory, so the sight of the collection is rewarding in itself. CD5 (Social Influence & Relatedness) benefits from conditioning notification cues to real social arrival (a real message from a real friend) rather than empty social signals (a badge or a like from someone the user does not know). CD8 (Loss & Avoidance) is the one Core Drive where designers should be the most cautious about conditioning — pairing arbitrary cues with loss-avoidance is how you build anxious products that users secretly resent.

Practical Steps to Apply Classical Conditioning

If you want to put classical conditioning to work in a product you are shipping this quarter, the framework collapses into a short sequence of design decisions.

Step 1. Inventory the candidate conditioned stimuli in your product. List every sound, animation, color, haptic, icon, and timing pattern your product exposes the user to on a repeated basis. Each of these is a candidate CS, whether you planned for it or not. Most teams have never done this audit, and are surprised how many candidate CSs the product has accumulated.

Step 2. Identify the US these CSs are being paired with. For each candidate CS, write down the outcome that follows it in practice. If the onboarding chime is followed by a clean first-run experience, the chime is acquiring positive valence. If the notification ping is followed by a generic growth message the user did not want, the ping is acquiring aversive valence. The CS does not care what you wish it were paired with; it is pairing with whatever actually happens next.

Step 3. Check contingency, not just contiguity. Rescorla-Wagner says the CR depends on how informative the CS is about the US. Count the trials in which the CS is presented without the US following (non-contingent trials). If those are a significant fraction of your exposures, the CS is extinguishing. This is the most common way product teams accidentally destroy a conditioned asset they built on purpose.

Step 4. Stack preparedness where you can. Pick CSs the nervous system is prepared to condition to. A short, crisp, pitched sound conditions faster than a long tonal wash. A sudden motion conditions faster than a slow one. A color change with high saturation contrast conditions faster than a subtle tint shift. You can condition any CS given enough trials, but prepared CSs condition faster and hold their conditioning longer.

Step 5. Control the timing. The CS must precede the US by a short interval (hundreds of milliseconds to a few seconds for most response systems). A notification sound that arrives at the same time as the content it is announcing will condition weakly or not at all; a sound that arrives after the content will not condition to that content and may actually condition to whatever comes next.

Step 6. Design for extinction resistance. If you want the conditioning to survive occasional failures, train under variable conditions. Present the CS across contexts (platforms, times of day, user states), so the extinction overlay that would inhibit the CR in a single context has nowhere to hide.

Step 7. Monitor the CR, not just the conversion metric. Set up telemetry that measures user behavioral responses to the CSs in isolation (notification open rates, sound-triggered session starts, animation-to-tap latencies). If the CR is degrading, the conditioning is in trouble, and the downstream conversion metrics will follow with a lag of weeks.

Step 8. Audit against the Black-Hat line. For every CS you are conditioning on purpose, ask whether the user would endorse the conditioning if they understood it. Conditioned anticipation for a product that serves the user is a gift. Conditioned compulsion for a product that does not serve the user is a trap. Neither Pavlov nor the Octalysis Framework can do the ethics for you; that decision belongs to the designer.

Closing Thoughts

Classical conditioning is the oldest framework in this library, and it is still the most under-used by working designers. Part of the reason is that the framework has been repackaged so many times — as predictive coding, as reward prediction error, as evaluative conditioning, as the Trigger phase of the Hook Model — that most teams think they know it when they only know one of its projections. Part of the reason is that the basic version (“pair the bell with the food”) feels too simple to respect. And part of the reason is that taking the theory seriously requires an uncomfortable admission: your product is already running conditioning experiments on your users, and the only question is whether you are steering them.

If you leave this post with one practical commitment, let it be this. Walk through your product the way Pavlov walked through his lab, and count the candidate conditioned stimuli. Trace each of them to the outcomes they are actually paired with. Fix the ones that are conditioning aversion. Strengthen the ones that are conditioning anticipation for experiences the user is happy to have. That audit is the single cheapest, highest-payoff exercise a behavioral designer can run, and most teams have never done it. The fact that they have not is why Pavlov still has so much to teach us.

Where to Go Next

• Map the Core Drives: Read The Octalysis Framework: 8 Core Drives of Human Motivation to see where classical conditioning installs Core Drive 7 and Core Drive 6.
• Go deeper on the book: Actionable Gamification is the field reference for the design moves above.
• The twin pillar: Pair this with Skinner’s Operant Conditioning — classical installs the cue, operant installs the action.
• Browse the library: The Behavioral Framework Library houses every framework that builds on top of Pavlov.

Walk through your product the way Pavlov walked through his lab. Count the cues, trace each to the outcome it actually predicts, and decide — one CS at a time — what your nervous-system-level promises to your users are. The framework is timeless; the audit is overdue.

Frequently Asked Questions

What is classical conditioning in simple terms?

Classical conditioning is the process by which a previously neutral cue comes to trigger a response by being reliably paired with a biologically significant stimulus. Ivan Pavlov’s classic demonstration: a metronome that has been presented many times just before food is delivered eventually triggers salivation on its own. The cue has become a conditioned stimulus and the response a conditioned response.

How is classical conditioning different from operant conditioning?

Classical conditioning teaches organisms that stimuli predict other stimuli; it works on involuntary, respondent behavior. Operant conditioning teaches organisms that their voluntary actions produce consequences; it works on emitted behavior. In a product, the notification sound that makes you reach for your phone is classical; the tap-and-reward loop that follows is operant. Most well-designed products use both in sequence.

Is classical conditioning still considered valid in modern psychology?

Yes. The phenomenon replicates across species and response systems and has a neural mechanism — dopaminergic reward prediction error — that is well established in systems neuroscience. The original 1927 stimulus-response framing has been superseded by the Rescorla-Wagner contingency model (1972) and its computational descendants, but those are refinements rather than refutations.

What is the Rescorla-Wagner model and why does it matter?

Rescorla and Wagner’s 1972 equation states that learning on any trial is proportional to the difference between the outcome that occurred and the outcome the organism was predicting. This explains why pairings that carry no new information fail to condition (blocking), why highly salient cues soak up the learning (overshadowing), and why the acquisition curve has its characteristic shape. The same quantity — prediction error — is what midbrain dopamine neurons are now known to signal, tying the psychology to neuroscience.

What is the difference between extinction and forgetting?

Forgetting would mean the conditioned association decays over time. Extinction is different: the organism learns a new, context-specific inhibitory association that says “in this context the cue no longer predicts the outcome.” Because the original association is still stored, it can come back through spontaneous recovery, renewal (context change), or reinstatement (delivering the unconditioned stimulus alone). Extinction masks; it does not erase.

What is biological preparedness?

Biological preparedness, introduced by Martin Seligman in 1971, is the observation that certain stimulus-outcome pairings condition much faster than others because evolutionary history has biased the nervous system toward them. Garcia and Koelling (1966) showed rats learn taste-nausea associations in a single trial at long delays but cannot easily pair taste with noise. For designers, it means some cues (brief, pitched sounds; sudden motion; color changes) condition far more efficiently than others.

How does classical conditioning apply to marketing and branding?

Evaluative conditioning — the repeated pairing of a brand with emotionally charged contexts — gives brands affective weight that the consumer cannot always articulate. Meta-analyses (Hofmann et al. 2010) show the effect is small but persistent and does not require conscious awareness of the pairings. For product designers, the in-product emotional context is running evaluative conditioning on the brand every day, for better or worse.

What is the ethical concern with using classical conditioning in design?

Conditioning installs responses at a level below deliberative consent and is unusually resistant to reversal. Using it to serve outcomes the user would not endorse — conditioned scarcity panic, conditioned compulsion to check a notification with no real payload — is the defining Black-Hat pattern in Octalysis vocabulary. White-Hat use requires that the conditioned response serve the user’s own stated goals and that the pairing be one the user would endorse if visible.

How does classical conditioning fit into the Octalysis Framework?

Classical conditioning is the installation mechanism for several Core Drives, most importantly Core Drive 7 (Unpredictability & Curiosity) and Core Drive 6 (Scarcity & Impatience). The anticipatory pull a user feels before a variable reward, and the resolution urgency a countdown produces, are conditioned responses trained across thousands of prior product interactions. The Octalysis Framework maps which Core Drives are active; classical conditioning explains why those drives can be triggered by cues that do not look motivationally loaded on their face.

Can I extinguish a conditioned response a user has learned from my product?

Partially. Extinction will reduce the immediate expression of the conditioned response, but the underlying association remains stored in the nervous system and can re-activate through spontaneous recovery, context change, or reinstatement (Bouton 2004). The durable fix is not to reverse the conditioning but to replace it — introduce a new cue paired with the desired positive outcome and condition against it from scratch in a deliberately varied context to support generalisation.

References

1. Pavlov, I. P. (1927). Conditioned Reflexes: An Investigation of the Physiological Activity of the Cerebral Cortex. Oxford University Press.
2. Watson, J. B., & Rayner, R. (1920). Conditioned emotional reactions. Journal of Experimental Psychology, 3(1), 1–14. https://doi.org/10.1037/h0069608
3. Rescorla, R. A. (1968). Probability of shock in the presence and absence of CS in fear conditioning. Journal of Comparative and Physiological Psychology, 66(1), 1–5. https://doi.org/10.1037/h0025984
4. Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and non-reinforcement. In A. H. Black & W. F. Prokasy (Eds.), Classical Conditioning II: Current Research and Theory (pp. 64–99). Appleton-Century-Crofts.
5. Garcia, J., & Koelling, R. A. (1966). Relation of cue to consequence in avoidance learning. Psychonomic Science, 4(1), 123–124.
6. Seligman, M. E. P. (1971). Phobias and preparedness. Behavior Therapy, 2(3), 307–320. https://doi.org/10.1016/S0005-7894(71)80064-3
7. Schultz, W., Dayan, P., & Montague, P. R. (1997). A neural substrate of prediction and reward. Science, 275(5306), 1593–1599. https://doi.org/10.1126/science.275.5306.1593
8. Bouton, M. E. (2004). Context and behavioral processes in extinction. Learning & Memory, 11(5), 485–494. https://doi.org/10.1101/lm.78804
9. Craske, M. G., Kircanski, K., Zelikowsky, M., Mystkowski, J., Chowdhury, N., & Baker, A. (2008). Optimizing inhibitory learning during exposure therapy. Behaviour Research and Therapy, 46(1), 5–27. https://doi.org/10.1016/j.brat.2007.10.003
10. Hofmann, W., De Houwer, J., Perugini, M., Baeyens, F., & Crombez, G. (2010). Evaluative conditioning in humans: A meta-analysis. Psychological Bulletin, 136(3), 390–421. https://doi.org/10.1037/a0018916
11. Berridge, K. C., & Robinson, T. E. (2016). Liking, wanting, and the incentive-sensitization theory of addiction. American Psychologist, 71(8), 670–679. https://doi.org/10.1037/amp0000059
12. LeDoux, J. E. (2000). Emotion circuits in the brain. Annual Review of Neuroscience, 23, 155–184. https://doi.org/10.1146/annurev.neuro.23.1.155
13. Kandel, E. R. (2001). The molecular biology of memory storage: A dialogue between genes and synapses. Science, 294(5544), 1030–1038. https://doi.org/10.1126/science.1067020
14. Norton, P. J., & Price, E. C. (2007). A meta-analytic review of adult cognitive-behavioral treatment outcome across the anxiety disorders. Journal of Nervous and Mental Disease, 195(6), 521–531. https://doi.org/10.1097/01.nmd.0000253843.70149.9a
15. De Houwer, J. (2019). Moving beyond system 1 and system 2: Conditioning, implicit attitudes, and habitual responses as explananda in need of explanation. Social Cognition, 37(2), 103–115. https://doi.org/10.1521/soco.2019.37.2.103

Related Reading

• Skinner’s Operant Conditioning: The Twin Pillar to Pavlov
• The Hook Model: Trigger, Action, Variable Reward, Investment
• BJ Fogg Behavior Model: B = Motivation + Ability + Prompt
• Dual Process Theory: System 1 and System 2 Thinking
• Social Cognitive Theory: Learning Without Direct Conditioning
• The Octalysis Framework: 8 Core Drives of Human Motivation
• The Behavioral Framework Library (Full Hub)

WOULD YOU LIKE YU-KAI CHOU TO WORK WITH YOUR ORGANIZATION?