The extinction timeline

The lie of the clean break

You imagined it as a switch. One day you do the thing, the next day you stop, and by the end of the week the urge has faded like a sunburn. You have heard the language that supports this fantasy — "quit cold turkey," "just stop," "draw a line in the sand." The metaphors all imply a binary transition: on, then off. And so when you decided to stop the behavior — the compulsive checking, the stress eating, the reflexive avoidance — you expected a brief period of discomfort followed by clean silence. A week, maybe two, of white-knuckling, and then freedom.

That is not what happened. What happened was messy, non-linear, and longer than you budgeted for. The urges faded and then returned. The behavior seemed gone and then appeared again at full strength on an ordinary Tuesday for no discernible reason. Your motivation to maintain the extinction eroded not because the process was failing but because you had no accurate model for how long the process takes and what the curve actually looks like. You were navigating with a map that showed a straight highway, and the real terrain was switchbacks.

This lesson gives you the real map. Behavioral extinction follows a predictable but non-linear curve that unfolds over weeks to months, not days. Understanding that curve — its shape, its features, and the variables that stretch or compress it — is the difference between completing an extinction attempt and abandoning one at the precise moment it was beginning to work.

The shape of the extinction curve

When you remove reinforcement from an established behavior, the subsequent decline in that behavior does not follow a straight line. Skinner first documented this in his operant conditioning experiments in the 1930s, and every subsequent extinction study has confirmed the basic shape. The curve has four distinct features, and knowing them transforms how you interpret your own experience during an extinction attempt.

The first feature is the extinction burst. You encountered this in Extinction bursts — the initial surge in frequency, intensity, and variability that occurs when reinforcement is first withdrawn. The behavior gets worse before it gets better. On the timeline, the burst typically occupies the first one to five days, though it can be shorter for simple behaviors and longer for deeply ingrained ones. The burst is the curve's opening spike, and it is the most dangerous phase because it feels like escalation rather than progress.

The second feature is the gradual decline. After the burst resolves, the behavior's frequency and intensity begin to fall. This decline is not uniform. It follows a roughly exponential decay pattern — steep at first, then increasingly shallow. In the first two weeks after the burst, you may notice dramatic reductions. By week three, the rate of improvement slows considerably. The behavior is still declining, but the daily change is small enough to be invisible without measurement. This is where most people make their critical error: they compare today to yesterday and see no improvement, when the meaningful comparison is this week to last week or this month to last month.

The third feature is spontaneous recovery. This is the phenomenon that destroys the most extinction attempts, and it is the one least understood by people outside behavioral science. Pavlov first observed it in the early 1900s with his conditioned dogs. After extinction appeared complete — the dogs had stopped salivating to the conditioned stimulus — he brought the dogs back to the laboratory after a rest period and presented the stimulus again. The conditioned response returned, at reduced strength, without any re-training. The behavior had not been erased. It had been overlaid by new learning, and the passage of time partially uncovered the original association.

Mark Bouton's research, spanning decades from the 1990s through the 2000s, established the theoretical framework for understanding spontaneous recovery. Bouton demonstrated that extinction does not destroy the original learning. It creates a new, competing association — an inhibitory memory that suppresses the old behavior. But this inhibitory memory is more context-dependent and more fragile than the original excitatory memory. When context changes — including the simple context change of time passing — the inhibitory learning loses some of its grip, and the original behavior resurfaces. This is why you can go three weeks without the old urge and then experience it at near-original intensity on a random afternoon. The original learning is still there, underneath. Spontaneous recovery is it showing through the cracks in the newer, inhibitory learning.

The fourth feature is the asymptote. Over sufficient time, the spontaneous recovery episodes become less frequent, less intense, and shorter in duration. The extinction curve approaches but never perfectly reaches zero. In practical terms, this means the behavior eventually becomes so rare and so weak that it no longer constitutes a meaningful pattern — but the original learning may never fully disappear. Robert Rescorla's 2004 research on extinction as decremental learning confirmed this view. Extinction does not return the organism to a naive state. It creates a new behavioral equilibrium that sits very near zero but retains, somewhere deep in the associative structure, a trace of the original contingency.

Put these four features together and you get the actual extinction curve: a sharp initial spike (the burst), followed by an exponential decline punctuated by periodic spontaneous recovery episodes of diminishing amplitude, gradually approaching an asymptotic floor. It looks nothing like a straight line from "behavior" to "no behavior." It looks like a heartbeat on a monitor that is gradually flatlining — still spiking occasionally, but each spike lower than the last, and the baseline between spikes dropping steadily toward silence.

How long the timeline actually takes

You want a number. Everyone wants a number. The honest answer is that extinction timelines vary enormously depending on several factors, but research gives you a reasonable range.

Phillippa Lally and her colleagues at University College London found in their 2010 study that habit formation — building a new automatic behavior — takes a median of 66 days, with a range spanning 18 to 254 days. Extinction is the mirror process: dismantling an automatic behavior. The research suggests that extinction timelines are comparable to or longer than formation timelines for equivalent behaviors, because extinction must contend with an existing, well-practiced neural pathway rather than building one from scratch. A simple behavior maintained by a straightforward reinforcement contingency might extinguish in three to six weeks. A complex behavior with a long reinforcement history, maintained on a variable schedule, embedded in multiple contexts, and supported by social reinforcement might take three to six months or longer.

These numbers are not precise specifications. They are the range within which your particular extinction curve is likely to fall. The point is not to know the exact day when the behavior will stop. The point is to calibrate your expectations so that week four of an extinction attempt does not feel like failure. It is not failure. It is, by the research, the early middle.

What determines where you fall on the timeline

Five factors dominate how long your particular extinction will take, and understanding them lets you make a rough prediction before you begin.

The first factor is reinforcement history — how many times the behavior has been reinforced and over what duration. A behavior you have been performing and getting rewarded for over ten years has a thicker, more deeply grooved neural pathway than one you picked up six months ago. Each reinforced repetition strengthened the association. More reinforced repetitions mean more extinction trials needed to overlay that association with inhibitory learning. This is straightforward: older habits take longer to extinguish than newer ones, all else being equal.

The second factor is reinforcement schedule, and this is where the timeline can stretch dramatically. Skinner demonstrated that the schedule on which a behavior is reinforced profoundly affects its resistance to extinction. Behaviors reinforced on a continuous schedule — every occurrence produces a reward — are actually the easiest to extinguish. The organism quickly detects that the contingency has changed because the contrast between "always rewarded" and "never rewarded" is stark. Behaviors reinforced on a fixed-ratio or fixed-interval schedule are moderately resistant. But behaviors reinforced on a variable-ratio schedule are extraordinarily resistant to extinction, and this is the schedule that governs most of the behaviors people struggle most to eliminate.

Variable-ratio reinforcement means the behavior is rewarded after an unpredictable number of occurrences. This is the schedule that operates slot machines, social media feeds, email checking, and most forms of compulsive digital behavior. You pull the lever (or refresh the feed) and sometimes you get a reward and sometimes you do not, and you cannot predict which pull will pay off. This schedule creates behaviors that persist through remarkably long stretches of non-reinforcement because the organism cannot distinguish "the contingency has permanently changed" from "I am in a normal dry spell and the next reward is coming." Every unreinforced trial is ambiguous. Skinner's pigeons on variable-ratio schedules would peck thousands of times after reinforcement was withdrawn, far more than pigeons on continuous or fixed schedules. The slot machine effect is not a metaphor. It is the literal mechanism by which variable-ratio reinforcement produces persistence.

If the behavior you are trying to extinguish was maintained by variable reinforcement — and most persistent unwanted behaviors are, precisely because this schedule makes them persistent — your timeline will be at the longer end of the range. Plan accordingly.

The third factor is the complexity of the behavior. A single, discrete behavior (checking one specific app) extinguishes faster than a behavioral chain (waking up, reaching for the phone, opening the app, scrolling, reading, reacting). Chains have multiple links, each of which is its own cue-behavior-reward loop. Extinction must operate on each link. Complex behaviors also tend to have more contextual triggers, meaning the extinction learning must generalize across more situations before the behavior feels truly resolved.

The fourth factor is environmental consistency. Extinction learning is context-dependent — a point Bouton's research emphasizes repeatedly. If you extinguish a behavior in one context (at home, alone, during the evening), the extinction may not transfer to another context (at work, with colleagues, during a stressful meeting). Every new context where the old behavior is triggered but not reinforced requires its own extinction learning. The more contexts in which the behavior is established, the longer the overall extinction timeline, because you are effectively running parallel extinction processes. This is why behaviors that are triggered "everywhere" — like thought patterns or emotional reactions — take longer to extinguish than behaviors anchored to a specific situation.

The fifth factor is whether the behavior is being simultaneously reinforced through channels you have not identified. This connects directly to Social reinforcement of unwanted behaviors on social reinforcement. If you have removed the primary reinforcer but a secondary reinforcer is still operating — a friend who encourages the old behavior, an environmental cue that provides comfort, an internal thought pattern that rewards the behavior through anxiety reduction — then full extinction cannot proceed until all maintaining reinforcers are addressed. An extinction attempt that is unknowingly competing with ongoing reinforcement will plateau rather than continue declining, and you will interpret the plateau as a ceiling rather than a signal to search for hidden reinforcers.

Setting realistic expectations

The practical application of understanding the timeline is that you can set expectations before you begin, and those expectations will prevent you from misinterpreting normal progress as failure.

Before starting any extinction attempt, write down your answers to these questions. How long have I been performing this behavior? What reinforcement schedule maintains it — continuous, fixed, or variable? How many contexts trigger it? What is the behavior's complexity — single action or multi-step chain? Have I identified all reinforcers, including social ones?

Based on your answers, assign a rough timeline. A simple, recently acquired behavior with continuous reinforcement in a single context: three to six weeks. A moderate behavior with a year or more of history, some variable reinforcement, in two or three contexts: six to twelve weeks. A complex, long-established behavior on a variable-ratio schedule across many contexts with social reinforcement: three to six months, possibly longer.

Write this timeline down and put it where you will see it. On day twenty, when the urge resurfaces after a week of quiet, you will look at the timeline and see that you are a third of the way through your conservative estimate. You are not failing. You are in the second act. The spontaneous recovery episode you are experiencing right now is predicted by the model. It will be shorter and weaker than the last one. The curve is working. You just need to stay on it long enough for the asymptote to arrive.

Measuring progress on a non-linear curve

Because the extinction curve is non-linear, standard binary measurement — "Did I do the behavior today? Yes or no?" — is inadequate and potentially harmful. Binary measurement erases all the information that tells you extinction is working. A day where you experienced the urge three times at intensity 4 and resisted each time looks identical, in binary measurement, to a day where you experienced the urge twelve times at intensity 9 and resisted each time. By any reasonable assessment, these are radically different days. But "Did I do the behavior? No." makes them the same.

You need dimensional measurement. Track three variables daily: frequency (how many times the urge or behavior occurred), peak intensity (the strongest instance, rated on a simple 1-to-10 scale), and duration (how long the longest episode lasted). These three numbers, plotted over weeks, reveal the extinction curve in your own data. You will see the burst in the early data points, the decline in the subsequent weeks, the spontaneous recovery blips, and the gradual approach to your asymptote.

Dimensional measurement also protects you from the most common cognitive distortion during extinction: "I am back to square one." When a spontaneous recovery episode hits in week four, it feels subjectively like the behavior has returned at full strength. But if you check your data, you will almost certainly find that the frequency is lower, the intensity is lower, or the duration is shorter than it was during the burst in week one. The recovery only feels like square one because you are comparing it to the quiet days immediately preceding it rather than to the actual beginning. Your data tells the real story. Trust it over your feelings.

This mirrors the insight from Habit formation takes weeks not days on habit formation timelines. In that lesson, you learned to measure automaticity on a 1-to-10 scale to track the formation curve for new habits. Here you are doing the inverse: measuring urge strength on a 1-to-10 scale to track the extinction curve for old behaviors. Formation and extinction are symmetrical processes. Both are non-linear. Both take longer than intuition predicts. Both require dimensional measurement rather than binary assessment. And both are sabotaged by unrealistic timeline expectations that transform normal progress into apparent failure.

The Third Brain

An AI assistant with access to your extinction tracking data can perform a function that your own cognition systematically fails at during extinction: objective trend analysis. You are inside the process. Your perception of your own progress is distorted by recency bias (the spontaneous recovery episode yesterday looms larger than the three quiet weeks before it), negativity bias (you remember the intense urges more vividly than the mild ones), and the fundamental human tendency to evaluate progress by comparing today to yesterday rather than this month to last month.

Feed your daily frequency, intensity, and duration numbers to your AI assistant. Ask it to plot the extinction curve from your data. Ask it to fit a trend line and identify whether the overall trajectory is declining, plateauing, or increasing. Ask it to compare your current spontaneous recovery episodes to the initial burst and quantify the difference. These are straightforward calculations, but they are calculations you will not perform honestly on yourself during a difficult week, because during a difficult week you do not want to calculate — you want to either feel better or give up.

The AI can also help you identify plateau patterns that signal hidden reinforcers. If your extinction curve stops declining and holds steady at a level above your asymptotic target, this may indicate that a reinforcer you have not identified is maintaining the behavior at its current level. An AI reviewing your logs might notice that spikes correlate with particular contexts, times of day, emotional states, or social interactions — correlations you miss because you are too close to the data. It can ask you targeted questions: "Your urge intensity spikes on Tuesday and Thursday evenings. What is different about those evenings?" This kind of pattern detection turns your tracking data from a passive record into an active diagnostic tool.

Perhaps most importantly, the AI can normalize your experience. When you report that the behavior resurfaced after three weeks of absence, the AI can respond with context: "Spontaneous recovery at the three-to-four-week mark is one of the most well-documented features of extinction. Your recovery episode peaked at intensity 6, compared to your burst peak of intensity 9. This represents a 33 percent reduction in peak amplitude. The curve is progressing normally." This is not reassurance. It is data interpretation. And it is data interpretation that your own mind, in the grip of a recovery episode, cannot perform reliably.

The timeline is the strategy

Everything in this lesson reduces to a single operational insight: if you do not know how long extinction takes, you will quit before it is finished. Not because you are weak. Not because the behavior is too strong. Because you will interpret normal, predicted, on-schedule progress as evidence of failure, and you will act on that interpretation by abandoning the attempt or — worse — reinforcing the behavior at burst or recovery intensity, making the next attempt harder.

The timeline is not a passive fact about extinction. It is the active strategy. Knowing that your particular behavior, given its reinforcement history and schedule and complexity and environmental spread, will take approximately eight to twelve weeks to reach asymptote transforms every day within that window from a test of willpower into a data point on a predictable curve. You are not enduring. You are waiting — with measurement, with understanding, and with the knowledge that the curve has an end.

But the curve's end does not mean the behavior can never return. Spontaneous recovery can occur even after long periods of apparent extinction, and context changes can temporarily revive responses you thought were gone. This is not a design flaw in extinction. It is a feature of how the brain stores and retrieves associative learning. The next lesson takes this on directly: relapse is not failure. It is a normal, predicted phase of the extinction timeline — one that, like the burst and the recovery episodes, has its own structure, its own logic, and its own strategies for navigation.