Why Old Habits Come Back (and What to Do Instead)

6 min read · Updated 2026-07-18

Old habits come back because they were never deleted. In Graybiel's rats, extinction training reversed the neural pattern marking a learned habit, but the pattern reappeared as soon as retraining started. The practical consequence: replacing a habit works better than trying to remove one, because removal isn't really available.

People treat a returned habit as a character failure. Six months clean, one bad week, back where you started, and the explanation is usually some version of not wanting it badly enough.

The neuroscience says something less personal. The old pattern is still physically there. It was suppressed, not removed.

What happens in the brain as a habit forms

Ann Graybiel's 2008 review in the Annual Review of Neuroscience describes a process she calls chunking, where a sequence of separate actions gets bundled into a single releasable unit. That's what lets a habit run while your attention is somewhere else.

The recording data is genuinely counterintuitive. Researchers running rats through a T-maze expected striatal activity to be low during early training and to climb as the behavior became habitual. It didn't. Activity was already strong early on, and with overtraining it didn't increase - it redistributed, concentrating at the beginning and the end of the run while quieting in the middle. Recordings in monkeys running trained sequences showed the same bracketing, as though marking the boundaries of the action sequences.

So a well-formed habit isn't a louder signal. It's a signal that has collapsed to its edges. Graybiel suggests the quiet middle is why a familiar routine needs so little attention while you're performing it.

The result that explains relapse

Here's the finding worth the whole article. When the rats were given extinction training, the acquired start-and-end pattern didn't hold. It gradually reversed, which looks like the habit being unlearned.

Then reacquisition training began, and the pattern reappeared. Graybiel's phrasing is blunt: they were not lost or forgotten.

The lesion work points the same way. Inactivating the infralimbic prefrontal cortex after a habit has been trained reinstates goal-directed behavior, which means the deliberate system was being actively suppressed rather than overwritten. Wood and Runger describe the behavioral version of this: old memory traces aren't necessarily replaced when new habits are learned, and the old response is still available to its original cue.

Their conclusion is worth stating plainly, because it contradicts how most people narrate their own relapses. Difficulty changing an established behavior doesn't indicate a continuing desire to perform it, and it isn't a failure of willpower.

Which is why substitution beats deletion

If the old pattern stays encoded, aiming at erasure is aiming at something that isn't on offer. Duhigg's formulation of this is the most usable one out there: keep the old cue, keep the old reward, and change only the routine in between. The structure stays intact and you swap the middle.

The reason this works better than white-knuckling is that you're not fighting the cue. The cue is going to fire. The question is only what it releases.

The hard part is usually identifying what the reward actually is, because it's often not the obvious thing. Graybiel's framing is useful here: a habit is performed not in relation to a current goal but in relation to a previous one, and the behavior that once satisfied it. You're running an old solution to a problem you may not have anymore.

What a log gives you that memory does not

• The cue, identified rather than assumed. Most people guess wrong about their own triggers, and a timestamped record of when the behavior happened beats introspection.
• Whether the swap is holding at the original cue. A new routine that only runs on easy days hasn't replaced anything yet.
• How long the old pattern stayed quiet before it came back, which tells you far more about the risk than a streak number does.
• Enough history that a single bad week reads as a single bad week. Without a record, one relapse tends to overwrite your memory of the months that preceded it.

One clarification worth making

Habit learning and skill learning are different systems, and Graybiel is direct about it: a regularly practiced habit may actually not be very skilled. Learning to ride a bike isn't the same kind of thing as having the habit of biking home every evening. Repetition makes a behavior automatic. It doesn't necessarily make it good.

That matters for tracking, because a long streak proves consistency and nothing else. It's not evidence that the thing you're consistent at is working.

HabitSync keeps your full history rather than resetting it, which is the point. When a habit lapses and restarts, the record shows the whole arc - how long it held, what the gap looked like, and whether the pattern around it changed. That's a more honest picture than a counter that returns to zero and implies you're starting over.