TL;DR. The habit loop splits an automatic behaviour into parts so you can edit one of them. Duhigg uses three (cue, routine, reward), Clear uses four (cue, craving, response, reward). The extra step is not a correction, it converts a description into a design checklist. To use either one: find your cue from five categories, run an experiment to find the reward you are actually chasing, then keep the cue and the reward and change only the middle. Both models are popular-science synthesis on top of real research, not settled science.
There is a version of this article on about forty book-summary sites, and every one of them draws the same circle with three arrows and stops there. The circle is not the useful part. What to do once you have drawn it is.
Two versions of the same diagram
Charles Duhigg, The Power of Habit (2012). Cue, routine, reward. Something triggers you, you run the behaviour, you get a payoff, and the payoff teaches your brain to run the same thing next time the trigger shows up. Duhigg does talk about craving, but it sits outside the three-part diagram as the thing that makes the loop spin.
James Clear, Atomic Habits (2018). Cue, craving, response, reward. Same shape, one extra stage, and craving is promoted into the loop itself.
That extra step is not a correction, and I would be suspicious of anyone selling it as one. It is a change of purpose. By making craving a stage, Clear gets four places to intervene, and the four laws fall straight out of them: make it obvious, make it attractive, make it easy, make it satisfying. Duhigg's version is a diagnostic tool for a habit you already have and cannot explain. Clear's is a build spec for one you want to install.
Use whichever matches what you are doing. Do not blend them into a five-step hybrid, which is what most summaries end up doing by accident.
One honest caveat before the practical part. The research these books draw on is real. Studies of the basal ganglia, including Ann Graybiel's work at MIT, show that as a sequence becomes routine the brain increasingly treats it as one chunk, with neural activity concentrating at the beginning and end of the sequence rather than throughout. That is genuine, and it is mostly from tightly controlled animal experiments. The clean loop diagram is a popular-science synthesis layered on top. It is a good map. It is not a mechanism anyone has photographed, and any article telling you exactly which brain region "controls" your evening scrolling is overselling.
Step 1: find the actual cue
Almost everyone gets this wrong on the first guess, because the obvious cue is usually a prop rather than a trigger.
Duhigg's five categories cover nearly everything:
- Location. Where are you standing or sitting?
- Time. What time is it?
- Emotional state. What are you feeling, in one plain word?
- Other people. Who is around, or is it the absence of people?
- Immediately preceding action. What did you do in the thirty seconds before?
The method: the next five times you catch yourself mid-habit, write down all five. Not a description of your feelings, just five short answers. After five rounds the column that repeats is your cue.
When I did this for evening phone use, I expected the answer to be "phone within reach". It was not. Four of my five entries had the same emotional state and the same preceding action: the flat had just gone quiet after my daughter was down, and I had just closed the laptop. The phone was incidental. Leaving it in the hallway moved the habit to the iPad within a week, because the cue was untouched.
Step 2: find the reward you are actually getting
The stated reward is rarely the real one. Nobody scrolls for information, and almost nobody snacks at 22:00 because they are hungry.
Guessing does not work here, so run Duhigg's experiment instead. When the cue fires, deliberately do something else, then wait about fifteen minutes and ask one question: is the urge gone?
- Urge gone: that alternative delivered the same payoff. You have found the reward category.
- Urge still there: wrong candidate. Try another one tomorrow.
Write two or three words immediately after the alternative behaviour, before your memory tidies it up. Cycle through a handful of candidates over a week: rest, stimulation, distraction, connection, a sense of control, an ending to the day.
For my evening scroll, the answer was "nothing is being asked of me". Not information, not novelty. The absence of demand. That single sentence explained why every replacement I had previously tried had failed.
Step 3: keep the cue, keep the reward, change the middle
This is Duhigg's golden rule and it is the only part of the whole framework I would call load-bearing.
You are not removing a habit. You are rerouting one. The cue will keep firing regardless of your opinion about it, and the need behind the reward does not disappear because you disapprove of how it is currently met. Take out the routine and put nothing in its place and you have built a gap that the old behaviour fills the first time you are tired.
Worked example 1: evening scrolling
- Cue: flat goes quiet, laptop just closed, tired, alone.
- Routine: phone, an hour, no memory of any of it.
- Reward: nothing is required of me.
The first replacement I tried was reading non-fiction, which failed inside four days. Obvious in hindsight: a book about anything useful is a demand. What worked was fiction and a specific chair, plus the phone charging in another room so the old routine needed a decision instead of a reach. Same cue, same reward, different middle. It is not perfect, and I still lose an evening most weeks. Rerouting a habit reduces frequency, it does not delete the behaviour.
Worked example 2: installing a habit that does not exist yet
Building forward is a different job. There is no cue to discover, so you have to assign one, and the reliable way to assign a cue is to borrow a behaviour you already perform every day. That is exactly what habit stacking is for: after I close the laptop, I will write one paragraph.
Then design the other three stages before you start:
- Craving: what makes the first thirty seconds appealing rather than virtuous? Usually pairing it with something you already like.
- Response: shrink it until a bad day cannot beat it. One paragraph, not thirty minutes.
- Reward: it has to arrive immediately, because a payoff in six months cannot teach anything today. Checking the thing off counts, which is unglamorous but genuinely how tracking earns its place.
That last point is where most designed habits quietly fail, and the numbers are unkind: across more than 7,800 tracked habits on Loggd, about 45% were never checked off once and another 23% exactly once, which I broke down in habits people quit fastest. A loop with no immediate reward never completes even once.
Where the model stops being useful
- Not everything is a habit. Some behaviour is a decision, or a symptom of something the loop cannot reach. Rerouting a routine will not fix burnout, grief or an untreated condition, and framing those as bad habits is worse than useless.
- Understanding a loop does not dissolve it. You can describe your own cue and reward perfectly and still do the thing that evening. Awareness is a prerequisite, not a treatment.
- Cues are usually plural. One habit often has three triggers, and killing one leaves the other two working fine.
- The loop has no cost side. It explains why a behaviour repeats, not whether it is worth its price. That is a different exercise: see how a dopamine menu handles the trade between cheap payoffs and expensive ones.
Mapping it without a notebook
The habit loop analyzer runs this as a five-step wizard in one of two modes, build or break. It asks you to name the habit, pick a cue type from those five categories and describe it, choose the craving from seven options, describe the response, then the reward. Build mode also asks for the time, the location, a two-minute version and a temptation bundle; break mode asks what environment change and what friction you will add.
At the end it draws the loop, writes your implementation intention as a sentence, and lays out the four laws (or the inverted four laws in break mode) as strategy cards. You can save analyses, copy the whole thing, or download it as a plain text file. Everything stays in your browser and nothing is sent anywhere.
Draw the circle if you like. Then go and find out what your actual cue is, because I promise it is not the one you would name right now.
Last updated: November 2026.
Written by Eusebiu, the solo founder building Loggd in public. I build the app in the evenings around contract work and a small daughter, and the evening scrolling example is mine, unresolved. I share what the aggregate data keeps showing on Threads.
Close the loop with a reward that shows up today. Loggd turns each check into a visible mark on a contribution grid, so a missed day stays one lighter square instead of resetting you to zero. Start free.
