Implementation intentions: how to write if-then plans that actually work

What an implementation intention is

The psychologist Peter Gollwitzer separated two kinds of intention. A goal intention says what you want: “I want to read more.” An implementation intention adds when, where and how you will start: “If I sit down on the sofa after dinner, then I open my book.” In Gollwitzer’s 1999 paper the template is “When situation x arises, I will perform response y.”

The idea is to make the decision once, in advance. When the situation shows up, you don’t have to weigh options or even remember the goal, because the situation itself prompts the action. Gollwitzer described this as handing control of the behaviour to a cue you have picked beforehand. Lab studies summarised in the 2025 meta-analysis by Sheeran, Listrom and Gollwitzer point to two things going on: people notice the cue more readily, and the cue-response link gets strong enough that starting needs less effort and less conscious intent.

The early studies were striking. In one, students were asked before Christmas to write a short report within 48 hours of Christmas Eve. Of those who had picked a specific time and place, about three quarters sent it on time; of those who hadn’t, about one third did. In an exercise study by Milne, Orbell and Sheeran, described in the same 1999 review, a motivational leaflet raised the share of students who exercised from 29% to 39%. Adding an if-then plan to the leaflet took it to 91%.

How well do implementation intentions work?

The number most articles quote comes from a 2006 meta-analysis by Gollwitzer and Sheeran: 94 tests, 8,461 participants, and an average effect of d = 0.65, which by the usual conventions sits between medium and large. 79 of those 94 tests were run with university students, and many used lab tasks.

The 2025 update is the better guide. It pooled 642 tests published up to 2019 and found an average of d = 0.36. It also found substantial publication bias: small studies with positive results were overrepresented. The most conservative correction the authors ran, a Robust Bayesian Meta-Analysis, put the effect at d = 0.15 while still finding strong evidence (a Bayes factor of about 319) that the effect is real. So the effect is reliable, smaller than the early hype, and varies a lot from one situation to the next.

For everyday behaviour the field studies matter most, and there the average was d = 0.27, against 0.49 in the lab. By behaviour: eating more fruit and vegetables 0.32, eating less unhealthy food 0.27, alcohol 0.21, quitting smoking 0.20, physical activity 0.14, taking medication 0.40. The effects held when outcomes were measured objectively instead of self-reported (0.43 vs 0.28), and in the 12 tests that followed people for six months or more.

A clean real-world example comes from a field experiment by Milkman and colleagues. 3,272 employees of a US utility got a mailer about free flu shots at work. In the control group 33.1% got vaccinated. A mailer that also invited people to write down the date and time they would go raised that by 4.2 percentage points; asking for the date alone added 1.5 points, which was not statistically significant. The prompt cost nothing extra to print. The study has limits (one employer, mostly men, the 2009 H1N1 season), and across five vaccination tests the 2025 meta-analysis found no reliable effect, which its authors put down to low motivation to get vaccinated in those samples.

How to write an implementation intention

The 2025 meta-analysis coded what was inside each plan, and the results translate into fairly concrete rules.

1. Start from a goal you actually hold. If-then plans work on top of motivation and don’t create it. When studies compared people with strong and weak intentions, the plans helped the motivated ones (Sheeran, Webb & Gollwitzer, 2005). In exercise studies the effect was d = 0.31 with strong intentions and roughly zero with weak ones.
2. Use the if-then wording. Plans phrased as a contingency (“if X, then Y”) averaged d = 0.43; plain schedules listing when, where and what averaged 0.29.
3. Make the cue a time and a place, or a moment in a routine. A time alone gave d = 0.25, a time plus a place 0.46. Cues tied to a point in a task, such as starting or finishing something, did at least as well (0.49 to 0.64).
4. Name the first action and stop there. Plans that also spelled out how, or for how long, averaged 0.24, about half the time-and-place figure. “Then I put on my running shoes” beats “then I run 5 km at a steady pace along the canal.”
5. Say it back to yourself once. Rehearsing the plan at least once raised the effect (0.45 vs 0.33 without), and repeating it more often added nothing.

One finding surprised me. Plans that participants wrote down and kept were less effective (0.31 against 0.44). The authors suspect that having the plan in front of you invites re-deciding it, which weakens the automatic link. It is a comparison across different studies rather than an experiment, so I wouldn’t stop writing plans down. I would stop renegotiating them every morning.

Implementation intention examples

Each plan below names a situation you will actually run into and one small first action. Swap in your own details; the structure is what carries over.

Exercise

• If I get home from work on Monday, Wednesday or Friday, then I change into my gym clothes before I sit down.
• If I walk into the office building, then I take the stairs.
• If the alarm goes off at 6:45, then I put on my running shoes and step outside.
• If I finish my morning coffee, then I do 10 squats next to the kitchen table.

Food and drink

• If I order lunch, then I add a side of vegetables.
• If I want something sweet in the afternoon, then I eat a piece of fruit first.
• If someone offers me a second drink, then I ask for sparkling water.
• If I unpack the groceries, then I wash the fruit and put it at eye level in the fridge.

Work and study

• If I open my laptop in the morning, then I work on the hardest task for 25 minutes before opening email.
• If I notice I’ve opened social media while studying, then I close the tab and read the next paragraph.
• If a meeting ends, then I write the next step in my task list before leaving the room.

Health admin and small routines

• If I fill the kettle in the morning, then I take my tablet with the first glass of water.
• If it’s the first Sunday of the month, then I book the appointment I’ve been putting off.
• If I put my toothbrush back in the cup at night, then I floss.

Sleep and screens

• If it’s 22:30, then I plug my phone in to charge in the kitchen.
• If I get into bed, then I pick up a paper book instead of my phone.

Feelings and self-talk

The then-part doesn’t have to be a physical action. In the meta-analysis, plans that told people what to think or say to themselves also worked. “Then I tell myself to stay calm” had one of the larger effects, d = 0.69 across 14 tests, mostly on emotional reactions.

• If I feel nervous before a presentation, then I tell myself: stay calm, I know this material.
• If I catch myself thinking “I’ll skip it today,” then I remind myself why I started.
• If I feel the urge to check my phone, then I ignore it and finish the sentence I’m writing.

Coping plans: if-then plans for obstacles

An action plan covers the normal case. A coping plan covers the predictable ways it goes wrong: you’re tired, it’s raining, the gym is closed, someone puts cake in front of you. The form is the same, with the obstacle in the if-part.

In a trial with 211 cardiac rehabilitation patients, Sniehotta, Scholz and Schwarzer compared action plans alone, action plans combined with coping plans, and standard care. Two months after discharge, the combined group exercised significantly more than the other two. In the 2025 meta-analysis, plans whose cue was a setback averaged d = 0.31, and plans telling people to ignore a distraction averaged 0.48.

To write one, list the two or three things that most often derail the habit and give each a fallback. Making the backup a smaller version of the habit works well, because the day still counts:

• If it’s raining at 7:00, then I do the 15-minute workout video in the living room.
• If I get home too tired to cook, then I make the eggs-and-salad dinner instead of ordering in.
• If I miss my morning walk, then I walk after lunch.
• If I’m too tired for 10 pages, then I read one page.
• If I’m ill, then I take the day off from the habit and start again the first day I feel better.

If-then plans and habit stacking

Habit stacking, attaching a new habit to one you already have (“After I pour my coffee, I write three lines in my journal”), is an implementation intention that uses an existing routine as its cue. That fits the meta-analysis: cues tied to a moment in an ongoing task were among the most effective. An old routine also supplies the cue at the same time and place every day, which is what a new habit needs to get repeated. How long it then takes to become automatic is a separate question, and the answer is usually longer than people hope.

Does it matter whether the new habit goes before or after the anchor? In a small study with 50 people, those told to floss after brushing had stronger flossing habits eight months later than those told to floss before. A larger follow-up with 118 people found no difference. So pick whichever order is easiest to remember, and choose an anchor that happens on every day you want the habit. If you are torn between a morning and an evening anchor, this piece on time of day goes through the trade-offs.

When implementation intentions don’t work

• The goal is weak. A plan amplifies motivation you already have. If you don’t really want the outcome, a perfect if-then sentence won’t change much, and the authors think this is why the effects on exercise, vaccination and voting were small or absent.
• The cue never shows up. “If I have some free time” is not a cue. Neither is a situation that happens twice a month, if the habit is meant to be daily.
• Too many plans at once. People asked to make three to five plans did worse (d = 0.07 to 0.22) than those who made one (0.41). The authors warn this may be tangled up with the kind of task studied, so treat it as a hint: one plan per habit, plus a coping plan or two.
• The plan is over-engineered. Adding how and how long cut the effect roughly in half.
• Lab results don’t fully transfer. Field effects were about half the size of lab effects, and most of the evidence comes from university students in the UK, Germany and the US.

Because of publication bias, the true average is probably at the lower end of the published estimates. What the evidence supports is narrower and still useful: if you want something and keep failing to start, deciding the exact moment and the first action in advance makes it more likely that you will. A plan doesn’t make a behaviour automatic by itself. That takes repetition, and the occasional missed day is part of it (what a missed day actually does).