Can an if-then plan close the gap between intention and action?

A goal does not decide the moment

I want to exercise more describes a direction. It leaves the next moment open: when exactly, under what conditions and what counts as starting? An implementation intention closes part of that gap by linking a cue to a response: If I finish my last meeting on Tuesday, then I put on my shoes and walk for twenty minutes.

The cue can be a time, place, situation, feeling or preceding action. The response can start a desired behavior or block an unwanted one. What matters is contingency: one recognizable moment is connected in advance to one move.

642 tests asked whether cues can trigger action

Researchers searched psychology databases, citations and earlier reviews for experiments published or available between 1994 and September 2019. They found 294 eligible reports containing 642 independent treatment-versus-control tests. Studies had to isolate a planning intervention and report a later behavioral, cognitive or affective outcome.

The corpus was broad but uneven. Most tests involved adults, participants were 60.4% women on average and the mean age was 28.6. The typical experiment was small: the median arm contained 30 planning participants and 28 controls, although a few were much larger. The review itself was not preregistered; the authors say they developed its component categories inductively.

The headline effect shrank under tougher analysis

The conventional random-effects meta-analysis estimated d = 0.36, with a 95% confidence interval from 0.33 to 0.40. On that analysis, if-then planning improved outcomes by a small-to-moderate amount on average.

But the literature showed extreme heterogeneity and strong evidence of publication bias. A robust Bayesian model that jointly allowed for an effect, heterogeneity and selection estimated d = 0.15, with a 95% interval from 0.08 to 0.22. Its evidence favored a non-zero effect, but the estimated size was less than half the conventional headline.

Both results belong in the answer. The evidence supports a non-zero average effect, while the conventional headline probably overstates how much help a new plan should be expected to provide.

A true contingency mattered more than a schedule

Across studies, plans in an explicit if X, then Y format produced a larger estimate than schedule-style plans: d = 0.43 versus 0.29 in the conventional moderator analysis. Rehearsing a plan at least once was also associated with a larger result, while further repetitions did not add a detectable benefit.

These are comparisons across groups of studies, not perfect head-to-head experiments holding every other feature constant. They make the cue-response logic plausible; they do not prove that one sentence template is optimal for every goal.

Planning cannot manufacture motivation

Implementation intentions worked better when people already had strong, active or self-concordant goal intentions. In the physical-activity subset, the estimate was d = 0.31 when intentions were strong and approximately zero when they were weak.

That distinction prevents a common misuse. A precise plan can help execute a goal someone values. It cannot decide whether the goal is worthwhile, create commitment that is absent or remove structural barriers such as time, money, health and care responsibilities.

Pre-decide the smallest next action

Our editorial inference is deliberately narrow. Choose a cue you are likely to notice and a response you can actually perform. Keep the response concrete enough that the cue answers what now? without another negotiation.

Then treat the plan as a hypothesis. If the cue rarely occurs, choose one you are more likely to encounter. If the response is too large to perform, reduce it. Planning can make action more executable. The next unresolved question is whether observing progress after action begins helps carry the goal through.

Standardized mean difference · d

Conventional random effects95% CI 0.33 to 0.40

0.36

Robust Bayesian model95% interval 0.08 to 0.22

0.15

Sheeran, Listrom and Gollwitzer (2024), 642 experimental tests. Heterogeneity and publication-bias evidence were substantial.