Behaviour changeAI nutrition coachHabitsDigital healthProduct

    Why a Coach That Messages You First Beats One That Waits

    Shideh HeraviShideh Heravi·16 August 2026

    Every nutrition tool ever built has assumed the same thing: that the user's problem is not knowing what to do. So the tools got better at knowing. Bigger food databases, smarter recipe engines, and now chatbots that can answer almost anything about nutrition.

    And people still do not eat the way they meant to. Because for most of us the missing ingredient was never information. It was the moment.

    The gap that no amount of advice closes

    This is one of the most replicated findings in health psychology, and it should be more famous than it is.

    Roughly half of people who intend to adopt a health behaviour fail to follow through (Sheeran & Webb, Social and Personality Psychology Compass, 2016). Not people who do not care. People who genuinely intended to. In physical activity specifically, one analysis found 46 percent of intenders never acted on the intention.

    The size of the leak is measurable too. Pooling experimental studies, a medium-to-large increase in someone's intention (d = 0.66) produced only a small-to-medium change in their actual behaviour (d = 0.33). Strengthening the intention is simply not the same as producing the behaviour, and the people responsible for the gap are the ones who fully meant to and did not.

    If half of motivated people fail anyway, then a tool whose entire theory is "give motivated people better information" is optimising the wrong half of the problem.

    This reframes what a nutrition coach is for. The bottleneck is not the plan. It is the distance between the plan and the fridge on a Tuesday evening.

    Why waiting to be asked is a design flaw

    A tool that waits for you to open it inherits the exact failure it should be solving.

    Think about when you would need to consult a nutrition app. Standing in a supermarket, deciding what to order, at 4pm when the biscuits are calling. Now think about when you actually remember it exists: rarely, and almost never in those moments. The moments of highest need are precisely the moments of lowest self-regulation, which is when you are least likely to stop and go looking for advice.

    This is why on-demand tools show such steep drop-off. It is not a UX failure. The interaction model requires the user to supply the very thing they are short of, which is the initiative to act well at the difficult moment.

    A tool that speaks first inverts that. It does not need you to remember it. It arrives.

    The evidence for well-timed prompts

    The formal name for this is a just-in-time adaptive intervention: support delivered at the moment of need, adjusted to your current state and context, rather than on a fixed schedule or on request.

    A meta-analysis of 33 studies found moderate to large effects. Against waitlist controls, Hedges's g = 1.65 (k = 9). Against non-JITAI active treatments, which is the harder and more meaningful comparison, g = 0.89 (k = 21). Within-participant changes ran at g = 0.79 (Wang & Miller, Health Communication, 2020;35(12):1531-1544).

    That middle number is the interesting one. Beating a waitlist is easy. Beating an active intervention by that margin says the timing itself is doing work, not merely the presence of support.

    The mechanism is not mysterious. A prompt at the right moment does several things at once: it removes the need to remember, it interrupts an automatic behaviour before it completes, it reduces the decision to a yes or no rather than an open question, and it arrives while the relevant context is still true.

    The honest counterweight

    If this post only cited the number above, you should distrust it. The evidence is promising, not settled.

    A systematic review and meta-analysis of just-in-time adaptive interventions for mental health and psychological well-being found only a small between-group effect, g = 0.15 (2025 systematic review). That is a long way from 0.89, and it is a useful reminder that these interventions work much better for some targets than others.

    Across this literature, risk of bias for adherence and missing outcome data has been rated moderate to high, definitions of what counts as a JITAI vary between studies, and long-term effects are genuinely under-researched. Reviews in this area consistently ask for better control conditions and longer follow-up (systematic review of JITAIs for physical activity).

    So the defensible claim is narrower than the headline: timing plausibly matters a great deal for concrete, action-shaped behaviours such as taking a supplement, hitting a protein target, or choosing a lunch. It is much weaker evidence for anything resembling mood or psychological change.

    Nutrition happens to sit mostly in the first category, which is why we think this design is right for it. But we would rather state the boundary than let a strong effect size from one meta-analysis stand in for the whole field.

    Where proactive goes wrong

    Being proactive is not the same as being noisy, and this is the failure mode worth naming because most apps get it wrong in the same direction.

    Alert fatigue is real. Prompts that arrive too often, or at moments that are not actually relevant, stop being read. Worse, they train you to dismiss the notification reflexively, which destroys the channel for the messages that would have mattered. Research on prompt dosing is explicitly concerned with finding the level that helps without habituating people into ignoring it (microrandomised dosing study).

    Irrelevant timing is worse than silence. A reminder to drink water while you are in a meeting is not support, it is noise wearing the costume of support. The "adaptive" half of the term is doing as much work as the "just-in-time" half.

    Nagging damages the relationship. In a domain as loaded as food, a tool that reads as disapproving will be deleted, and probably should be. Tone is a safety feature here, particularly for anyone with a difficult history with eating.

    The practical implication is that proactive support has to be earned and sparing. Fewer messages, better placed, in a tone that assumes you are an adult.

    How we apply this

    HealthBlokk's coach is built on the premise that the interaction model matters more than the answer quality. It lives in your messages rather than an app you have to remember to open, and it initiates rather than waiting.

    In practice that means the morning plan arrives before the day gets away from you, the supplement reminder lands with a meal rather than at a random hour, and the restaurant help shows up when you are actually out. The vitamin D example is the cleanest illustration: the entire benefit depends on starting in September rather than December, and nobody spontaneously does that in a month when they feel fine. That is not an information gap. It is a prompting gap, and it is exactly the shape of problem this evidence says timing can solve.

    What it does not mean is a stream of notifications. The counterweight section above is a design constraint for us, not a rhetorical gesture. A coach that has to be muted has already failed.

    Practical takeaways

    • The bottleneck is rarely knowledge. About half of people who intend to change a health behaviour do not, so better advice alone predictably underperforms.
    • On-demand tools require the thing users lack. Remembering to seek help at your worst moment is the hardest part of the task.
    • Timing carries real effect. Just-in-time interventions beat active comparison treatments at g = 0.89 across 33 studies, though the field's evidence quality is still uneven.
    • It works better for actions than for moods. The same approach shows only g = 0.15 for mental health outcomes.
    • Fewer, better-placed prompts beat more. Alert fatigue turns a working channel into ignored noise.

    Frequently asked questions

    Why do nutrition apps stop working after a few weeks? Because most require you to open them at exactly the moments you are least likely to. The problem is usually not motivation or knowledge, it is that roughly half of people who intend to act on a health behaviour never do (Sheeran & Webb, 2016).

    Is there real evidence that well-timed prompts work? Yes, with caveats. A meta-analysis of 33 studies found just-in-time adaptive interventions outperformed active non-JITAI treatments with an effect size of g = 0.89 (Wang & Miller, Health Communication, 2020). Evidence quality across the field is uneven and long-term data is limited.

    Does proactive coaching work for everything? No. The same approach shows a much smaller effect for mental health and psychological well-being, around g = 0.15. It appears best suited to concrete, action-shaped behaviours rather than mood.

    Isn't a coach that messages you first just annoying? It is if it is badly designed. Prompts that are too frequent or poorly timed cause alert fatigue and get dismissed reflexively, which destroys the channel. Effective proactive support means fewer, better-placed messages.

    What is a just-in-time adaptive intervention? Support delivered at the moment of need and adapted to your current context and state, rather than on a fixed schedule or only when requested.


    Sources

    1. Sheeran P, Webb TL. The intention-behavior gap. Social and Personality Psychology Compass, 2016.
    2. Wang L, Miller LC. Just-in-the-moment adaptive interventions (JITAI): a meta-analytical review. Health Communication, 2020;35(12):1531-1544.
    3. Effectiveness of just-in-time adaptive interventions for improving mental health and psychological well-being: a systematic review and meta-analysis. PubMed, 2025.
    4. A systematic review of just-in-time adaptive interventions (JITAIs) to promote physical activity. PMC, 2019.
    5. The dosing of mobile-based just-in-time adaptive self-management prompts for caregivers: preliminary findings from a pilot microrandomized study. PMC, 2023.

    Educational content, not medical advice. If you have a history of disordered eating, discuss any food-tracking or prompting tool with your clinician before using it.

    Shideh Heravi

    Written by

    Shideh Heravi

    Newsletter

    Gut health & clean eating, in your inbox

    New posts on food intolerances, gut health and safer eating no spam, unsubscribe anytime.

    By subscribing you agree to receive emails from HealthBlokk about gut health, food intolerances, and product updates. You can unsubscribe anytime.