← All posts
FrancisAccountability
The psychology of accountability: the key to adherence, and how AI can bring it to everyone
The psychology of accountability explains why people stick to health plans. What the research shows, and why AI may be the way to reach everyone.

I came to the psychology of accountability the hard way. After a decade as a software engineer, I was in the worst shape of my life. I would start a program on Monday and quit by the weekend. Then I paid an online accountability coach $400 a month, and he texted me about my health every day for three years. I got into the best shape of my life.
I already knew the basics. I was paying for his attention.
That is why I care about this. My claim is that accountability is the key to adherence, which means actually following a health plan, and that AI is the only realistic way to give it to everyone who needs it. The evidence that accountability helps is good. The evidence that AI can deliver it is early, and I will show you where it runs out.
Adherence is the unsolved problem in health
Most people already know the advice. They know to eat more vegetables, move most days, sleep enough and take the medicine their doctor prescribed. The hard part is doing it on an ordinary Tuesday when nobody is watching.
Psychologists call this the intention-behavior gap. Rhodes and de Bruijn pooled 10 studies with 3,899 people and found that nearly half of the people who intended to exercise at guideline levels did not follow through (Rhodes and de Bruijn, 2013).
Medicine has the same problem. In 2003 the World Health Organization reported that adherence to long-term treatment among people with chronic diseases in developed countries "averages only 50%." That figure is about medicines, and it is a rounded summary of earlier reviews.
A review of 17 studies of apps for chronic disease found that an average of 43% of participants dropped out (Meyerowitz-Katz et al., 2020).
Most of the people in these studies had a plan or a prescription. I think adherence is the biggest unsolved problem in health, because most treatments for long-term conditions depend on someone doing them at home for months.
Why "try harder" is weak advice
When people fall off a plan, the usual explanation is a lack of motivation or discipline. I think that explanation is incomplete, and several lines of research point the same way.
Raising motivation moves behavior less than you would expect. Webb and Sheeran looked at 47 experiments that changed people's intentions. A medium-to-large change in intention produced a small-to-medium change in behavior, about half the size (Webb and Sheeran, 2006).
The popular idea that willpower is a tank that runs dry has also failed two large tests. Repeat studies of the classic lab task, one run in 23 labs and one in 36, found almost no effect (Hagger et al., 2016; Vohs et al., 2021).
Self-control seems to work mostly through habits. In six studies, mostly of students, habits explained much of the link between self-control and good results. In one of them, people with more self-control had stronger habits around snacks, exercise and sleep, and said they had to resist temptation less (Galla and Duckworth, 2015).
To be fair, motivation matters. Intentions do predict behavior, and a review of weight-loss research found that an autonomous, self-motivated style was among the best predictors of success (Teixeira et al., 2005). My point is narrower. Motivation gets people started, but it does not reliably carry them through, so telling someone to want it more is a weak lever for the gap between meaning to and doing. I think anyone who starts a new plan on a Monday already wants it.
I wrote an earlier post on accountability and willpower in weight loss.
What the psychology of accountability says
When I say accountability, I mean that you expect to report what you did to someone, and that person can see whether you did it. Four lines of research show how that changes behavior.
Reporting your progress to someone
Harkin and colleagues pooled 138 randomized trials with nearly 20,000 people. Tracking progress toward a goal helped people reach it. The effect was larger when people reported their results to someone or made them public, and when they wrote them down (Harkin et al., 2016).
Someone being able to see your data
Butryn and colleagues randomized 322 adults in a 12-week weight-loss program. For half of them, the coach could see their scale, activity and food data. Those people logged food on more days (98% of days vs 94%), lost a little more weight (6.2% of body weight vs 5.3%), and said they felt more accountable (Butryn et al., 2024). The effect is small. I still think it is the cleanest test of the idea, because the only thing that changed was whether someone could see.
Being accountable to a person you trust
Mohr, Cuijpers and Lehman proposed a model they called supportive accountability. It says people stick with a program because they feel accountable to a coach they see as "trustworthy, benevolent, and having expertise" (Mohr et al., 2011). The authors present it as a model to be tested. It also says that the more motivated someone already is, the less support they probably need. As written, it is about human coaches. I will come back to that.
Deciding in advance
Plans that name when and where you will do something, which psychologists call implementation intentions, help people follow through. A review of 94 studies found these plans had what the authors call a medium-to-large effect (Gollwitzer and Sheeran, 2006).
Commitment devices, where you lock yourself into a choice ahead of time, also work for the people who use them. In a smoking trial with 2,538 people, only 13.7% accepted a program where they put down their own money. For people willing to accept either program, though, the deposit raised quit rates by about 13 percentage points over a plain reward (Halpern et al., 2015). I think of a daily commitment made to someone who will ask about it later as a much lighter version of the same idea.

Coaching works when people get it
In the POWER trial, 415 patients with obesity and cardiovascular risk factors were followed for two years. The self-directed group lost 0.8 kg. The group coached remotely by phone, web and email, with no face-to-face visits, lost 4.6 kg, and the group coached in person lost 5.1 kg (Appel et al., 2011). About twice as many in the remote coaching group lost at least 5% of their weight (38.2% vs 18.8%), and remote and in-person coaching did not differ. The coaching came bundled with a website and education, so the trial cannot isolate accountability. What I take from it is that regular contact by phone and email was enough.
The Diabetes Prevention Program is the best-known example. Among 3,234 people at high risk, the lifestyle program cut new cases of diabetes by 58% compared with placebo over about three years (Knowler et al., 2002). That 58% came from a whole package, so I can't credit accountability alone. The package was this: each participant had their own lifestyle coach, met 16 times in the first 24 weeks, recorded fat and calories every day during those weeks, and stayed in regular contact after that (DPP Research Group, 2002).
There are not enough coaches for everyone
In 2023, 76.4% of US adults, roughly 194 million people, reported at least one of 12 common chronic conditions, such as high blood pressure, diabetes or depression (Watson et al., 2025). The US has about 86,300 dietitian and nutritionist jobs (Bureau of Labor Statistics) and about 12,000 board-certified health and wellness coaches, by the board's 2024 count (NBHWC). Not everyone with a chronic condition needs or wants a coach, and doctors, nurses and pharmacists also follow up with patients. Treat the numbers below as an illustration.
By my own arithmetic, that is about 2,250 adults with a chronic condition for every dietitian, and about 16,000 for every certified coach. If a coach spent five minutes a day with each person over a 7.5-hour workday, they could reach about 90 people. Covering 194 million people that way would take around 2 million coaches. Personal trainers and coaches without the certification add more people, and I still don't think the numbers come close.
Right now, daily accountability from a person is something only a few people can buy. I paid for it. Even if everyone could pay, there would not be enough coaches.
Can AI provide accountability?
Whether AI can provide accountability is the part of my thesis with the least evidence, so I want to be careful with it.
What has been shown
The strongest trial I found, published in JAMA in 2025, randomized 368 adults with prediabetes to either a fully automated program with no human in the loop or a human-led Diabetes Prevention Program delivered over group video. The same share of each group hit the trial's goal, 31.7% with the automated program and 31.9% with human coaches. More people in the automated group started the program (93.4% vs 82.7%) and finished it (63.9% vs 50.3%) (Mathioudakis et al., 2025).
That app was not a chatbot. It used a reinforcement learning algorithm that did not use large language models, along with push notifications and a Bluetooth scale. The human program it was compared against met as a group over video, so the trial tells us nothing about one-to-one coaching. And about two-thirds of people in both groups missed the goal. Even so, in the first randomized trial to compare them head to head, a fully automated program did as well as human-led group coaching at helping people with prediabetes reach weight, activity or blood sugar goals.
What has not been shown
Mohr's model says the coach has to be someone people trust, and the authors had a human in mind. No health trial has tested whether people feel accountable to software in the way they feel accountable to a person, so whether software can be that trusted someone is still unknown. In the same diabetes trial, people rated human coaching as more acceptable in every area measured, and 46.5% of the automated group said they would have preferred a human coach, while 31.5% of the human-coached group would have preferred the automated program (Lalani et al., 2026). In an older weight-loss trial, automated feedback matched human email counseling at three months, and the human group pulled ahead by six months, 7.3 kg lost vs 4.9 kg (Tate et al., 2006).
Chatbot trials for weight and diet are thin. A 2025 review of seven randomized trials of natural-language chatbots found they helped adults with smoking, but found no such evidence for diet or physical activity (Chen et al., 2025). In a review of conversational agents for weight management, only three of eight studies reported substantial weight loss, of 1.3 to 2.4 kg, and overall quality was low (Noh et al., 2023).
Put together, software has matched human group coaching once, on short-term goals for people with prediabetes. Whether people feel the same accountability to software is still an open question, and in the one trial that asked, many did the work but wanted a person. For weight and diet the evidence is still thin, and two trials of automated programs found no effect on blood pressure or on taking heart medicines (Persell et al., 2020; Ho et al., 2025).
The bet
BodyBuddy is a bet that a coach that texts you every day, remembers what you said you would do, and can see what you logged creates enough of that feeling of being expected to change what you do. One finding gives me some hope. In digital mental health programs, guides without clinical credentials did about as well as clinicians (Leung et al., 2022). That study did not test accountability directly, so I count it as a small reason for hope.
Here is what would prove me wrong. If a fair randomized trial of a conversational AI coach for diet and activity found no difference against tracking alone, or found that the benefit fades after a few months the way automated feedback did in the Tate trial, then the "at scale" half of my thesis would be wrong. If that happens, I want to know.
How BodyBuddy does it
BodyBuddy is an AI health coach that texts you every day. You commit to what you're doing for your health that day, and it texts you to hold you accountable, to make sure you actually did it, even on the bad days.
- You text it what you ate, or send a photo.
- If a doctor, dietitian, physical therapist or trainer gave you a plan, you can hand BodyBuddy the plan and it follows up with you between appointments.
- It texts you first, so you do not have to remember to open an app.
BodyBuddy is built on this psychology, so it does not ask you to find more motivation or discipline. Those features copy the parts of accountability the research points to: deciding in advance, reporting what you did, and having your log seen. In those studies the one doing the seeing was a person, so whether software gets the same effect is the open question above.
BodyBuddy is fully AI. There is no human behind the chat. It does not diagnose anything, and it does not replace your doctor. I built it to give everyone the daily accountability I paid a coach for.
Frequently asked questions
Why does accountability work?
When you expect to report what you did to someone, you are more likely to do it. In a meta-analysis of 138 trials, tracking progress helped people reach their goals, and it helped more when they reported the results to someone (Harkin et al., 2016).
What is the intention-behavior gap?
The intention-behavior gap is the distance between what people mean to do and what they actually do. In exercise research, nearly half of the people who intended to be active at guideline levels did not follow through (Rhodes and de Bruijn, 2013).
Can an AI hold you accountable?
The evidence is early. One rigorous trial found that a fully automated diabetes prevention program did as well as human-led group coaching (Mathioudakis et al., 2025). That program was not a chatbot, and no health trial has yet tested whether people feel the same accountability to software that they feel to a person.
Where to start
Write down one thing you will do for your health today. Make it specific enough that someone could ask you tonight whether you did it. Then tell a person who will actually ask.
If you don't have someone who will ask every day, that is the job I built BodyBuddy to do.
Sources
Related BodyBuddy guides
- Can AI Analyze Your Fitbit Data? Insights Are Not the Bottleneck
- Will AI Replace Doctors? The Better Role for Healthcare AI
- Meet Future You: BodyBuddy’s 90-Day Health Vision
- Accountability Partner vs. Accountability App: Which One Actually Helps You Lose Weight?
- BodyBuddy vs Poke: Which AI Is Better for Health?
Everyone gives you a plan.BodyBuddy texts you until it happens.
BodyBuddy keeps you on track—from meals to doctor's appointments and everything in between.