What an AI life coach is, and what it cannot do

An AI life coach is software that helps you set a personal goal, break it into steps, and keep working at it, using a language model to do the conversational part.

The useful distinction inside the category is not which model it runs on. It is whether the thing initiates. A tool that responds when you open it is an assistant. A tool that contacts you on a schedule and asks about a commitment you made is doing something structurally different, whatever it is called.

Turning a vague want into a sequence. Models are unreasonably good at this, and it is the part most people skip. A goal written as a category cannot be started; a goal written as eleven minutes of work can.

Being available at the moment you are actually thinking about it. Human coaching happens on a calendar. The doubt does not.

Not getting bored of you. A model will take the same goal seriously in week nine, which friends and family reliably will not, through no fault of theirs.

Removing the daily decision. Most days are lost in the negotiation about what today is for. Arriving with the list already written removes that negotiation.

It is not care. AI coaching is not therapy, counseling, or medical treatment. It cannot diagnose or treat a condition and it is not a substitute for a licensed professional. Any product in this category that implies otherwise is one you should be suspicious of.

It cannot read the room. It has your words and nothing else. No tone of voice, no history it was not told, no sense that today is a bad day to push.

It can be confidently wrong. Models produce plausible text. Plausible is not the same as correct, and a plan that sounds reasonable can still be a bad plan for your actual life.

It cannot want it for you. This is the real ceiling. Nothing in this category makes you do anything. The most it can do is make not doing it a conscious choice rather than a drift.

A good human coach is better than any software at the parts that need judgment: noticing what you are avoiding, hearing what you did not say, adapting to the week you have actually had.

What they cannot do at a normal price is show up daily. Most coaching relationships run weekly or fortnightly, because that is what people can afford, and the gap between sessions is exactly where the plan quietly stops.

The two are not really substitutes. Frequency is the axis they differ on, not quality.

AI coaching is one of the few corners of self-improvement with randomised controlled trials behind it, and they are worth knowing before you believe anyone's marketing, ours included.

In a six-month randomised trial, people using a purpose-built goal-attainment chatbot increased their goal attainment by 55 percent, against 24 percent for a control group pursuing the same kinds of goals unaided (Terblanche, Molyn, de Haan and Nilsson, International Journal of Evidence Based Coaching and Mentoring, 2022). The more people used it, the more they gained, and the gains were still there three months after the trial ended.

The same group ran the harder test: two ten-month randomised trials against human coaches (PLOS ONE, 2022). On goal attainment, the chatbot matched the humans.

Two boundaries keep those results honest. The effect stayed on goal attainment: resilience, wellbeing and perceived stress did not move, so a product sold on feeling better in general is selling past the evidence. And the samples leaned on students, mostly from one research group, testing systems built for research rather than for sale.

None of it is a study of Odyssey. What we take from it is the design the trials validated: coaching built around named frameworks, with structured goals and scheduled follow-through, which is what Odyssey is. The full read, including the limits, is here.

Different problem, not a cheaper version of the same one.

Therapy tends to the injury. Coaching trains the athlete.

Coaching starts from the assumption that things are basically working and you want to get somewhere. Clinical care exists for when they are not. If what is in the way is a health problem, coaching is the wrong tool and no amount of daily structure fixes that.

The full comparison, and how to know which one you need, plus where we draw the line in practice.

Four questions that separate them quickly:

• Does it start conversations, or only answer them? This is the main one.
• Does it remember a goal from six weeks ago without you pasting context back in?
• Does it follow up on specific commitments, by name, or does it send generic reminders?
• Is it clear about not being therapy? Products that blur this are telling you something about how carefully they were built.

Largely, yes, and it is worth knowing that before paying for anything. A well-written prompt gets a real coaching conversation out of a general chatbot, and none of the frameworks involved are secrets. Anyone willing to do the setup can get most of the way.

A blank prompt starts over. A coach remembers.

The structural catch is that every prompt requires you to open the app and start it, in a fresh session that remembers none of it. In the week where nothing moved, which is the week a review matters most, that does not happen.