DUNCAN McLEOD: How ChatGPT became my health coach

For the past few weeks, I’ve been running a quiet experiment on myself.

Every time I have a meal or reach for a snack, I pull out my phone, snap a photo and describe to ChatGPT what’s on the plate. Sometimes I type it. More often I just talk to it: “A cup of chickpeas, a handful of baby spinach, a floret of broccoli, a drizzle of olive oil, a bit of feta.”

The app logs it. And slowly, without me filling in a single spreadsheet cell or using an infuriatingly slow traditional diet-logging app, a picture of how I actually eat has started to form — one that I can query at any time.

For years, I’ve half-heartedly tried the usual suspects — MyFitnessPal, Cronometer — and abandoned all of them. The friction was always the killer: scanning barcodes, hunting for the right food in a database, guessing portion sizes from menus. It was always too much effort. ChatGPT has removed that admin overhead. I snap a picture, I describe what I’m eating, and it logs it. That’s it.

What has surprised me is how much more the tool does than simply keep a record. When a day’s meals have been light on vegetables, it tells me. When my sodium intake starts creeping up, it flags it and suggests where the salt is sneaking in — usually the processed stuff I didn’t think twice about before. It’s driven me towards raw, unprocessed plant food.

I’ve deliberately created an ongoing “diet project” inside ChatGPT, which means it remembers context across sessions. It knows I run. It knows my weight and the kind of food I gravitate towards. Over time, it has steered me, without being preachy about it, towards more whole foods, more fibre, more plants and fewer refined carbs.

The diet log is only half of it. I’ve also been feeding the same conversation a stream of health and fitness data. My smartwatch tracks my heart rate around the clock, including my resting rate, which has become one of the most useful numbers in my life.

It counts my steps too. I aim for a minimum of 10,000 a day and although I sometimes fall short, on running days I sail well past it. An app-linked blood pressure cuff gives me a reading I can share with ChatGPT in seconds — and the readings have plunged from borderline hypertension to “optimal” in a matter of weeks. Strava, which collects data from my smartwatch, handles the exercise side, logging parkruns and trail runs and adding a layer of social accountability that is always motivating.

It connects the dots I’d otherwise miss. And I can interrogate it with questions

Feeding all of this into one conversation turns ChatGPT into something I hadn’t quite expected: a pattern matcher. It connects the dots I’d otherwise miss. And I can interrogate it with questions. When my resting heart rate ticked down a couple of beats per minute over a fortnight, it pointed out the likely correlation with a run of consistent training.

When I mentioned feeling shaky after a particularly tough effort at my local parkrun, it suggested — correctly, I think — that I’d probably gone out underfuelled and my blood sugar had crashed.

None of this is rocket science. A decent coach or a GP with time on their hands could tell me the same things. The difference is that ChatGPT is in my pocket at 9pm on a Sunday when I’m wondering what goals to focus on in the week ahead. And it costs me only $20 a month.

Of course, there are also real questions about privacy: I’m handing a US technology company a remarkably intimate record of what I eat, how I move and how my body is behaving. That’s a trade-off each user has to weigh for themselves.

But the productivity lesson here, for a business readership like the FM’s, is clear. The technology doesn’t have to be smarter than a human expert. It just has to be available, patient and good enough — and to meet me where I already am, which is talking into my phone while the kettle boils.

That, I suspect, is where a lot of the real consumer value of generative AI is going to be: in the quiet removal of small annoyances from things we already wanted to do. One of the use cases I’ve unearthed is using it to eat better and live a healthier life.

And that is perhaps the truest measure of AI’s power — not the problems it solves for us, but the ones it finally helps us solve for ourselves.

McLeod is editor of TechCentral