← Back to recent questions

Mostly real, with some hype around the edges. AI does not change the underlying job, which is the boring fundamentals of eating less and moving more. What it does, very well, is take the friction out of the bits that most people give up on, especially food tracking. The single biggest unlock for me was a tool that lets you photograph your plate and have it logged for you in seconds.

I am writing this with two long walks behind me. The 800km Camino in 2024 and the 1,200km Shikoku 88 Temple Pilgrimage some years before. I lost roughly fifteen kilograms on each one. I put it back on again, both times. So while AI was not what got me through the walking, it has been useful for the harder, slower, less heroic problem that comes after. Which is keeping it off.

How I'd approach it

Start with photo-based tracking. The app I have actually used and stuck with is MyNetDiary. You take a photo of your plate, it estimates the calories and macros, and it logs them. That is it. Compared to typing in "one cup of brown rice, half a chicken breast, broccoli", which is what every food tracker has asked of you for the last fifteen years, the photo approach is a different sport. You can do it at the table without anyone noticing. You can do it for a takeaway meal where you would normally have given up. The free tier is generous; the paid tier is A$8.99 a month, and they run regular yearly offers that average around A$60, which is worth taking if you already know you will use it. It also has a very strong App Store rating, which I will admit was part of what made me try it in the first place. Lose It! has a near-identical feature called Snap It, and the bigger name MyFitnessPal has added photo logging too. Try one for two weeks and see whether it sticks before you start comparing.

Make the chatbot your sounding board. Set up a ChatGPT Custom GPT or a Claude Project as a personal coach. The prompt below is the one I use. A real coach is better. A chatbot is available at 10 pm on a Sunday and does not judge you.

You are my weight-management sounding board. Keep replies short.

About me:
- Starting weight: [kg]. Goal: [kg]. Realistic timeline: [months].
- Height, age, activity level: [fill in].
- Food I won't give up: [coffee, wine, dessert on Sundays, etc].
- Food preferences / allergies: [vegetarian, gluten-free, etc].
- Things that have worked before: [walking, meal prep, etc].
- Things that have NOT worked: [shake diets, "no carbs ever", etc].

How I want you to behave:
- Be honest, not motivational. I don't need pep talks.
- When I tell you what I ate, give a short, plain-language read on
  the day. Where would a small change have helped? Where was it fine?
- When I ask about a specific situation (a wedding, a long lunch,
  a bad week), help me plan around it, not feel guilty about it.
- If I tell you something that sounds disordered or unsafe (very
  low calorie targets, skipping meals routinely, weight obsession),
  say so and suggest I talk to a GP or dietitian.
- Never give me an exact daily calorie number as if it's a
  prescription. Aim for direction.

Ask me one short question to get us started.

Use a sample week to test it. Ask the questions you would otherwise google: "I have a wedding Saturday and a long lunch Tuesday, how should I think about this week?", "I am stuck at the same weight for three weeks, what should I look at?".

Use voice mode for the walks. Long walks are one of the most effective things you can add to a weight-loss week. Voice mode on ChatGPT or Gemini is a perfectly good walking companion: talk through what you ate today, what you are hungry for tonight, what is hard about this week. It is not therapy, but it is more useful than most podcasts for actually thinking about the thing.

Fridge-photo trick, healthier version. The same photo-of-the-fridge trick from the recipes question works in this context too. Add to the prompt: "high-protein, lower-energy meals I can make from this for one person, around 500 calories per serve, that take less than thirty minutes". You will get specific dinners that match what is actually on the shelf rather than a list of generic "diet recipes" pulled from somewhere on the internet.

What I'd avoid

Three things to keep an eye on. First, do not trust any AI calorie estimate to the kilojoule. The same plate logged a dozen times will give a dozen slightly different numbers. That is fine for following a trend over weeks. It is not fine for "I have exactly 300 calories left to spend before bed". Aim for direction, not precision.

Second, this is not medical territory. If there is a thyroid condition in play, a medication that affects weight, a history of disordered eating, or anything else that makes weight loss complicated, see a GP or an accredited practising dietitian first. AI is not a clinician and it should not be the first line of advice for any of those situations. The Health and Wellbeing page covers how to use AI to prepare for those appointments rather than to replace them.

Finally, do not get pulled into the "best diet" debates with a chatbot. Ask any AI which approach works best and you will get a confident, well-formatted answer that is not actually the answer. The honest answer is the approach you can stick to for six months. AI cannot tell you which one of those that is for you. It can help you stay with the one you have already chosen, which is most of the value.

The bit AI does not solve

The bit no app fixes is the bit you already know. When the leftover slice of cake in the fridge calls your name at ten at night, no AI is going to save you. The tools take the friction out of tracking, of food choice, of asking a question at the wrong hour. They do not take the friction out of the actual moment. That is built by repetition, not by app. After a week on the Camino you stop thinking about whether to walk. The walking is just what you do that day. The fifteen kilos came back both times because that repetition ended when the walks did. AI does not change that arithmetic. It can, sometimes, slow it down.

As of May 2026. Tool features and pricing change quickly; if you are reading this much later, check the current state before relying on the specifics.