Companion to the book

Lose Weight With AI

Every prompt from the book, ready to copy into Claude, ChatGPT, Gemini or Copilot — plus the printable 90-day planner. Free, no sign-up, nothing to buy.

Five minutes of setup

The difference between a generic answer and a useful one is almost entirely context. Do this once.

1. Make a dedicated space. In Claude, create a Project called “Weight” — it gets its own isolated memory. In ChatGPT, create a Project or one long conversation you return to. In Gemini, turn on memory under Personal Intelligence.
2. Paste the setup prompt. It's the first one in the Prompts tab. Fill in your details honestly — especially the line about what has failed before.
3. Come back weekly. Paste your numbers in with the “Weekly data drop” prompt. That's what turns a chatbot into something resembling a coach.

Three rules that matter more than the prompts

Consistency beats intensity. In a 2022 study by Payne and colleagues, what predicted weight loss was not how completely people logged — completeness didn't reach significance — but how many weeks they logged consistently at all.

The next meal is always the restart point. Not tomorrow. Not Monday. The hours between a slip and “I'll start again Monday” are where nearly all the damage happens.

Have a floor, not a ceiling. Define the minimum you'll do on your worst day. On a catastrophic day, hitting the floor is a complete success — that's what keeps the streak, and the streak is what gets you to day 200.

The daily page

Two minutes. It collects data — it does not evaluate you. That distinction is the whole design.

Date__________
Weight__________
7-day average__________
Protein targethit / missed
Steps__________
Sleep (hours)__________
Movement__________
Day out of 10__________

One line about today

Tomorrow's one thing

The weekly review

Ten minutes, same day each week. This is the only place you draw conclusions. A single day has too much noise in it, and judging days is how the all-or-nothing spiral starts.

This week's average weight__________
Last week's average__________
Change__________
Protein hit___ / 7 days
Average steps__________
Training sessions___ / ___

What worked

What didn't

ONE change for next week

How to read the number

What you seeWhat it means
Down 0.3–1% of body weightOn track. Change nothing.
Flat for one weekNormal noise. Change nothing.
Flat for 3+ weeksAudit your logging for a week before cutting anything. Drift in portion size is the usual cause.
Down more than 1.5%Too fast. More muscle loss, more metabolic adaptation, worse odds of keeping it. Eat a bit more.
UpCheck context first — salt, a big meal, poor sleep, hormonal phase, new training. Only act if it's up three weeks running.
The question that matters most at each review: could I have done this week for the next year? If not, it doesn't matter that the number went down — you've just demonstrated something temporary.

Tip: use your browser's print function on this page to get a clean printable copy of the planner.

Things the industry won't tell you

AI cannot count your calories

Four independent research teams across three model families, 2025–2026, reached the same conclusion. Food identification is good — 85–93% accurate. Portion estimation is not. O'Hara and colleagues (Nutrients, 2025) found AI estimated 530g for meals that actually weighed 798g, and all models underestimate more as portions grow.

The trap: at group level, their energy error was −0.1%, which looks perfect. The mean absolute error across nutrients was 26.9%. Errors cancel in aggregate; they never cancel for the meal in front of you.

Use photo logging for consistency, not precision. Weigh the calorie-dense things — oil, nuts, cheese, butter. Assume the number is low. And add text context to the photo: describing the ingredients and cooking method measurably improves the estimate.

Most weight-loss supplements don't work

The FDA states plainly that it “does not have the authority to approve dietary supplements before they are marketed.” No proof of efficacy is required. No proof of safety is required.

ClaimWhat the evidence says
Garcinia cambogiaIn the definitive JAMA trial (1998), the placebo group lost more — 4.1 kg vs 3.2 kg.
Green coffee beanStudy retracted 2014. FTC fined the company $3.5m; the investigator had altered subject weights.
Apple cider vinegarThe big 2024 trial was retracted in September 2025 — analyses couldn't be replicated, dataset missing.
Raspberry ketonesNo published human trial of the ingredient alone exists.
Proprietary blendsLegally need only disclose a combined total weight. If the dose isn't disclosed, treat it as absent.

And the safety data is worse than the efficacy data. Weight-loss and energy products accounted for 71.8% of all supplement-related cardiovascular emergencies in a ten-year NEJM surveillance study — most of them in adults aged 20–34. Of 776 adulterated supplements found in FDA warnings 2007–2016, 317 were weight-loss products, and 269 of those contained sibutramine, a drug withdrawn in 2010 over heart attack and stroke risk.

Exercise is a weak weight-loss tool

Exercise alone produces roughly 1.5–3.5 kg of weight loss. Pontzer and Trexler's 2026 review found total daily energy expenditure rises by only about 30% of what additive models predict — the rest is compensated away, mostly by moving less for the remainder of the day. And compensation is higher in people carrying more weight.

That's not a reason to skip it. It's a reason to stop measuring it by the scale. Exercise determines what your body can do, how you age, and what you look like at whatever weight you land on — and resistance training appears to trigger less compensation than cardio.

And the famous number is wrong: muscle burns about 13 kcal per kg per day, not 50 per pound. Gaining 3 kg of muscle buys roughly 40 calories a day. Lift for body composition, strength, and ageing — not for metabolism.

HIPAA does not cover any of this

HIPAA governs doctors, hospitals and insurers. Consumer health apps and AI assistants sit entirely outside it. A food photo also carries EXIF location and timestamp data, and frequently contains faces and surroundings. Check the “Data Used to Track You” section of any app's store listing before installing.

Where the line is

The tools here — tracking, weighing, logging — work, and for some people they're the mechanism by which a healthy intention becomes a disorder. If your forbidden-food list keeps growing, if the number sets your mood for the day, if you're avoiding social events because of food, or if you're thinking about this most of the day, stop tracking and talk to someone.

In the US, the National Alliance for Eating Disorders helpline is 1-866-662-1235, staffed by licensed clinicians. Eating disorders occur at every body size.