Describe a meal (AI)

Type or speak what you ate, even a whole day, and AI estimates the calories and macros.

Didn’t take a photo? Just write down, or say, what you ate. Describe a meal takes plain English, one snack or an entire day, splits it into individual foods, works out which meal each belongs to, and estimates the calories and macros for every one. It’s the fastest way to catch up a day you forgot to log.

How it works

  1. On the Food & water page, tap Describe a meal (it’s also in the sidebar).
  2. Type what you ate, or tap Dictate and say it out loud (see below), for example: “Porridge with honey and a flat white for breakfast, chicken caesar salad at lunch, curry and rice with a beer for dinner.”
  3. Tap Estimate calories & macros. If something you wrote could go two very different ways, you’ll get a quick question or two first (see below): otherwise it goes straight to the estimate.
  4. Each food comes back as its own line with a meal, an amount, and its calories, protein, carbs, fat and fiber, plus a running total for everything.
  5. Fix anything that’s off: change the meal, correct the amount (the macros rescale automatically), add or remove items.
  6. Pick the day to log it all to, then Log to diary.

When it asks you a question

Some descriptions genuinely can’t be estimated without one more detail, and guessing wrong would be badly wrong, not slightly wrong. “A cup of sunflower seeds” is the classic: in the shell, most of that cup is shell, so it’s roughly a third the calories of a cup of kernels. Rather than quietly picking one, the AI asks.

  • You’ll get at most a few questions, each with one-tap likely answers, or you can type your own.
  • Anything you leave blank just falls back to a sensible assumption, and Skip: just estimate moves straight on.
  • It only asks when the answer would move the number a long way. Ordinary portions, brands and preparations are assumed, not queried.

Typical questions: shelled or in the shell; rice and pasta weighed cooked or dry; how big “a bowl” was; whether a curry was cream-based or tomato-based; roast in oil or boiled.

Dictating instead of typing

If your browser supports it, a Dictate button appears above the box. Tap it, allow microphone access the first time, and just say what you ate: the words appear in the box as you speak. Tap it again to stop. It’s often quicker than typing out a whole day, and handy while you’re cooking or on the move.

  • Your speech is transcribed to text and then estimated exactly as if you’d typed it: you still see and can edit everything before logging.
  • Dictation adds to whatever’s already in the box, so you can mix speaking and typing, or dictate a bit, pause and carry on.
  • It uses your device’s built-in speech recognition. Support varies by browser: most phones and Chrome-based browsers have it; if the button doesn’t show, just type instead.

Getting better estimates

  • Say which meal: “…for breakfast”, “…at lunch”. Without it the AI guesses from the food and time of day, and you can change it afterwards.
  • Give portions where you know them: “two eggs”, “a large bowl”, “200g of chicken”, “a pint”.
  • Head off the obvious questions: “in the shell”, “cooked weight”, “with the skin on”: and it won’t need to ask.
  • Name the place and the item for anything from a chain: “a Wake-Up Wrap from Dunkin’”, “a Big Mac”, “an 11oz sirloin at Texas Roadhouse”. Where it would make a real difference, the chain’s own published nutrition is looked up and used instead of an estimate.
  • Mention how it was cooked: fried, grilled, takeaway: for home-cooked and generic restaurant food, where it changes the fat a lot.
  • List drinks too: juice, lattes, beer and wine are easy to forget and add up quickly. Say the add-ons the way you ordered them (“large iced coffee, 4 cream, no sugar”).

Where the numbers come from

Estimates are aimed at the true number, never padded “to be safe”: an inflated figure would eat a calorie budget you hadn’t actually spent. Three things sit behind each line on the review screen.

  • Published nutrition, wherever we can find it. Name a chain and its menu item and we go looking for the figure that chain publishes, rather than rebuilding the item from its ingredients. That is the number printed on the menu you were reading, so it wins outright. Big national chains are often covered by public nutrition databases; where they aren’t, we may go further afield for anything substantial enough that being wrong would matter.
  • Restaurant portions read the way restaurants mean them. A weight on a menu is the weight before cooking. An “11oz sirloin” is about 8oz by the time it reaches your plate, and pricing all 11 as cooked meat would overstate it by roughly a third. Chains also publish figures that already reflect how they actually cook a dish, so we don’t add butter or oil you didn’t order.
  • A plausibility check on anything still estimated. Calories have to agree with the protein, carbs and fat given for the same item, and no food is allowed an energy density it couldn’t physically have.

When a figure gets looked up or corrected, a short What we checked note appears above your items saying what changed and why. Nothing is ever altered silently.

It’s still an estimate, so sense-check anything that looks off before logging, and if a figure is wrong just correct it on the review screen. Independent restaurants are the hardest case: there is no published figure to look up, so those lean entirely on the estimate.

Everything you log this way goes into your food diary and counts toward your daily targets, including your fiber goal, exactly like any other entry, so you can edit or delete individual items later from the Food & water page.

A note on fiber: it’s counted inside the carb figure rather than on top of it, so an item’s carbs already include its fiber. Whole grains, beans, fruit and veg carry most of it; meat, fish, eggs, dairy and oils have none. Your fiber ring on the Food & water page only appears once you’ve set a fiber goal under Targets.

Estimates from a description are approximate: they can only be as accurate as what you describe, so check anything that looks off before logging. Not a substitute for a label or a measured portion.

These guides are for using the app. They’re not medical advice: the medication-level estimates in particular are a model for personal tracking only. Talk to your clinician about your treatment.