Transparent method

How KnowsYou estimates a meal from words or a photo.

The AI agent identifies likely foods, uses the quantities and preparation details available, and saves an approximate meal record for you to review. A photo can be a useful starting point, but it is not a scale or nutrition label.

The journal entry is the useful record. Review can correct assumptions, but values remain estimates unless you replace them with verified information.

Example approximate meal record with calories and macronutrients
Lunch · 12:42

Chicken wrap and apple

Approximate

One small tortilla, grilled chicken, vegetables, no sauce, and one medium apple.

Estimated energy520 kcal
Protein
38 g
Carbs
58 g
Fat
15 g
Fiber
8 g

Example only. Review and correct the estimate when details matter.

In short: more specific input reduces avoidable assumptions, but no description or image makes an estimated meal exact. · About 8 min read

What the result means

A KnowsYou meal record is a structured approximation of what you reported. It may contain individual food items, an estimated quantity and available estimates for calories, protein, carbohydrates, fat and fiber. It is not a chemical analysis of the meal and does not establish exactly what you consumed.

The primary value is the journal entry: preserving what, when and roughly how much you ate so that you can revisit the day. Nutrition numbers add context, but they should not be treated as more precise than the information used to create them.

Approximate stays approximate

A detailed-looking number can still come from uncertain food identification or portion assumptions. Display precision is not the same as measurement accuracy.

Inputs the estimate can use

The agent works from the information in your message and any current image you ask it to use. Useful inputs can include:

  • Food and drink names in ordinary language.
  • A stated amount such as grams, milliliters, pieces, servings or package size.
  • Ingredients, brand, recipe, preparation method, sauces and cooking fats.
  • Visible food and relative portion cues in an attached photo.
  • A correction you provide after reviewing the first record.

User-supplied quantities and nutrition facts are more useful than visual guesses, but the agent can still misunderstand what you wrote. Check them in the saved record.

The estimation workflow

The current workflow follows four practical stages:

1 · Identify

Break the meal into likely food and drink items using your description and, when supplied, visible image cues.

2 · Quantify

Use amounts you supplied. Where an amount is missing, infer an approximate portion from the available context.

3 · Estimate

Produce approximate nutrition per item and calculate the meal's saved totals from those items.

4 · Review and correct

Show the saved record so that you can replace a wrong food, quantity, ingredient, time or meal label through conversation.

This is an AI estimation process, not a laboratory procedure. KnowsYou does not publish a universal accuracy percentage because error changes substantially with the meal and the input quality.

What one meal photo cannot reveal

A two-dimensional image can suggest foods and relative sizes. It usually cannot determine exact mass, volume or composition. Important unknowns commonly include:

  • Weight, serving depth and food hidden underneath other items.
  • Oil, butter, dressing, sugar and other ingredients mixed in.
  • Brand, recipe, substitutions and nutrition-label values.
  • Preparation method and how much cooking fat was absorbed.
  • Whether everything visible was actually eaten.
  • The scale of the plate when no reliable reference is present.

Similar-looking dishes can therefore have materially different calories and nutrients. Multiple images may provide more visual context, but they do not remove these basic limitations.

Details that make a record more useful

Add detail in proportion to why you are logging. A quick food journal can stay simple. If a nutrition estimate matters, the following information can reduce avoidable assumptions:

DetailExampleWhy it helps
Measured amount150 g chickenReplaces a visual portion guess
Package or labelOne 200 ml bottleClarifies product and serving size
PreparationFried in one teaspoon of oilAdds information a photo may hide
Recipe detailDressing served separately; used halfDistinguishes what was served from what was eaten

More detail can improve usefulness, but it does not turn an AI estimate into an exact measurement.

Corrections are part of the method

The first estimate is not final. Ask the AI agent to find the recent meal and correct the full record—for example, “The rice was 200 grams, not one cup,” or “Remove the dressing and add one tablespoon of olive oil.”

When food items change, the meal totals are recalculated from the updated items. Review the result again: a correction can still be misunderstood, and earlier uncertainty may remain.

A correction improves the record, not its certainty

Corrected values remain approximate unless you supplied verified values for every relevant item and amount—and even then, the record only reflects what was entered.

Missing details and daily summaries

Missing is not the same as zero. An individual food can have no numeric value for a nutrient. The current summary calculation treats that missing contribution as zero when building meal and period totals. A displayed zero—or any total built from incomplete nutrient values—therefore does not confirm that the nutrient was absent or that the total is complete. This is a current summary-interface limitation. A daily total also cannot prove that every meal, snack or ingredient was logged.

  • A day can have some logged food and still be incomplete.
  • A day with no calorie values is excluded from calorie-intake averages rather than silently counted as zero.
  • A period summary cannot detect an omitted meal on a day that otherwise contains entries.
  • Energy-balance error can combine uncertain intake with uncertain expenditure.

Read the separate daily energy-expenditure method before interpreting an intake-minus-expenditure value.

What the app currently shows about sources

KnowsYou saves the AI agent's structured estimate and the meal information you provided. The app does not currently show a food-by-food citation or source history for each nutrition number. Do not assume a displayed value came from a particular public database, manufacturer or laboratory.

When a number must be verified, use the product label, recipe, measured quantity or another authoritative source and give the corrected value to the AI agent. Because per-food citations are not currently available, KnowsYou does not claim database-backed precision for an individual estimate.

Important limitations and unsupported uses

  • Food recognition can fail, especially for mixed dishes, visually similar foods or partially hidden items.
  • Portion error can materially change calories and every derived nutrient estimate.
  • Logging patterns can provide context, but they do not establish a medical cause or effect.
Never use an estimate for medical decisions

Do not use photo or meal estimates to dose insulin or other medication, manage an allergy or medical condition, or replace a prescribed clinical-nutrition method. Use verified nutrition information and guidance from a qualified professional when the consequence of an error is medical.

Food and calorie tracking can also be distressing or harmful for some people, including people with a current or past eating disorder. Stop tracking if it feels harmful and seek appropriate support.

References

Published · Method version 1.0 · Last changed

Changelog

1.0 · July 22, 2026 — Initial publication of the current inputs, estimation stages, correction workflow, missing-data rules, source boundary and safety limitations.

Keep the journal useful and the estimate in perspective.

Start with the information you have, review the saved foods and correct the details that matter.