How the Score Works

What you see

Every scanned meal gets a letter grade from A+ to F and a 1–10 inflammation-awareness score. Higher grades mean the meal looks more aligned with an anti-inflammatory way of eating; lower grades mean the AI spotted more ingredients commonly associated with inflammatory load.

How the score is produced

When you take a photo, advanced AI vision models analyze the image to identify the foods on the plate, estimate portions, and infer likely ingredients. The score then weighs factors such as:

• Level of processing (whole food vs. ultra-processed)
• Added sugars and refined sweeteners
• Refined grains vs. whole grains
• Industrial seed oils and fried preparations
• Overall balance of the plate (protein, fiber, vegetables)

Barcode scans work differently: packaged-food data comes from the Open Food Facts database, and the same scoring framework is applied to the listed ingredients.

What the score is NOT

• It is not personalized to your symptoms. The grade reflects general anti-inflammatory eating principles, not your individual reactions. The symptom journal exists precisely because your triggers are personal.
• It is not a medical measurement of inflammation in your body. NutriAI estimates the inflammatory potential of foods. It does not measure CRP, cytokines, or anything happening inside you, and it is not a diagnostic tool.

Estimates can be wrong

AI vision analysis of a photo is an estimate. Accuracy varies with meal complexity — a single apple is easy; a mixed curry with hidden oils is hard. Portions, cooking methods, and ingredients the camera can't see (sauces, sweeteners, oils) are inferred, and sometimes inferred wrong. That's why every result is editable before you save it — if the AI misidentifies something, correct it and the log reflects your fix.

Scores may change over time

As the underlying AI models improve, the same meal may receive a slightly different grade than it did months earlier. We consider that a feature — the scoring gets better — but it does mean grades are not frozen in time.

Validation status

NutriAI's scoring has not been externally validated. Nobody independent has checked it.

We do now test it ourselves. Every change to how meals are analysed is scored against a set of real meals before it ships, so we can tell whether a change made estimates better or worse instead of guessing. That has already caught two mistakes worth naming: the estimator was pricing whole cooking pans rather than the portion someone actually ate, and it once counted a tub of Parmesan sitting on the counter as part of the meal.

What that testing cannot yet tell you is how accurate the numbers are in absolute terms. Most of what we compare against is our own earlier output that a user accepted as reasonable — useful for catching a change that breaks things, useless as proof of correctness. Only a handful of meals so far have a verified answer behind them.

So we are not publishing an accuracy percentage. We would rather say nothing than quote a number that sounds authoritative and isn't. When enough meals have verified answers to make a figure meaningful, it will appear here along with how it was measured. Until then, treat every grade as an estimate — a starting point for your own observation, not a verdict.

Questions?

We're happy to explain more. Email support@trynutriai.com.