Transparency Report

How Nourva Calculates Your Nutrition

A complete account of every data source, algorithm, and AI system used to estimate the nutrients in your meals, including their known limitations.

Last updated: June 2026 · Nourva Nutrition App

Why we built a multi-source system

No single nutrition database covers every food consumed by pregnant women worldwide. Australian and New Zealand foods are often absent from US-centric databases. Many regional dishes have no published nutritional record. Standard food databases carry macronutrients, iron, and folate, but not iodine or DHA: two nutrients that are critical in pregnancy and difficult to track.

Nourva uses a priority waterfall: the most accurate, laboratory-measured source is always consulted first. AI estimation is used only as a last resort when no measured data is available, and every AI-derived value is clearly labelled as an estimate on screen.

Nutrition data priority order
1
USDA Nutrient Database (SR Legacy 2018) Built-in
Primary source for priority pregnancy nutrients: calcium, protein, iron, folate, iodine, and DHA. Also powers intake limit tracking for sugar, sodium, fat, calories, and caffeine. Sourced from the USDA National Nutrient Database for Standard Reference Legacy (2018) and compiled from the NIH/FDA Iodine Database Release 3 (2023) and USDA SR28 DHA data. Fetched instantly with no API call.
2
Edamam Nutrition API API
Secondary source for broader nutritional data when a food is not matched in the built-in database. If daily quota is reached, the system falls through to Chomp.
3
Chomp Food Database API API
Tertiary fallback activated when Edamam's daily quota is reached. Covers a wide range of packaged and branded foods. If quota is also reached, the system falls through to Gemini.
4
Google Gemini AI Estimation Premium
Last resort when all API sources are unavailable or exhausted. Instructed to reference USDA, NIH, and WHO data. Values below minimum thresholds are discarded and clearly marked "AI est." in the app.

USDA Nutrient Database (SR Legacy 2018)

🏛️
USDA National Nutrient Database for Standard Reference Legacy (2018)
US Department of Agriculture · nal.usda.gov

The primary nutrition database powering Nourva is compiled from the USDA National Nutrient Database for Standard Reference Legacy (SR Legacy, 2018), supplemented by the NIH/FDA Iodine Database Release 3 (2023) and the USDA SR28 DHA nutrient file. Together these datasets cover all priority pregnancy nutrients and intake limit tracking nutrients without requiring an API call.

SR Legacy is the culmination of the USDA's decades of food composition analysis, representing laboratory-measured values for approximately 8,800 foods across 150 nutrient components. Values are standardised per 100g and scaled to the serving size logged by the user.

Priority nutrients tracked: calcium, protein, iron, folate, iodine, and DHA. These are the nutrients most clinically significant in pregnancy and most commonly deficient in pregnant women.

Intake limit nutrients tracked: sugar, sodium, saturated fat, total calories, and caffeine. These are surfaced as soft daily limits the user configures at setup.

Calcium Protein Iron Folate Iodine DHA Sugar Sodium Saturated fat Calories Caffeine Vitamin D Choline Zinc Magnesium
Accuracy note USDA SR Legacy values reflect standardised laboratory-measured reference data compiled in 2018. Actual nutrient content varies by cooking method, food variety, brand, and growing conditions. Nutrient retention during cooking, particularly for folate and vitamin C, is not always accounted for in reference values.

Spoonacular Recipe API

🍽️
Spoonacular Food and Recipe API

Pregnancy-safe recipes displayed in Nourva are sourced from the Spoonacular Food and Recipe API. Spoonacular provides a large database of recipes with full ingredient breakdowns and nutritional information per serving. Nourva queries Spoonacular with pregnancy-relevant filters to surface recipes that are appropriate for each trimester.

Nutritional values for recipes are calculated by Spoonacular based on ingredient quantities and their corresponding USDA nutrient data. As with all nutrition estimates, actual values vary by specific brand of ingredient used, preparation method, and portion size.

Recipe ingredients Per-serving nutrition Pregnancy-safe filters
API Reference
Spoonacular. Food and Recipe API. https://spoonacular.com/food-api/docs

Local Iodine Database

🫘
USDA / FDA / ODS-NIH Iodine Content of Common Foods
Release 3, March 2023 · NIH Office of Dietary Supplements

The iodine database embedded in Nourva is compiled from the NIH Office of Dietary Supplements (ODS) Database for the Iodine Content of Common Foods per Serving, Release 3 (2023). This dataset aggregates analytical chemistry measurements from the USDA FoodData Central, the US FDA Total Diet Study, and published peer-reviewed laboratory analyses. It represents the most comprehensive publicly available per-serving iodine reference for common foods.

This is the same dataset cited by US federal health agencies and referenced in clinical pregnancy nutrition guidelines. Values represent iodine per typical household serving as determined by food-grade analytical methods. Baby foods have been excluded. All remaining categories including dairy, eggs, seafood, iodised salt, bread, grains, vegetables, and meat are included, covering over 380 foods.

Iodine content in food is inherently variable: it depends on soil iodine content in the growing region, iodine supplementation in animal feed, and whether iodised salt was used during processing. Published values represent population averages and will differ from specific products and brands.

Iodine (mcg/serving)
Primary Reference
US Department of Agriculture, US Food and Drug Administration, and NIH Office of Dietary Supplements. USDA, FDA and ODS-NIH Database for the Iodine Content of Common Foods per Serving, Release 3. March 2023.

https://www.ars.usda.gov/ARSUSERFILES/80400535/DATA/IODINE/IODINE_DATABASE_RELEASE_3_PER_SERVING.PDF
Geographic variability in iodine content Iodine is the most geographically variable nutrient in food. The same food grown in iodine-rich soil can contain up to ten times more iodine than the equivalent food from iodine-depleted soil. Australian and New Zealand dairy typically differs from US measured averages due to differences in farming practices and iodophor use in dairy processing. Nourva's database uses US FDA/USDA measured values as the best available published international reference.

Local DHA Database

🐟
USDA National Nutrient Database for Standard Reference, Release 28
Standard Reference Release 28, 2016 · USDA FoodData Central

DHA (docosahexaenoic acid, 22:6 n-3) data is sourced from the USDA National Nutrient Database for Standard Reference, Release 28 (2016). This is a comprehensive laboratory-measured dataset of omega-3 fatty acid content in foods, compiled by the USDA Agricultural Research Service using standardised analytical chemistry methods.

The DHA dataset compiled into Nourva includes DHA values for over 530 foods with a DHA content of 10 mg or more per serving. Covered categories include fish, shellfish, fish oils, eggs, poultry, and select organ meats. Each entry records the exact gram weight of the reference serving, allowing Nourva to scale DHA proportionally when a different quantity is entered. Foods with DHA below 10 mg per serving are excluded as nutritionally insignificant for pregnancy tracking purposes. Baby foods and infant formula are excluded.

DHA / 22:6 n-3 (mg/serving)
Primary Reference
US Department of Agriculture, Agricultural Research Service. USDA National Nutrient Database for Standard Reference, Release 28: Nutrient 22:6 n-3 (DHA) Content of Selected Foods per Common Measure. 2016.

https://ods.od.nih.gov/pubs/usdandb/DHA-Content.pdf

USDA FoodData Central API (Iodine & DHA gap-fill)

🏛️
USDA FoodData Central API
US Department of Agriculture · fdc.nal.usda.gov

When a food is logged in exact weight units (grams, kilograms, ounces, or pounds) and iodine or DHA is not found in the built-in datasets described in Sources 2A and 2B, Nourva queries the USDA FoodData Central API specifically for iodine and DHA using a two-step approach:

  1. Step 1: Search. The food name is submitted to the FoodData Central search endpoint, requesting only nutrients 1100 (iodine) and 1272 (DHA). Foods that appear to be concentrated oils or extracts are filtered from results to avoid misleading values.
  2. Step 2: Portion lookup. The food detail endpoint is queried to retrieve foodPortions data, recording the actual gram weight per serving as measured by the USDA. This step eliminates the common error of assuming one serving equals 100 g.

USDA FoodData Central data is limited to SR Legacy and Foundation foods: standardised reference measurements rather than branded products. This source is not used for serving-based units (cup, piece, tablespoon) because the gram weight of such servings cannot be determined reliably without portion data.

Iodine (nutrient ID 1100) DHA 22:6 n-3 (nutrient ID 1272)

Google Gemini AI Estimation

Last resort only Gemini AI estimation is triggered only when the built-in datasets and the API fallbacks both return no result for iodine or DHA. This source is available to Premium subscribers. All AI-estimated values are clearly marked "AI est." in the app and should be treated as approximations.
🤖
Google Gemini 2.5 Flash
Google DeepMind · deepmind.google

When no database source has iodine or DHA for a logged food, Nourva submits a structured prompt to Google Gemini 2.5 Flash requesting an evidence-based estimate. The model is explicitly instructed to reference peer-reviewed data from USDA FoodData Central, the NIH Office of Dietary Supplements, the WHO, and the NHMRC, and is prohibited from guessing.

Exact prompt submitted to Gemini

The following prompt is transmitted verbatim for every AI estimation request. The food name, quantity, and unit are substituted at runtime:

Gemini request prompt
// Sent to: google/gemini-2.5-flash
// Model temperature: 0.1 (minimal creative variance)
// Max tokens: 100

"You are a clinical nutritionist. Use only
peer-reviewed nutritional data from USDA
FoodData Central, NIH Office of Dietary
Supplements, WHO, or NHMRC as your reference.
Do not guess.

For [QUANTITY] [UNIT] of "[FOOD_NAME]":

Return ONLY this JSON:
{"iodine_mcg": <number>, "dha_mg": <number>}

Rules:
- iodine_mcg: iodine in micrograms per
USDA/NIH data. Good sources: dairy, eggs,
seafood, iodized salt. Use 0 for most
plant foods, coffee, tea, spices.

- dha_mg: DHA omega-3 in milligrams per
USDA FoodData Central. ONLY found
meaningfully in: fatty fish (salmon,
sardines, mackerel, tuna), shellfish,
and algae/seaweed.
Use 0 for ALL other foods including
dairy, eggs, meat, grains, vegetables,
tea, coffee, fruit, spices.

- Use 0 (not null) when the food is not a
real source per nutritional data.
- Do not estimate. Only report values
consistent with published food databases."

Post-processing filters applied to AI output

Nourva applies minimum thresholds to discard values that are too small to be nutritionally meaningful, or that are likely to reflect model error:

NutrientMinimum thresholdRationale
DHA 20 mg per serving Below 20 mg is nutritionally negligible. The WHO recommends a minimum of 200 mg/day during pregnancy; 10 mg represents 5% of that target and is likely a trace amount or model error.
Iodine 3 mcg per serving Below 3 mcg is trace-level. The WHO pregnancy target is 220 mcg/day; values under 3 mcg do not constitute a meaningful contribution and are not displayed.

Known limitations of AI estimation

Large language models can produce plausible-sounding but incorrect nutritional values, a phenomenon commonly referred to as hallucination. Nourva mitigates this through several measures:

  • Using AI only when all measured sources have returned no result
  • Providing explicit rules about which foods contain DHA and iodine
  • Anchoring the model to named scientific databases (USDA, NIH, WHO, NHMRC)
  • Setting model temperature to 0.1 to minimise speculative responses
  • Applying hard minimum thresholds to the output before values are displayed
  • Clearly labelling all AI-derived values as "AI est." throughout the app

Despite these measures, AI-estimated values should be treated as approximate. Nourva does not warrant the accuracy of AI-generated nutritional estimates.

How we handle data conflicts and quality

Source priority: why measured data takes precedence over AI

Laboratory-measured data from the built-in USDA SR Legacy database, along with the dedicated iodine and DHA datasets, is always preferred over AI estimates. Measured values are obtained by standardised analytical chemistry methods and are reproducible across studies. AI estimates are derived from pattern-matching across text corpora and may not accurately reflect the specific food consumed. When a measured source returns a value, AI estimation is not invoked.

Food matching

Built-in dataset lookups use keyword matching with a minimum four-character word length to prevent false positives. If a keyword match would be ambiguous, the lookup returns no result and passes control to the next source in the waterfall.

Serving size and gram weights

For serving-based entries such as "1 serving" or "2 cups," Nourva uses USDA's foodPortions data to determine the actual gram weight of that serving for the specific food queried, rather than assuming a fixed 100 g default. This improves accuracy for foods where the standard serving differs substantially from 100 g.

Cooking method

Nourva does not currently adjust nutrient values for cooking method. DHA content is modestly lower in cooked fish than raw; folate and vitamin C are reduced by heat. For the most accurate tracking, log the cooked form of a food where possible.

Pregnancy nutrition targets used in Nourva

These are population-level reference values, not personal prescriptions The targets shown in Nourva are based on WHO, NHMRC, and NHS recommendations for pregnant women. Individual requirements vary. Always follow the advice of your obstetrician, midwife, or registered dietitian.
NutrientPregnancy targetGuideline source
DHA200 mg/day minimumWHO 2012; European Food Safety Authority
Iodine220 mcg/dayWHO/UNICEF/ICCIDD; NHMRC Australia
Iron27 mg/dayWHO; NHMRC Australia and New Zealand
Folate600 mcg DFE/dayWHO; NHMRC (inclusive of folic acid supplementation)
Calcium1,000 mg/dayWHO; NHMRC Australia and New Zealand
Protein+25 g above pre-pregnancy baselineNHMRC; NHS UK (varies by trimester)

Frequently asked questions

Why does the same food show different values in different apps?

Different apps use different underlying databases and different serving size assumptions. Some platforms draw substantially from user-submitted data, which can be highly inconsistent. Nourva prioritises laboratory-measured data from USDA FoodData Central, which provides standardised reference values but represents typical values for a food category rather than the exact product or brand purchased.

Additionally, nutrient content varies naturally between batches, growing regions, seasons, and cooking methods. No nutrition tracking application can produce an exact value; all results are approximations of varying quality.

Can I trust AI-estimated DHA and iodine values?

Treat them as an informed estimate rather than a precise measurement. Gemini is instructed to reference USDA and NIH data in its response, and Nourva applies minimum thresholds to discard implausible values. However, AI models can and do produce incorrect nutritional figures.

If DHA or iodine tracking is relevant to a clinical situation, we recommend cross-referencing against the USDA FoodData Central database directly at fdc.nal.usda.gov.

Why does DHA show no value for some foods?

DHA is present in meaningful quantities only in fatty fish (salmon, sardines, mackerel, tuna), shellfish, and algae-derived products. For foods such as vegetables, grains, dairy, tea, or coffee, DHA is either zero or below the 20 mg threshold that Nourva considers nutritionally significant for pregnancy tracking. Displaying a value of 1 to 2 mg for a food that is not a DHA source would be misleading.

When AI estimation would be required to produce a value and the user has not subscribed to Premium, DHA is shown as not available.

What data is transmitted to Google when AI estimation is used?

When AI estimation is triggered, Nourva transmits the following to the Google Gemini API:

  • The food name as entered (for example, "grilled salmon")
  • The quantity and unit (for example, "1 serving")
  • A structured prompt requesting iodine and DHA values only

No personal information, user identifiers, health data, or account details are transmitted to Google. Each request is stateless; Google does not retain conversation history from these API calls. Please refer to Google's Privacy Policy and Gemini API Terms of Service for details on data handling.

Does iodine content in dairy vary by country?

Yes, significantly. Iodine content in dairy products depends primarily on iodine levels in cattle feed and the use of iodophor disinfectants in dairy processing, both of which vary by country, producer, and season. Australian and New Zealand dairy may contain higher iodine than the US averages used in Nourva's database.

Nourva's local iodine database uses US FDA and USDA measured averages from the NIH ODS Release 3 (2023). If you are in Australia or New Zealand, actual iodine intake from dairy is likely to be higher than shown. The Australian NHMRC recommends a 150 mcg/day iodine supplement for all pregnant women precisely because dietary iodine sources are variable and difficult to track accurately.

How does Nourva validate that a food is real before fetching nutrition?

For foods entered manually rather than selected from search results, Nourva first submits a validation query to Gemini: "Is '[food name]' a real food or drink that humans eat? Reply with only YES or NO." This prevents nonsense entries from generating nutrition data. If Gemini is unavailable, validation is skipped and nutrition is fetched directly from the databases.

What happens when no internet connection is available?

Nourva's local iodine and DHA databases are embedded in the app binary and function without an internet connection. The USDA FoodData Central API, Spoonacular API, and Google Gemini all require an active connection. If network requests fail, the app displays whichever nutrients could be retrieved and shows no value for those that could not be fetched. A retry option is available, and values can be entered manually.

Important limitations

Not a medical device or clinical tool Nourva is a personal wellness application. It is not a registered medical device, has not been approved by the TGA (Australia) or the FDA (United States), and is not a substitute for professional medical, dietary, or clinical advice.

Nutritional data in Nourva is provided for general informational and personal tracking purposes only. The values shown are estimates based on standardised food databases and may not accurately reflect the nutritional content of the specific food consumed.

Nourva should not be used to:

If you have concerns about your nutrition during pregnancy, particularly regarding iron, iodine, folate, or DHA, consult a registered dietitian or your healthcare provider. Many pregnant women require supplements in addition to dietary intake; Nourva tracks food intake only and does not account for supplement use unless manually logged.

Third-party data sources including USDA FoodData Central, Spoonacular, and Google Gemini are provided by independent organisations. Nourva accepts no responsibility for the accuracy or availability of data from these services.

AI-generated nutritional estimates are produced by Google Gemini and are clearly labelled in the app. These estimates may contain errors. Nourva makes no warranty, express or implied, regarding the accuracy of AI-estimated values.

Corrections and feedback

If a nutritional value in Nourva appears significantly incorrect, we want to know. Errors in food database entries can affect many users. Please contact our support team with the food name, quantity, nutrient in question, the value displayed, and the value you believe is correct together with your source.

All data correction requests are reviewed, and our source databases are updated when systematic errors are identified.