Skip to content

AI does three jobs. You review each one.

AI matches each ingredient to a USDA food, estimates a weight only when no USDA portion fits, and in Studio suggests which ingredients add sugars. The compliance engine calculates.

Start freeRead the guide for AI assistants
A sample recipe pasted into ChatGPT: each line matched and weighed, the added sugars confirmed, the label exported

Your assistant asks. BetterMenu calculates.

We asked ChatGPT for a label for the sample granola in fresh chats. Two chats had BetterMenu not connected. Two had the BetterMenu app connected and used the same saved recipe.

ChatGPT on its own Same prompt, two fresh chats

Chat 1

The granola prompt and the top of ChatGPT’s answer in the first chat: an estimated Nutrition Facts label, about 7 servings per recipe, serving size 60 g, Calories 250, Total Fat 9 g.
See the full answer in Chat 1

Chat 2

The same prompt and the top of ChatGPT’s answer in the second chat: an estimated Nutrition Facts label, about 8 servings per recipe, serving size 60 g, 260 Calories, Total Fat 11 g.
See the full answer in Chat 2
  • Chat 1: about 7 servings, 9 g fat. Chat 2: about 8 servings, 11 g fat.
  • No source for any number
  • Called its own numbers estimates

ChatGPT, logged out, same prompt in fresh chats, Nov 10, 2026. Across six chats it gave 240, 250 and 260 calories. Results can vary by model and plan.

With BetterMenu connected Same saved recipe, two fresh chats

Chat A: the recipe pasted and saved

Part of the Nutrition Facts label BetterMenu made in the chat where the granola recipe was pasted and saved: Calories 240, Total Fat 9g 11%, Saturated Fat 1g 4%, Trans Fat 0g, Cholesterol 0mg 0%, Sodium 170mg 7%, Total Carbohydrate 36g 13%, Dietary Fiber 5g 18%, Total Sugars 11g, Includes 10g Added Sugars 20%.
See the full answer in Chat A: the recipe pasted and saved

Chat B: a new chat asks for the saved recipe

Part of the Nutrition Facts label for the saved recipe in a new chat: Calories 240, Total Fat 9g 11%, Saturated Fat 1g 4%, Trans Fat 0g, Cholesterol 0mg 0%, Sodium 170mg 7%, Total Carbohydrate 36g 13%, Dietary Fiber 5g 18%, Total Sugars 11g, Includes 10g Added Sugars 20%.
See the full answer in Chat B: a new chat asks for the saved recipe
  • Both chats: 240 calories, 9 g fat, 170 mg sodium
  • Each value traced to a USDA food and an FDA rule
  • Every version kept, and the label as a PDF, PNG or JPEG file

ChatGPT with the BetterMenu app connected, same saved recipe in two fresh chats, Nov 10, 2026. Chat B asked: “Show the Nutrition Facts label for my saved BetterMenu recipe Sample Granola. Don’t change the recipe.”

What the AI does

Three jobs. Each is a suggestion you review, change or confirm, never a calculation.

  1. Matches each ingredient. Finds the USDA food each recipe line means.
  2. Estimates a weight when it must. Only when no USDA portion fits a unit, like "1 handful", and marked for you to change.
  3. Suggests added sugars. In Studio, flags which ingredients add sugars, for the Added Sugars line.
What the AI does: matches each ingredient, estimates a weight when it must, suggests added sugars. You confirm each one, and the compliance engine calculates.Three panels, one per AI job, each ending at your confirm. 1. Matches each ingredient: for the line 1/2 cup cacao nibs, AI picks the USDA food Cacao nibs over Cocoa powder, unsweetened. 2. Estimates a weight when it must: 1 handful raisins has no USDA portion for its unit, so AI estimates about 40 g, marked with ≈ as an estimate you can change. 3. Suggests added sugars, in Studio: of honey, sliced almonds and rolled oats, AI suggests honey adds sugars and the others do not. Then a heavy line, AI stops here: the compliance engine, fixed FDA rules from 21 CFR 101.9, calculates every number on the label.Matches each ingredientRecipe line1/2 cup cacao nibsCacao nibsAI pickCocoa powder, unsweetenedYou confirmEstimates a weight when it mustRecipe line1 handful raisins≈ AI estimateno USDA portionRaisins≈ 40 gYou confirmor change itSuggests added sugarsIngredientAdds sugars?HoneyAI suggests: adds sugarsYesSliced almondsNoRolled oatsNoYou confirmAI stops hereCompliance engineFixed FDA rules, 21 CFR 101.9Calculates every number on the labelWhat the AI does: matches each ingredient, estimates a weight when it must, suggests added sugars. You confirm each one, and the compliance engine calculates.Three panels, one per AI job, each ending at your confirm. 1. Matches each ingredient: for the line 1/2 cup cacao nibs, AI picks the USDA food Cacao nibs over Cocoa powder, unsweetened. 2. Estimates a weight when it must: 1 handful raisins has no USDA portion for its unit, so AI estimates about 40 g, marked with ≈ as an estimate you can change. 3. Suggests added sugars, in Studio: of honey, sliced almonds and rolled oats, AI suggests honey adds sugars and the others do not. Then a heavy line, AI stops here: the compliance engine, fixed FDA rules from 21 CFR 101.9, calculates every number on the label.Matches each ingredientRecipe line1/2 cup cacao nibsCacao nibsAI pickCocoa powder, unsweetenedYou confirmEstimates a weight when it mustRecipe line1 handful raisins≈ AI estimateno USDA portionRaisins≈ 40 gYou confirmor change itSuggests added sugarsIngredientAdds sugars?HoneyAI suggests: adds sugarsYesSliced almondsNoRolled oatsNoYou confirmAI stops hereCompliance engineFixed FDA rules, 21 CFR 101.9Calculates every number on the label

Where does the AI stop?

Once you have reviewed and confirmed it. From there the compliance engine calculates every number, with no AI in it.

Two recipe lines, four steps: AI suggests each match, a USDA portion or an AI estimate gives the weight, you confirm, the compliance engine calculatesTwo recipe lines go through four steps. 1. Match: AI suggests a USDA food for each line: Cacao nibs for 6 Tbsp cacao nibs, Raisins for 1 handful raisins. 2. Weigh: 6 Tbsp cacao nibs has a USDA portion, 3 Tbsp = 30 g, so 2 × 30 g = 60 g, with no AI. 1 handful raisins has no USDA portion for its unit, so AI estimates the weight, about 40 g, marked as an estimate you can change. 3. Confirm: you check each line. Then a heavy line marked "AI stops here". 4. Calculate: the compliance engine, with no AI, takes both lines and rounds by 21 CFR 101.9(c): 80.3 calories per serving prints as 80, and it fills the Nutrition Facts panel (per 1/6 of the batch: Total Fat 4.5g at 6%, and Protein 2g).AI suggests, you confirm1Match2Weigh3ConfirmRecipe line6 Tbsp cacao nibsUSDA foodCacao nibsAI pickUSDA portion: 3 Tbsp = 30 g2 × 30 g= 60 gYouCacao nibs60 gRecipe line1 handful raisinsUSDA foodRaisinsAI pick≈ AI estimateno USDA portionYou can edit it≈ 40 gYouRaisins40 gAI stops here4CalculateCompliance engine, no AICalories per serving101.9(c)80.3→ 80nearest 10Nutrition FactsPer 1/6 batch (17g)Calories80Total Fat 4.5g6%Protein 2gTwo recipe lines, four steps: AI suggests each match, a USDA portion or an AI estimate gives the weight, you confirm, the compliance engine calculatesTwo recipe lines go through four steps. 1. Match: AI suggests a USDA food for each line: Cacao nibs for 6 Tbsp cacao nibs, Raisins for 1 handful raisins. 2. Weigh: 6 Tbsp cacao nibs has a USDA portion, 3 Tbsp = 30 g, so 2 × 30 g = 60 g, with no AI. 1 handful raisins has no USDA portion for its unit, so AI estimates the weight, about 40 g, marked as an estimate you can change. 3. Confirm: you check each line. Then a heavy line marked "AI stops here". 4. Calculate: the compliance engine, with no AI, takes both lines and rounds by 21 CFR 101.9(c): 80.3 calories per serving prints as 80, and it fills the Nutrition Facts panel (per 1/6 of the batch: Total Fat 4.5g at 6%, and Protein 2g).AI suggests, you confirm1Match2Weigh3ConfirmRecipe lines6 Tbsp cacao nibs1 handful raisinsCacao nibsAI pickUSDA portion: 3 Tbsp = 30 gCacao nibs: 2 × 30 g= 60 gCacao nibs60 gRaisinsAI pick≈ AI estimateRaisins: no USDA portion≈ 40 gRaisins40 gAI stops here4CalculateCompliance engine, no AICalories per serving80.3→ 80Over 50: nearest 10101.9(c)Nutrition FactsPer 1/6 batch (17g)Calories80Total Fat 4.5g6%Protein 2g

Can I make a label from ChatGPT or Claude?

Yes. The ChatGPT app and Claude connector build the recipe and export the label from a chat. Agents never count as a seat.

Read the guide for AI assistants
Things you can ask, each run against your real recipes
You ask
“Make a label for my granola.” Builds the recipe, sets a serving, exports the label.
“What changed since version 3?” Compares two versions, ingredient by ingredient.
“Set the serving to 1/2 cup.” Updates the serving and recalculates the panel.
“Which ingredients add sugars?” Opens the added sugars review for you to review and confirm.
“Export the label as a PNG.” Returns a download link to the file.
“Go back to last Tuesday’s version.” Restores that version. The others stay.

Questions

More questions

ChatGPT and Claude can already give me nutrition. Why use BetterMenu?

On their own, they estimate nutrition from what the model learned, with no source for any number, and the label can come out different from one chat to the next. With BetterMenu connected, every nutrient value comes from USDA data or your own ingredients and fixed FDA rules: same saved recipe, same label. You also get the FDA serving size, rounding and %DV, every version kept, and the label as a PDF, PNG or JPEG file. When a retailer, co-packer or inspector asks how a number was made, you can show the recipe version, the source behind each ingredient and when the label was exported. You keep working in ChatGPT or Claude.

How does the assistant connect to my account?

Through MCP, signed in with OAuth. The assistant acts as you, with your access and nothing more.

Do BetterMenu’s AI providers keep or train on my recipes?

No. We send recipe lines only to AI providers that do not train on them and keep none of them after answering (Privacy Policy 3.5). In ChatGPT or Claude, that assistant’s provider handles your chat under its own terms and your settings there.

Read next

Make your first label today.

No card needed. See pricing.

Start freeSee pricing