What AI Calorie Counters Miss: Oils, Sauces and Drinks

Why do AI calorie counters miss oils, sauces and drinks?

An AI calorie counter estimates what it can detect. Oil soaked into vegetables, butter melted into mashed potatoes, dressing tossed through a salad and sugar stirred into coffee are hard or impossible to see. A review of image-assisted dietary assessment by Höchsmann and Martin notes that hidden ingredients, such as butter in mashed potatoes, look-alike items, such as diet versus regular soda, and preparation methods, such as fried versus baked, often require the person eating to add information.

A 2024 scoping review by Chotwanvirat and colleagues makes the same point with fried rice: the rice and egg are visible, but seasonings and cooking oil have to be estimated. Even trained nutritionists' estimates of those amounts differ, which is why the authors note that research often uses two or three analysts to reduce individual bias. The direction of the error is not always the same, either. In a 2026 study of hospital meal photos by Isobe and colleagues, all ten AI models overestimated fat by more than 20% on average, while energy and carbohydrate estimates from the best models were much closer. Fat is simply hard to judge by eye.

Voice and text logging have a different version of the same problem: the AI can only count what you mention, and few people remember to say "and a tablespoon of oil in the pan."

Why do small amounts add up so fast?

Because fat and alcohol are energy-dense. According to the FAO's report on food energy conversion factors, the general Atwater system counts about 9 kcal per gram of fat and 7 kcal per gram of alcohol, compared with about 4 kcal per gram for protein and carbohydrate. Cooking oil is essentially pure fat, so a splash that barely shows in a photo can carry more energy than the vegetables it coats.

How many calories are in a tablespoon of oil, butter or sauce?

The values below come from the SR Legacy data in USDA FoodData Central, using its listed gram weight for one tablespoon (about 15 ml) of each food. Brands vary, so check the label for anything packaged.

Food (USDA description)	1 tbsp weight	kcal per tbsp	kcal per 100 g
Oil, canola	14 g	124	884
Oil, olive	13.5 g	119	884
Butter, salted	14.2 g	102	717
Mayonnaise, regular	13.8 g	94	680
Caesar dressing, regular	14.7 g	80	542
Ranch dressing, regular	15 g	64	430
Honey	21 g	64	304
Heavy whipping cream	15 g	51	340
Hummus, commercial	15 g	36	237
Italian dressing, regular	14.7 g	35	240
Barbecue sauce	17 g	29	172
Ketchup	17 g	17	101
Soy sauce (shoyu)	16 g	8	53

Two more worth knowing: FoodData Central lists 2 tablespoons (32 g) of smooth peanut butter at 191 kcal, and 1 teaspoon (4.2 g) of granulated sugar at 16 kcal. Tablespoons in real life are rarely level, so a generous "spoonful" of peanut butter or a free pour of oil can be well above these figures.

How many calories are in common drinks?

Drinks are easy to forget entirely, and a camera cannot tell regular cola from diet or know how strong a glass of wine is. These values are also from USDA FoodData Central:

Drink (USDA description)	Serving	kcal
Cola, regular	About 355 ml (12 fl oz can)	155
Beer, regular	About 355 ml (12 fl oz can)	153
Whole milk (3.25%)	About 240 ml (1 cup)	149
Red table wine	About 148 ml (5 fl oz)	125
Orange juice (chilled, from concentrate)	About 240 ml (1 cup)	122
Spirits, 80 proof (gin, rum, vodka, whiskey)	About 44 ml (1.5 fl oz)	97

UK servings are often bigger. The NHS page on calories in alcohol says a standard 175 ml glass of 12% wine can contain up to 158 kcal, a pint of 5% beer up to 222 kcal, and a double measure (50 ml) of 40% spirits about 100 kcal. The same page reminds readers that the low-risk guideline is not to regularly drink more than 14 units of alcohol a week.

How much can hidden calories add to a day?

Here is a simple worked example using the figures above. Suppose your AI log captures your meals correctly but misses four extras:

• 1 tablespoon of olive oil in the pan at dinner: 119 kcal
• 2 tablespoons of ranch dressing on a lunch salad: 128 kcal
• 1 tablespoon of mayonnaise in a sandwich: 94 kcal
• One 175 ml glass of 12% wine: up to 158 kcal

That adds up to as much as 499 kcal that never reached your log. Not all pan oil ends up on your plate, so the real figure for cooking fat may be lower, but the scale is clear. Missed extras like these are one of the common reasons behind a stall, which our guide on why a calorie deficit might not be working covers in more depth.

How do you log hidden calories with an AI app?

1. Tell the app how the food was cooked. Add "fried in 1 tbsp olive oil" or "with butter" to a photo note or voice entry. In a 2025 study by Rodríguez-Jiménez and colleagues, average energy error from meal photos fell from 30.5% with the image alone to 13.9% when a detailed ingredient list was added.
2. Log fats, sauces and dressings as separate items so they are visible and easy to adjust.
3. Measure oil with a spoon for a week, or weigh the bottle before and after cooking. If the bottle is 20 g lighter, that is about 177 kcal of oil (20 g at 8.84 kcal per gram), shared across however many portions you served.
4. Log every drink with its size and type. Say "330 ml can of regular cola" or "large glass of white wine," not just "drink."
5. Use the label for bottled sauces and dressings. Our guide on reading nutrition labels explains serving sizes and per-100 g values.
6. Add a buffer for restaurant food. You cannot see how much fat a restaurant kitchen used, so add a line for cooking oil or butter if your log looks light.
7. Check the AI's breakdown. If a stir-fry or roast vegetables show almost no fat, something is missing.

Common mistakes with hidden calories

• Logging the salad, not the dressing. The leaves are rarely the main source of energy.
• Assuming "a drizzle" is small. Free-poured oil is hard to estimate by eye.
• Forgetting milk and sugar in coffee and tea, which repeat several times a day.
• Letting the app guess regular versus diet. A photo cannot tell; specify it.
• Skipping weekend drinks. Alcohol often goes unlogged, and the NHS figures above show how quickly it adds up.

For how photo, voice and barcode inputs work in general, see how AI food tracking works. MyMacro AI, which is not yet released, is designed so you can say a meal in a normal sentence, including the oil and the drink, and have it become a structured entry you can edit. You can join the waitlist to hear when it launches.

Frequently asked questions

Can AI calorie counters see cooking oil?

Usually not reliably. Oil and butter are often absorbed or spread thinly, so research reviews describe them as hidden ingredients that need to be added by the user.

How many calories are in a tablespoon of olive oil?

USDA FoodData Central lists 1 tablespoon (13.5 g) of olive oil at 119 kcal. Canola oil is 124 kcal per 14 g tablespoon.

How many calories are in a glass of wine?

The NHS says a standard 175 ml glass of 12% wine can contain up to 158 kcal. USDA data lists about 148 ml (5 fl oz) of red table wine at 125 kcal.

Should I log cooking oil if most of it stays in the pan?

Yes, log at least the share you think ended up in your portion. Weighing the bottle before and after cooking gives you a better estimate than guessing.

How do I make an AI calorie counter more accurate for restaurant meals?

Describe the cooking method, log sauces and drinks separately, and consider adding a line for cooking fat, since added fats are hard for any app to see.

Sources

1. FoodData Central (SR Legacy), U.S. Department of Agriculture, Agricultural Research Service (2018)
2. Calories in alcohol, NHS (2023)
3. Food energy: methods of analysis and conversion factors, Chapter 3: Calculation of the energy content of foods, Food and Agriculture Organization of the United Nations (2003)
4. Review of the validity and feasibility of image-assisted methods for dietary assessment, International Journal of Obesity (Höchsmann and Martin) (2020)
5. Advancements in Using AI for Dietary Assessment Based on Food Images: Scoping Review, Journal of Medical Internet Research (Chotwanvirat et al.) (2024)
6. Accuracy of AI-Based Nutrient Estimation from Standardized Hospital Meal Images: A Comparison with Registered Dietitians, Nutrients (Isobe et al.) (2026)
7. Image-Based Dietary Energy and Macronutrients Estimation with ChatGPT-5: Cross-Source Evaluation Across Escalating Context Scenarios, Nutrients (Rodríguez-Jiménez et al.) (2025)

Not medical advice. This article is general information for healthy adults and is not a substitute for advice from a doctor or registered dietitian. Talk to a qualified professional before changing your diet, exercise or supplements, especially if you are pregnant, under 18, have a medical condition such as diabetes or kidney disease, take medication, or have a history of disordered eating. See our health disclaimer.