Yes, you can track calories with ChatGPT: it can estimate calories from a photo or a description of your meal, and add up a day’s food if you keep it all in one chat. But it is not a food diary. Its estimates can change when you ask again, its memory does not keep every detail, it only knows your target if you tell it, and in published tests its photo estimates were off by roughly a third on average.
That does not make it useless. It makes it a helpful assistant rather than a reliable log. Here is what the research shows and how to use it sensibly.
Can ChatGPT count calories from a photo?
It can estimate them, and it is good at the first step: working out what is on the plate. In a 2025 study using 114 photos of weighed meals, ChatGPT-4 identified the foods with 93% precision (O’Hara et al., 2025).
The harder step is how much food there is. In the same study it underestimated the weight of the meal in 76.3% of photos, and agreement with the real values was poor for medium and large portions. A separate 2025 study found ChatGPT-4o’s calorie estimates from standard photos were off by 35.8% on average, with underestimation growing as portions got bigger (Fridolfsson et al., 2025).
Newer models do better when you give them more to work with. In a study of ChatGPT-5 on 195 dishes, the average calorie error was:
| What ChatGPT-5 was given | Average calorie error per dish |
|---|---|
| Photo only | 123 kcal (30.5%) |
| Photo plus details such as type and amount of fat | 92 kcal (24.4%) |
| Photo plus full ingredient list with quantities | 53 kcal (13.9%) |
| Ingredient list with quantities, no photo | 67 kcal (18.1%) |
Source: Rodriguez-Jimenez et al., 2025.
The lesson is simple: a photo on its own is a rough guess, and a photo with a good description is much better.
How accurate is ChatGPT when you type what you ate?
A 2023 study in JAMA Network Open asked ChatGPT for the calories and macronutrients of 222 foods, in English and Chinese, five times each, and compared the answers with a nutritionist’s values (Hoang et al., 2023):
- Accuracy: calorie values were within 10% of the nutritionist’s for about 39 to 47% of items, depending on the version and language.
- Consistency: asking the same question again gave a slightly different answer, with an average variation of around 5 to 6% for calories.
- Protein: GPT-4 tended to overestimate protein.
These were 2023 versions, and newer models may do better. But the pattern still matters for tracking: if you ask “how many calories in a plate of rajma chawal?” twice, you may get two different numbers. In a dedicated tracker, a food picked from its database gets the same value every time, which keeps your daily totals comparable.
O’Hara et al. also point out that a general chatbot does not look up a food composition database the way a nutrition app does; it predicts a likely answer.
What can ChatGPT not do that a calorie tracker does?
Keep a reliable log. ChatGPT has a Memory feature that can use details from past chats, but OpenAI’s own Memory FAQ says that “Memory does not retain every detail from every conversation.” Your food history lives inside chats rather than in a structured diary sorted by date.
Track targets automatically. It does not know your calorie or protein goal unless you tell it, or it happens to remember. It will not warn you that you are 400 kcal over unless you ask.
Show trends. Weekly averages, weight graphs and macro charts only appear if you paste in your data and ask for them each time.
Stay consistent. As the Hoang study showed, the same food can get a different estimate on a different day.
What it does well is flexibility. You can describe a meal in your own words, ask why your weight is stuck, get high-protein vegetarian ideas, or ask it to explain a food label. For many people, that is the most useful part.
How do I track calories with ChatGPT if I want to try?
If you prefer a chatbot, these steps make it far more reliable:
- Set your numbers first. Work out your maintenance and target with the TDEE calculator and calorie deficit calculator, then tell ChatGPT your daily calorie and protein target.
- Use one chat per day, and ask it to keep a running table with a total after every meal.
- Describe portions and cooking fat. Grams are best; otherwise use standard sizes such as a 200 ml katori, and always mention the oil or ghee.
- Check the maths. Glance at the totals; adding up is where chat logs can quietly go wrong.
- Save the daily total yourself in a note or spreadsheet. That becomes your real log.
A prompt that works well:
My target is 1,600 kcal and 90 g protein a day. Keep a table for today
with columns: meal, food, portion, kcal, protein. After each entry,
show the running total and what is left.
Breakfast: 2 moong dal chillas (made from about 60 g raw dal) cooked
with 1 tsp oil, 1 katori (200 ml) curd.Our guide on how to count calories in Indian food explains katori sizes, the oil problem and recipe maths, which all make chatbot estimates better too.
ChatGPT or a dedicated calorie tracker: which should I use?
It depends on what you need. A fair comparison:
| General chatbot (such as ChatGPT) | Dedicated calorie tracker | |
|---|---|---|
| Calorie estimate from a photo | Yes | Yes, in AI photo trackers |
| Estimate from a typed description | Yes | Yes, in apps with text logging |
| Daily food log saved by date | No, it lives in chats | Yes |
| Calorie and protein targets | Only if you tell it | Built in |
| Same value for the same food each time | Not guaranteed | Yes for database foods; AI estimates can vary |
| Weight and macro trends | Only if you paste in data | Built in |
| Open-ended nutrition questions | Strong | Varies by app |
A chatbot suits people who log occasionally, like to ask questions, and are happy to keep their own records. A tracker suits people who want a daily log, targets and progress charts without extra effort. Calo AI, for example, combines photo, voice and text logging with calorie and macro targets, analytics and an AI nutrition coach, free with ads. You can try a photo estimate on the photo calorie calculator.
Whichever you choose, both give estimates. For what the research says about accuracy across apps, see are AI calorie trackers accurate?
Safety note: the authors of the JAMA study concluded that AI chatbots cannot replace nutritionists. If you are pregnant, under 18, or have diabetes, PCOS, kidney disease or a history of an eating disorder, get your targets from a doctor or registered dietitian, and do not let any app or chatbot set you below 1,200 kcal a day.
Practical takeaway
- ChatGPT is a good estimator and explainer, but not a reliable food diary.
- Photo alone is a rough guess. Add portions, ingredients and cooking fat for much better numbers.
- Expect some variation if you ask about the same food twice.
- Keep your own record of daily totals, whether in a note, a spreadsheet or a tracking app.
- Use the tool that fits your habits. The best tracker is the one you will use every day.