Can ChatGPT estimate body fat from a photo?
Asking an AI for your body fat has become normal. Here’s what the estimate is good for, how far off it can be, and how to get a number you can track.
Short answer
Yes, roughly. In an informal test with reference photos, GPT-4o’s median error was about 2.4 points for men and 5.7 for women. In practice estimates swing with lighting, pump, pose and clothing, and different AIs often disagree: in our analysis of Reddit posts, people quoting two AIs saw gaps of 2 to 8 points. Use an AI estimate as a first guess, then track change with a consistent method such as the US Navy tape formula.
What the evidence says
- Informal test (2025): GPT-4o estimating body fat from published reference photos: median absolute error about 2.4 points for men and 5.7 for women, with underestimates at higher body fat.
- Purpose-built smartphone methods: research systems trained against DXA report errors of roughly 3–4.5 points in controlled conditions.
- Real life: in our Reddit analysis, 22 posts quoted an AI body fat estimate and still asked strangers to check it. Reported pairs included Google AI 14% vs ChatGPT 18–20%, and Gemini 12–13% vs ChatGPT 16–17%.
Why AI estimates swing
- Overhead light carves abs; front light flattens them.
- A pump, a flex or holding your breath can take several points off visually.
- Clothing, camera height and distance change proportions.
- The model has no idea how much muscle you have under the fat, which is what really changes the look.
How to get a better estimate
- Use the same spot, light and distance every time, relaxed, morning, no pump.
- Give the AI your height, weight and waist measurement as well as the photo, and ask for a range.
- Cross-check with a tape-based estimate from the body fat calculator.
- Compare against the body fat chart.
- Most important: track the change. Same method, every two to four weeks.
What about AI body fat apps?
Apps such as GainFrame, trackBod and bodyfatAI wrap the same idea in a tracking app. They’re convenient but share the same sensitivity to light and pose. Compare them in the best AI body fat apps. ProgressLens is built around the consistency problem: a ghost overlay makes every photo match the last one, its AI combines the photo read with your tape measurements and history, and the same photos always give the same result. For a free first estimate, use the body fat estimator, or see ProgressLens vs ChatGPT.
See also: The ChatGPT body fat prompt · Is there an app that measures body fat? · What is my body fat percentage?
FAQ
Frequently asked questions
Is ChatGPT accurate at guessing body fat?
Roughly, within a few points on average, but individual estimates can be much further off and change between photos.
Why do ChatGPT and Gemini give different body fat numbers?
They weigh visual cues differently and have no reliable way to judge the muscle under the fat. Lighting and pose also change their reading.
What is the most accurate way to measure body fat?
A DEXA scan is the most accurate common option, typically within about 1–2 points. For home tracking, consistency matters more than accuracy.
Sources
- Riedl A (2025). Can ChatGPT accurately estimate body fat percentage? annaleptikon.substack.com.
- ProgressLens, What lifters ask about body fat (2026).