I Logged Every Meal by Photo for 30 Days — Here's What Surprised Me

I am, historically, terrible at logging food. Not because I don’t care — I care a lot, in bursts, for about nine days at a time. Then a busy week hits, I skip one meal in the app, then three, then I’m back to vaguely promising myself I’ll “start Monday.” So this time I tried to remove the part that always breaks me: the typing. The plan was simple and a little ridiculous. For 30 days, I would log every single meal by photo. Point the camera at the plate, let the app figure it out, move on with my life.

What surprised me wasn’t whether it worked. It was the one thing a photo cannot see — and how much that one thing quietly changes the number at the bottom of the screen.

Week one: the honeymoon

The first few days felt like cheating, in the best way.

Breakfast was the easiest test case, and it nailed it instantly — oatmeal with banana and a scoop of peanut butter, photographed in about two seconds while the kettle was still going. Lunch the same. I’d plate a sandwich and some fruit, snap it, glance at the estimate, and that was the entire interaction. No searching a database for “whole wheat bread, store brand, 2 slices.” No scrolling past forty near-identical entries to find the one that was probably right.

By day four I was a convert. The friction that had killed every previous attempt — the open-app-search-scroll-pick-guess-the-portion loop — was just gone. I logged a coffee, a handful of almonds, leftover pasta, all without breaking stride. I remember thinking, smugly, that I’d cracked it.

I had not cracked it.

Week two: the drift I couldn’t explain

Here’s where it got interesting, and where my smugness took a hit.

About ten days in, the numbers stopped feeling right. Not wildly off — just low, consistently, in a way I couldn’t put my finger on. I was eating roughly what I always eat, my weight wasn’t doing what the math said it should, and yet every day’s log looked tidy and reasonable and under target.

So I started paying attention to which meals felt suspicious. And a pattern emerged. The clean, simple plates — grilled chicken, plain rice, a piece of fruit — those the camera handled beautifully. The trouble was everything that had been cooked.

I made a stir-fry one night, the kind I make all the time: vegetables, chicken, a bit of rice. Photographed it. The estimate came back looking almost virtuous. But I’d cooked that stir-fry in a generous glug of oil — easily a couple hundred calories of oil — and the photo had no idea. Oil disappears into the food. The camera sees glossy vegetables; it does not see the three tablespoons that made them glossy.

Once I noticed it, I saw it everywhere. The salad that looked like a diet plate was swimming in a dressing the photo read as “a bit of moisture.” The “just some vegetables” side had been roasted in butter. The pan sauce, the drizzle, the spoon of mayo holding the tuna together. All the calorie-dense stuff is invisible, because it’s either absorbed, hidden underneath, or transparent. A photo logs the visible meal. Real food is full of things you can’t see.

That was the genuine surprise of the month, and it reframed the whole experiment. Naive photo logging isn’t inaccurate because the AI is dumb about what’s in the picture. It’s inaccurate because the most fattening ingredients aren’t in the picture at all.

A confession: I wasn’t actually photographing everything

I should be honest about something, because “I logged every meal by photo” was the goal, not a religion.

There were meals the camera was simply the wrong tool for. A protein bar grabbed between meetings — I scanned the barcode, because the barcode is the answer, no estimating required. A late-night bowl of cereal where I knew the exact box and the exact splash of milk — I just typed it, faster than fishing my phone into camera position over a bowl. And a packaged microwave meal I scanned rather than photographed a tray through its film.

So the real-world version of “log every meal by photo” turned out to be photo for most meals, barcode when there’s a barcode, manual when I already know the answer. I’d resisted apps that pushed photo-only because that’s not how eating works — sometimes the fastest, most accurate path is a scan or three taps. The point was low effort, not photo purism.

But the core problem remained for all the cooked, mixed, restaurant, homemade meals that make up most of how I actually eat. And those are exactly the meals where the hidden oil-butter-dressing problem lives. I needed something that could look at a plate of real food and reason about what was probably in it, not just what it could see.

Week three: the app that asked the right question

I’d been bouncing between a couple of photo apps in the first two weeks, and most of them did the same thing: take the photo, spit out a confident number, done. Confident and wrong is the worst combination, because you trust it.

The one I ended up settling on — PlateLens — behaved differently in a way that took me a day or two to even register, and then I couldn’t unsee it.

When I photographed that stir-fry, it didn’t just label the vegetables and chicken. It recognized the dish — that this was a stir-fry, a category of food that is almost always cooked in oil — and instead of silently guessing, it asked me. A little prompt: cooked in oil? roughly how much? The salad got how much dressing — light, medium, heavy? The pasta asked whether there was butter or oil in it. It was inferring the hidden, calorie-heavy ingredients from what the dish actually is, and then confirming with me whenever it wasn’t sure instead of pretending it knew.

That confirm-on-doubt loop was the turning point of the entire month. Suddenly the invisible calories weren’t invisible anymore — the app knew to look for them based on the type of food, and a one-tap answer from me filled in the part the camera physically couldn’t. My logs stopped reading suspiciously low. They started matching reality, and my weight trend finally lined up with the numbers.

It’s a subtle bit of intelligence and it’s easy to undersell, so let me put it plainly: the difference between a photo app and a good photo app is whether it understands that the dish implies ingredients the lens can’t see — and whether it has the humility to ask rather than guess.

Week four: trust, and the honest caveats

By the last week, logging had become the thing I’d always wanted it to be — boring, in a good way. Most meals: snap, confirm the one thing it asked about, done. The annoying meals: scan or type. I wasn’t dreading it, I wasn’t drifting, and crucially I believed the number.

I’m not going to pretend it’s flawless. A chaotic restaurant plate — a mixed curry, a loaded burrito — is still a harder guess than a tidy home plate, and on those I sometimes nudge the portion myself. The free tier capped how many photo scans I got per day, so on heavy days I leaned on the barcode and manual options more (both unlimited), and eventually decided the unlimited scans were worth paying for. And yes, it asks you to confirm things, which is a half-second of friction — but that half-second is exactly the feature, so I can’t really complain about it.

None of that came close to outweighing the upside. For the first time in years I logged for 30 straight days without quitting.

The verdict

I went in thinking the surprise would be can a photo really count calories? The real surprise was that a photo can’t — not on its own — because the food that actually moves the number is the oil, butter, dressing, and sauce that the camera will never see. Naive photo logging drifts for exactly that reason.

What fixed it wasn’t a better camera. It was an app that reasons about what the dish is, infers the hidden ingredients, and asks me to confirm when it’s unsure — plus the freedom to scan a barcode or type when that’s simply faster. That combination is what made photo logging trustworthy rather than just convenient.

If you want to try the thing that actually made it work, it’s PlateLens — you can grab it on the App Store or Google Play and start on the free tier. My slightly ridiculous month taught me one thing above all: the best photo calorie app isn’t the one that gives you a number fastest. It’s the one that knows what it can’t see — and bothers to ask.