Plateknow When we measured what people actually type looking for an app in this category, a striking share of them ended the query with the name of a forum — reddit, ptt, dcard. They are not looking for a comparison page. They are looking for somebody with no stake in the answer.
We have a stake in the answer, obviously — we build one of these. So this page does the only useful thing left: it hands you the questions, tells you what we do about each one, and then says where we lose. You can take the questions to any app in the store, including ours.
Why this page names nobody
Every app in this category publishes the same comparison page: a grid of nine competitors, the publisher’s own row a column of green ticks, a paragraph on what is wrong with each rival, and a pricing column that was true in the month it was written. You have seen it. It is the reason the word reddit is on the end of the query.
- A vendor’s scoreboard is not evidence. The company writing it picks the rows. Any app can win a table it designed, and the rows a product would lose on simply are not in it.
- Competitor pricing goes stale and becomes untrue. Plans change quietly. A table stating what another company charges is a claim about somebody else’s business that nobody is maintaining.
- Criticising named products by name is not research, it is marketing with footnotes. We have a rule against it, written down before this page existed.
So the category descriptions below are deliberate. What matters is not which logo is in the row; it is whether the app you are about to set up does the thing or does not.
Seven questions worth asking before you commit an evening
| What to check | What you will often find | What Plateknow does |
|---|---|---|
| Can you see how it read the plate? | A number, and no working. If it decided your chicken was tofu there is no way to tell. | The segmentation is shown. You watch the plate get cut apart, which is the only audit trail a photo estimate can give you. |
| How many taps to fix a wrong portion? | A correction lives behind an edit mode, two or three screens in. | The correction control is on every result card and always visible. It is the second thing you can touch, not the ninth. |
| Can you say "about two thirds of that"? | Whole servings. Two thirds of a bowl becomes one bowl or nothing. | Portions adjust in fractions, because two thirds of a bowl is the single most common real correction. |
| Can you use it with no goal at all? | Goal setup is mandatory onboarding — target weight, rate, deficit — before the app will show you anything. | A no-target mode that sets no calorie goal and simply records the day. Every number can also be switched off: the record stays, the counting goes. |
| What exactly is free — in numbers? | "Free to download", which is not a free tier and is not an answer. | The free version comes with an allowance: ten AI estimates a day, and no account at all — the app does not sign you in to anything. The allowance sits on the estimate, never on the diary: log every meal you eat, type in what it was once the allowance is spent, and the check-ins still come. Paying is for people who run past it — it raises the estimate allowance and opens the longer recaps. |
| Do you need an account before you see it work? | A signup wall before the first result, which optimises for the install chart rather than for you. | No account before you have seen it work on your own food. |
| What leaves your phone? | Rarely stated plainly. Usually the whole photograph goes to a server. | Your phone separates the food from everything around it first, and only that cut-out — the plate on a blank background — is sent for an estimate. The table, the room and the people at it never leave the device. |
Where we lose
A comparison page whose author never loses a row is not a comparison page. Here is the honest list, and none of it is coming off this page when it becomes inconvenient.
- There is no Android build yet. It is being written alongside the iOS one. If you are on Android today, an established tracker will serve you and we will not.
- The estimate needs a connection. Logging, check-ins and recaps all work offline; the AI estimate is the one part that does not. An app that ships its model on the device beats us on a plane.
- Packaged food is a barcode’s job, not a camera’s. A barcode is an exact match to a known product; a photograph is an educated guess. If most of what you eat comes out of a packet, the incumbent trackers and their barcode databases will beat us, and it is not a close call.
- A decade of user-contributed food entries is a real moat, and we do not have one. The oldest trackers in this category have databases built by millions of people over years. That is worth a great deal for searching food by name, which is the half of the problem we are not trying to win.
- We are new. Fewer reviews, a shorter track record, and no answer to "has it been around long enough" except that it has not.
The half nobody is competing on
Every app in this category now has roughly the same camera model, and the gap between them on the number is closing. So the interesting question stopped being whose estimate is best a while ago.
What almost nobody does is ask what happened afterwards. A few hours after a meal, Plateknow asks one question — energy, focus, mood, digestion, fullness — and the answer sits next to that meal in the record. Over a few weeks that is a different kind of thing from a calorie total: it is your own record of what showed up alongside what.
It reports co-occurrence and never why. It does not name a condition, find a cause, or tell you to cut a food, and any app that claims to do those from a photograph is claiming something it cannot do.
That is the target. The hard part is the other half — knowing what you actually ate.
Download freeWhich app is actually best?
It depends on what you eat and why you are logging, which is why this page is a checklist rather than a ranking. If most of your food is packaged, something barcode-first will beat a camera. If you eat food that came out of a kitchen, photo estimation is the only thing that scales. If counting itself is the part you dislike, look for the no-goal mode before anything else.
Why should I trust a comparison written by one of the apps?
Do not trust it — check it. Every row above is a question you can answer yourself in a few minutes on any app in the store, which is the whole reason it is written as questions. The section where we lose is the part worth judging us on: an app that will not tell you what it is bad at is telling you something anyway.
Is accuracy not the thing that matters most?
It matters less than how fast you can fix it. An app that is 15% off and corrects in one tap beats an app that is 8% off and buries the edit control, because you will still be using the first one in March. There is a fuller answer on the accuracy page, including what the published research does and does not support.
Do you compare yourself to specific apps anywhere?
No, and it is a written rule rather than a preference. Category descriptions only — no star ratings used as put-downs, no quoted negative reviews aimed at a named company. When we measured it, brand-versus-brand searching turned out to be about one percent of what people type in this category anyway.
Search demand from this repository’s own autocomplete corpus, harvested 31 August 2026 and committed as `research/SNAPSHOT-2026-08-31.jsonl`: of 3,218 suggestions across four markets, 270 en-us suggestions seek an app and 43 of those carry `reddit`; 139 zh-TW app suggestions carry 30 with `ptt` or `dcard`. Thirty-one suggestions in the whole corpus carry any brand name. Autocomplete shows phrasing and breadth, never volume, and no volume figure appears on this page. Claims about what this app does are checkable against the support and privacy pages.