How to choose a pill counting app

Eleven identical transparent capsules, spread on one tray under ordinary room light. We photographed that tray with two widely installed counting apps and with countRx, without moving anything between the shots. One app came back with 12. The other came back with 10. countRx came back with 11.

Be clear about what that is and is not. It is one tray, photographed once with each app, and it is not a benchmark. A single tray tells you nothing about how any of these apps behave across hundreds of photos, and clear capsules are a deliberately hard case: they are the thing most likely to split apps apart, which is why they were on the tray. What it does show is worth taking away anyway. Apps that all promise to count pills do not agree with each other, and it does not take a difficult scene to prove it.

None of that is visible from a store listing. Every listing claims to count accurately. Screenshots are chosen by the developer, and the count in them is always right. Star ratings mostly measure whether the app opened and whether it nagged. So if the number is what you actually care about, the choice has to rest on things you can check yourself.

Five things separate a counting app you can trust from one you are simply taking on faith. All five can be checked before you count a single pill.

Accuracy you can inspect, rather than a number in a marketing line. Almost every listing carries a percentage. Very few say what was measured, on how many photos, or whether those were photos the app had already been trained on. A figure with no method behind it is a claim rather than a measurement, and there is nothing to argue with because there is nothing to look at. What you want is a page that gives you the size of the test set, the date it was scored, and the cases where counting does worst. An app willing to publish its weak spots has usually looked for them.

Every pill marked on your own photo. This is the most useful thing a counting app can do and the easiest to check. If it returns a bare number, you have to take it on trust, and a total that is quietly wrong looks exactly like a total that is right. If it draws a marker on each pill it found, you can glance at the picture and catch the ordinary failures yourself: two markers sitting on one tablet, a marker on something that was never a pill, a pill cut off at the edge of the frame.

What happens to a broken half. Splitting tablets is ordinary at home, and it is where counting gets interesting, because half a tablet is still half a dose. An app that counts objects will tell you there are 15 things on the tray. An app that understands halves will tell you there are 13.5 doses. Those are different answers to the question you actually asked, which is how long the supply lasts.

A history that keeps the photo. A number you wrote on a sticky note is a number you cannot defend a week later. A saved count with its picture attached is something you can go back to without counting anything again. Check whether the history keeps the image or only the total, whether it is searchable, and whether you can get the data out at all as CSV or PDF.

What it costs after the first week. There are three common shapes and a store listing rarely makes it obvious which one you are in. Some apps are simply free. Some are free to install and then ask for a subscription at the moment you want to count something. Some sell counts as credits, so the app costs nothing but each photo spends one. None of those models is dishonest by itself, and the point is not to avoid any particular one. The point is to know which you picked before your medication is spread out on the table.

None of this needs a comparison table. Comparing pill counter apps turns into five questions you can ask about anything in the app store, this app included, and mostly answer before you install it.

If an app cannot answer the first one, that is information. If it cannot answer the last one in plain words, that is more.

Plainly, so it is not buried: countRx is ours. This page is on our site, so read what follows as our answers to our own checklist rather than as a verdict on anyone. The checklist above is written the way it is precisely because every answer below can be checked rather than taken on our word.

Our accuracy page gives a short summary of the benchmark. It was scored on 3,487 held-out photos containing 157,183 pills, none of which the model had been trained on, and it came to 99.26 percent net counting accuracy. The figure closer to what a single photo feels like is a different one, and day to day it matters more: about 93 in 100 photos came back with the count exactly right. The accuracy page explains the headline metric and offers practical photo tips.

On seeing what was counted, every pill the model found is marked on your own photo. Markers are tappable, so tap one to take that pill out of the total and tap it again to put it back. That is how you deal with a stray detection, or with a bottle cap that happened to be sitting on the tray.

On halves, a piece recognized as a broken half counts as 0.5 rather than 1, so a tray of wholes and halves reads in doses. Recognition is not a guarantee. A tablet with a small chip out of it can still come back as a whole one, which is exactly the sort of thing the markers let you catch.

On the record, every count is saved with its photo and a timestamp, the history is searchable, and it exports to CSV or PDF. On price, countRx is free. Not free for a trial period, not free up to a number of counts, not free until a plan appears later. There is no paid version and no account to create.

On what happens to your photo, counting is cloud-assisted and we would rather say so than imply otherwise. The picture goes to our server over an encrypted connection, a vision model finds the pills, the count comes back a few seconds later, and the copy on the server is deleted once it has been produced. It is not stored, not shared, and not used for anything else. It also means counting needs a connection, because the work does not happen on the phone.

You do not have to install a counting app to find out whether it copes with your pills. There is a version of the countRx counter that runs in this browser. Take or drop in one photo of a tray and you get the same count and the same markers back, with no account and nothing to download. It works on a laptop as well as on a phone.

Use the hardest tray you own for that test rather than the easiest. Clear capsules, small white tablets on a pale counter, a pile that includes a few halves, a tray big enough that you would not enjoy counting it by hand. An app that copes with your worst case is the one worth keeping, and a minute in a browser tells you more than any list of features will.

If it does what you need, the app adds the saved history, the barcode scan for the drug name, and the exports on top of it. If it does not, you have lost a minute and installed nothing.