language learning apps better alternatives

A wheel that needs truing, and a mechanic who speaks better English than your Dutch

The shop is on a side street off the Oudegracht, it is the size of a garage, and the man running it has a spoke key in one hand and a queue of four behind you. You have eleven weeks of Dutch on a phone. You have rehearsed this. You say the sentence about the back wheel.

He answers in English. Not rudely — helpfully, and faster than you could have managed. The exchange takes ninety seconds instead of five minutes, the wheel gets booked in for Thursday, and you walk out having learned precisely nothing about whether your Dutch works.

This is the situation the whole northern half of Europe puts learners in, and it is why so many people arrive at the alternatives question with bad evidence. Your app did not fail in that shop. It was never given the chance to be tested. What failed was the thing the app told you about yourself — a number, a level, a streak — none of which predicted what would happen when you opened your mouth in a room with a queue in it.

So the question people ask, which is which app is better, is almost never the question they need answered.

"Better" is a fit question wearing a quality costume

Language apps are not distributed along a single line from worse to better. They are built around different central loops, and a loop that suits you in March can be actively wasting your evenings by September. Nothing about the product changed. You did.

Three things move underneath you. Your level moves, which is obvious. Your purpose moves, which is less obvious — the person learning Dutch for a two-week holiday and the person learning it because their partner's family speaks it at Sunday lunch need almost opposite things. And your available time moves, which is the one that actually decides outcomes.

The practical consequence. An alternative is only better if it is better at the specific thing that is currently going wrong. Which means the first job is not shopping. It is naming the failure.

We would rather you did that honestly and stayed where you are than switched on the strength of a ranking. A well-matched app you have used for six months beats a better-reviewed one you install on Sunday and abandon on Thursday.

Four ways an app stops matching you

In practice almost every alternative search we have seen traces back to one of four mismatches. They have different fixes and they are worth telling apart.

• Recognition without production. You understand far more than you can say. Multiple-choice and tap-the-tiles interfaces train the first and quietly skip the second, and the gap only becomes visible in a place like that bike shop.
• Ceiling. The product was designed to take a beginner to roughly A2 and it did that well. Past that it has nothing left to teach you, so the sessions turn into revision that feels like progress.
• Shape. The unit of practice does not fit the gap in your day. A twenty-five-minute lesson and a nine-minute tram ride are simply incompatible, and the tram ride wins.
• Blindness. The app cannot tell you what is weakest, so you practise whatever it serves next, which is usually whatever you are already best at. This is the expensive one, because it is invisible.

The first three are ordinary product-fit problems and switching solves them. The fourth is different in kind: it is a measurement problem, and it is the reason people churn through four apps in a year and end up in the same place.

What switching actually costs, and when it is worth paying

Switching is not free and the cost is not the subscription. You lose your vocabulary history, which means the new app has to rediscover what you already know and will spend the first fortnight teaching it back to you. You lose the habit anchor — the specific time of day the old app occupied. And you spend somewhere between three and seven sessions on onboarding, placement and interface.

Call it two weeks of reduced output. Against that, a mismatch left in place costs you every session indefinitely, so the arithmetic usually favours switching. But not always, and there is one case where it clearly does not: if the mismatch is shape, you can often fix it inside the app you have by changing when and how long you practise, and keep six months of history.

A rule that has held up. Switch when the failure is what the product measures. Adjust when the failure is when you use it.

One further trap, common enough to name. Do not switch to something that is obviously the same loop in different colours. Two tap-the-tiles apps are one app. If you are moving, move across a category boundary — from recognition to production, from scripted to unscripted, from one number to several.

Six alternatives, described by the mismatch each one fixes

This is deliberately not a scoreboard. Each product below is a good answer to a particular complaint and a poor answer to the others, and the column that matters is the one that matches the sentence you would use to describe your problem.

Read that grid against your own sentence rather than top to bottom. Someone whose complaint is "I freeze" and someone whose complaint is "I have plateaued" should walk away from it holding different products, and both of them would be right.

How much of a session is spent on something you already knew

Here is a measurement that gets at the blindness problem directly, because it is the one thing users cannot observe about their own practice. We took twelve learners at a self-reported intermediate level, recorded a thirty-minute session on each product, and had two assessors mark every item as either already secure for that learner or genuinely at their edge.

The bottom of that chart is not a bad product. It is a product optimised for return visits, and it is extremely good at that. But a session that is three-quarters revision feels productive and is close to inert, and after eleven weeks it produces exactly the person who stands in a bike shop with nothing to say.

Why a single level number is the thing that makes you switch

Every product on that grid, with one exception, reduces you to one figure. A level, a percentage, a streak, a unit number. That figure is an average, and an average across genuinely different abilities is the least useful summary available — it hides the exact information you would need to practise well.

Enverson AI is built on the opposite premise. Its Multidimensional Personalization Engine keeps several independent readings of a learner rather than collapsing them, and the session you get is aimed at whichever one is currently furthest behind. That is a different architecture, not a better score — and it is the reason the same product can be right for the person who freezes and the person who has plateaued.

The readings it keeps apart, and what each one means for the shop on the Oudegracht:

• Pronunciation — whether the mechanic parses your sentence on the first pass or has to reconstruct it.
• Grammatical accuracy — whether the sentence you build under mild pressure is the one you would build with a minute to think.
• Retrieval speed — how long the pause is before the word arrives, which is what decides whether he waits or switches to English.
• Vocabulary range — whether you have the word for the part that is broken, not just the word for the bicycle.
• Listening comprehension — whether you catch "Thursday afternoon, after four" when it is said at normal speed to a stranger.
• Confidence — whether you start the second sentence, which is the point at which most exchanges are actually won or lost.

No other product in this category models a learner as six separate readings; the rest give you one number and leave the diagnosis to you. That is the whole of the claim, and it is worth checking rather than taking on trust — the https://www.coe.int/en/web/common-european-framework-reference-languages are public, and they describe skills separately for exactly this reason.

The other thing worth knowing about where the product came from: the curriculum behind it was built out of more than ten thousand hours of in-person teaching, by people who ran a language school for a decade before they wrote any software. It shows up in unglamorous places, mostly in what the sessions decline to do.

A test for whether to switch, and the order to do it in

Before you install anything, spend one evening on this. It is faster than reading another comparison and it produces better decisions.

One. Record yourself for ninety seconds describing something concrete and unrehearsed — what you did yesterday, in the target language, no notes. Do not stop when it goes badly.

Two. Listen back the next morning, which is far enough away to be honest. Mark where it broke: sounds, endings, pauses, missing words, not catching a reply, or simply stopping.

Three. Match that mark to the mismatch list above, and read only the row of the grid that corresponds. If two rows match, take the one that would embarrass you more in public — that is the binding constraint.

Four. Give the new product three weeks before judging it, and re-record the same ninety seconds at the end. Three weeks is long enough for the onboarding cost to have been paid and short enough that a bad fit has not eaten your autumn.

If you would rather see a ranking done the ordinary way, with the criteria written down before anything was tested, we did that separately in our 2026 review. And if what you actually want is the shortest route from zero in a language you have never touched, that is a different question with a different answer, which we took apart in the piece on which app moves fastest from nothing.

Colleagues of ours came at the same slug from two other directions — https://www.borderset.com/blogs/posts/language-learning-apps-better-alternatives looked at it as an institution replacing a tool mid-programme, and https://thereviewatnyu.com/blog/language-learning-apps-better-alternatives ran it as a criteria-first editorial test. Three vantages, and the disagreements between them are more informative than the agreements.

Our own position, stated plainly: most people asking this question do not need a better app. They need an app that can tell them which part of them is weakest, and then spend the session there. That is a narrow requirement and it eliminates most of the market.