A placement test has an impossible job. In twenty questions it has to stand in for everything you have learned - months or years of it - and put you on one rung of a six-step ladder. It will sometimes get that wrong. The interesting questions are what it can see, what it cannot, and what happens next.
What the questions actually probe
Ours is typical of the genre, so it makes a good example. Every language has twenty questions, four at each level from A1 to C1. Each is one sentence with a gap, four options and a short English hint so the early ones are not impossible. A German learner starts with:
___ Tag! Wie geht's? (Gute / Guten / Gut / Hallo)
and, a few levels later, meets:
Wenn ich Zeit ___, würde ich reisen. (habe / haben / hatte / hätte)
and, at C1:
Der Verlag hat uns die Daten freundlicherweise zur ___ gestellt. (Verfügung / Auswahl / Ordnung / Kenntnis)
Look at what changes as you climb. The A1 item is a set phrase and an ending. The B1 item needs the conditional mood - you have to know that if I had time is not a plain past. The C1 item is a fixed expression that you either know or you do not; no rule gets you there. A Portuguese B2 item works the same way: Assim que eu ___ da decisão final, aviso-te wants souber, the future subjunctive - one of the forms that sets Portuguese apart from Spanish.
So what the test really measures is recognition of grammar and usage in context, calibrated against the level at which each thing is normally learned. That is a genuinely good proxy. Grammar is the part of a language that develops in a fairly predictable order, and recognising the right form is a fast, cheap signal of how far along that order you are.
What it cannot see
Be clear about the limits, because they explain most "wrong" placements.
- Speaking. Recognising hätte on a screen is not producing it at speed in
a conversation. It is common to test a level above where you can speak.
- Listening. Nothing in a text item tells you whether you can follow a
native speaker at full pace.
- Vocabulary size. Four options in a sentence you can mostly read is a
very different task from finding a word you have never needed.
- Heritage and self-taught profiles. Someone who grew up hearing a language
often speaks far above the level their written grammar suggests - or the reverse, for someone who learned from textbooks.
None of that is a flaw in a particular test. It is what multiple choice is. A placement test sets a starting point, and it should be treated as one.
Why it should round down
Given that it will sometimes be wrong, which direction should it be wrong in? We made ours conservative on purpose: to be placed at a level you need three of its four questions right and at least half of everything below it. The test walks down from C1 and places you at the highest level that passes both checks; fail them all and you start at A1.
The reason is that the two mistakes do not cost the same. A strong learner placed one level low finds the material easy, and the fix is a short climb. A beginner placed too high is handed content they cannot engage with at all, and the natural response to that is to stop. We went through the same reasoning in what the CEFR levels actually mean.
The second half of the rule - half of everything below - is what stops a test this short being guessable. Three lucky answers at C1 are not enough on their own. Worked through exactly, someone answering every question at random is placed at A1 about 94% of the time and reaches C1 less than half a percent of the time.
What should happen after it
A placement test that sets your level once and never again is doing half a job. The rest is how you move when it was wrong.
Down is free. If the level feels too hard, you can step down on your profile at any time. Nobody needs to prove they are struggling.
Up needs evidence, and there are two routes. The short one is the level challenge: eight questions, four at your current level and four at the next, and you need three of four at both. It draws on a separate question bank from the placement test, so it is never a memory test of questions you have already seen, and you get one attempt a week per language - enough for a learner who slipped, too few for someone trying to guess their way up (random answers pass it about a quarter of a percent of the time).
The long route is simply working the course. When you have finished 80% of your level's units, kept 30 of its words long enough that they come round only every three weeks, and built a steady practice habit, the app offers you the next level. It never moves you silently: a tutor that suddenly speaks above you with no explanation is worse than one that never moves at all.
How to take one well
- Do not look things up. A placement that reflects your dictionary rather
than you puts you somewhere you cannot stay.
- Skip what you do not know rather than agonising. A blank is honest data.
- Judge it after a week, not after one lesson. If a week later it all
still feels easy, take the challenge.
Every one of our 19 languages runs its placement test, its path and its decks from A1 to C1, so wherever you land there is somewhere to go. Each level also has its own page - Portuguese B2 or German A1, for example - showing what that band covers and the words it teaches. Or skip the test entirely and try a conversation to see how it feels.

