Speak Review

An AI tutor that only talks

The whole app is one instruction: say it out loud

A friendly blue blob teaches you a phrase — ¡Mucho gusto!, ¡Buenas!, ¡Hasta luego! — then blanks part of it, the button says 'Speak now', and it waits for your voice. Get it right and the phrase turns green, with two buttons worth knowing about: 'Example' plays the model pronunciation and 'You' plays back your own attempt, so you can hear the difference yourself rather than trusting a score. Speak does use tapping exercises too, later on, but the default way to answer here is out loud — and that is the gap most learners have, plenty of reading and almost no speaking.

It asks you to commit before you begin

Before the first lesson, Speak asks you to pick a streak goal — 3 days ('baby steps') through 30 ('unstoppable') — and press 'I'm committed', telling you that choosing one makes you several times more likely to finish the course. This is a commitment device straight out of behavioural science, and the useful part is that the target is yours: you decide what you're aiming for before you start, which makes it easier to judge later whether the pace suits you.

Then it drops you into a scene with actual characters

The AI conversation opens with a situation — 'You're at a café with your friend. The barista is about to serve you' — and characters with names and faces: Kai the friend, Noa the barista, Lumi later on. Each line they say carries its own buttons to translate it, save it, hear it again, or report it. It reads like a graphic novel you're a character in, and it gives the practice a reason to exist beyond drilling.

Listen, then rebuild what you heard

Between spoken turns, Speak checks comprehension by hiding part of a line and asking you to reconstruct it from word tiles — 'Aquí están ___' with sus, cafés and su to choose from. It's a small thing, but it stops the conversation being a wall of audio you can nod along to without following, and it targets the words that carry grammar rather than the ones carrying meaning.

Rehearsal, then the real thing

Having walked you through the scene once, it runs the same situation again with a new barista and no script: 'Now it's your turn.' The mic listens, prompts tell you the goal of each turn — greet the barista, thank them, say no thanks and goodbye — and each attempt comes back marked 'Good response'. It is the closest thing to conversation practice available without another human being, and the two-pass structure means you are never asked to improvise something you haven't just seen modelled.

You finish having actually talked to someone

The lesson ends with an exchange you carried yourself — 'Gracias, hasta luego' answered by '¡Adiós!' — and a plain verdict: 'Now you're ready to greet and thank someone like a local.' No chest, no gems. The limitation is the one every AI partner has: it is endlessly patient, never confused, and never asks the thing you didn't prepare for. That makes it excellent rehearsal and slightly flattering, which is worth knowing before you conclude you're ready for Madrid.