What an AI English tutor can and cannot do for you

What an AI English tutor actually is

An AI English tutor is software you talk to. You speak, it turns your speech into text, judges what you said against a model of good English, and answers back: sometimes in writing, sometimes in a voice, sometimes as a face on a video call. The useful ones then tell you something specific about how you spoke.

That last part is what separates a tutor from a chatbot. Any general-purpose assistant will hold a conversation with you in English, and that alone is worth something: it is speaking practice with no audience and no consequences. But a conversation partner that never tells you your past tense slipped, or that you said "wery" for "very" eleven times, is a place to use English rather than a place to improve it.

So the question is not "does it talk?" Everything talks now. The question is what it notices, and what it does with what it notices.

What AI tutors are genuinely good at

Availability without judgement. The most common reason people stop practising is not difficulty, it is embarrassment. Software will let you restart the same sentence six times at eleven at night, and it will not sigh. For a lot of learners that is the whole difference, and it is not a small thing: most of the fear of speaking is the fear of being watched failing. If that is your situation, the guide to speaking anxiety goes into why it happens and what helps.

Consistency of measurement. A human teacher's attention moves around. Software applies the same checks every time, so a change over eight weeks is a real change and not a difference in mood. That makes progress visible in a way weekly classes rarely do.

Patience with the boring parts. Pronunciation improves through repetition of specific sounds in specific positions. No reasonable person wants to be the teacher who hears you say "thirty-three" forty times, and no reasonable learner wants to ask.

Volume. In a sixty-minute group class you might speak for four minutes. Fifteen minutes alone with an AI tutor is fifteen minutes of your own voice.

What they are still bad at

Be clear-eyed about this, because the marketing in this category rarely is.

Speech recognition is not neutral. Every one of these tools runs on a speech-to-text model, and those models are measurably less accurate on accented English than on the accents they were trained on most. The practical effect is that you can be marked wrong for a word you said correctly. A good tool shows you its transcript so you can see when it misheard you; a bad one hides the transcript and shows you only a score.

A score is not a diagnosis. "Fluency: 68" tells you nothing you can act on. What you need is the sentence you got wrong, what the right version is, and why your first language pushed you towards the wrong one.

Nothing here is an official result. Any tool that reports a CEFR level or an IELTS band is producing an estimate from software. It is useful for tracking your own progress and useless as a credential. Treat a product that blurs that line with suspicion.

Cultural and situational judgement is thin. An AI tutor can tell you a sentence is grammatical. Whether it is the right thing to say to a manager you have met twice is a harder question, and current tools are weaker at it than they sound.

How to choose one

Five questions, in order of how much they matter:

1. Does it make you speak, or can you finish by tapping? Check what a completed session actually requires. A lot of "speaking" apps are vocabulary apps with a microphone in one exercise.
2. Does it show you the transcript? If you cannot see what it heard, you cannot tell the difference between your mistake and its mistake.
3. Is the feedback specific enough to act on tomorrow? One named thing to fix beats five scores.
4. Does it remember your mistakes? A tool that starts fresh every session cannot notice that you have been getting the same conditional wrong since March.
5. Can you explain a confusion in your own language? At the moment you do not understand why something is wrong, English-only explanations stop working.

Two things that should not be on the list: how realistic the avatar looks, and how big the vocabulary is. Neither predicts whether you will speak more English.

Where Speakle fits, and where it does not

Speakle is built for one part of this: people who already understand English and lose it when they have to speak. Every session is answered out loud, scored on pronunciation, fluency, grammar, vocabulary and coherence, and returned with a CEFR level and an IELTS speaking estimate. Mistakes are written to a record that does not reset, so the drills you get are built from your own errors rather than a generic list. Explanations come in nine native languages, Arabic, Hindi, Japanese, Korean, Malay, Thai, Vietnamese, Turkish and Portuguese, because the moment you need your own language is the moment English stopped working.

The pronunciation work is ordered by the language you already speak rather than by a fixed syllabus: an Arabic speaker starts on /p/ and consonant clusters and skips the TH pair they have had since childhood, while a Japanese speaker starts on the L–R contrast. There is a page for each first language setting out which English sounds it tends to merge, and why.

What Speakle is not: it is not a beginner course, if you cannot yet form a simple sentence, a vocabulary app will serve you better first. It is not a substitute for a human teacher on anything that requires judgement about people. It does not issue credentials, and the IELTS estimate it reports is an estimate. And it does not claim its speech recognition is perfect, which is why every session shows you the transcript it scored.

If you want to see what any of this means in practice, the free speaking test takes one spoken answer and needs no account.

The honest summary

An AI English tutor is the best available answer to a specific problem: you need to speak far more English than your life currently gives you the chance to, and the chances you do get are too frightening to use well. It is a rehearsal room. It is not a teacher, not an examiner, and not a certificate.

Used that way, fifteen minutes a day, out loud, with feedback you actually read, it works. Used as something to tap through while thinking about something else, it is a vocabulary app with a microphone.

Five things worth checking before you pay

Free tiers in this category are generous enough to answer most of these in an evening. Spend the evening.

How long is a session, really? Some tools count a two-minute exchange as a session and some expect fifteen minutes. The number that matters is minutes of your own speech per day, not sessions per month.

What happens on your worst day? Every learner has evenings where they are tired and bad at English. A tool that only works when you are sharp will lose you in the third week. Look for a low-effort option: a short spoken note, a gentle version of the daily task, something you can face at eleven at night.

Is the correction level adjustable? Being corrected on every article and preposition is accurate and demoralising. Being corrected on nothing is pleasant and useless. The useful setting is somewhere between, and it changes depending on the week you are having, so it should be a dial rather than a fixed personality.

Where does your audio go? You are sending recordings of your voice to a company. Look for a plain statement about whether those recordings train models, whether they are sold, and how to delete them. If the answer is hard to find, that is the answer.

Does it work in your first language when you need it to? Not for the English, for the explanations. There is a specific moment, when you have just been told a sentence is wrong and cannot see why, where one line in your own language saves ten minutes of confusion.

What progress actually looks like

People expect fluency to arrive as a feeling. It mostly arrives as the absence of things.

In the first few weeks the change is not that you speak better, it is that you start speaking sooner. The gap between hearing a question and beginning an answer shrinks. That gap is where most of the anxiety lives, so closing it feels disproportionately good.

After that, the recoveries get quieter. You still lose words, but you route around them without stopping, and the person you are talking to does not notice anything happened.

Later still, and this is the one people find strangest, you stop translating. Not everywhere and not all at once, but in the topics you have spoken about many times, the English arrives directly.

None of those three show up as a score going up by five points. If you are measuring progress only by a number an app gives you, you will miss all of it and quit somewhere in week four. Keep your first recording. Nothing else makes the change as obvious.

The English speaking hub collects the rest of these guides.