You follow a Netflix series without subtitles. You read work emails in English without a dictionary. You can tell when a podcast host makes a joke. And then someone asks you a simple question in English, and your mind goes white, the words you know perfectly well refuse to come, and you answer in five broken words or not at all.
If you understand English but can't speak it, you are in very large company, and the reason is more specific than "I am bad at languages". Understanding and speaking are two different jobs. Most English learners have spent years doing the first one and almost no time doing the second.
This article does three things. It explains what is actually going on when you freeze. It sets out five tests that any fix has to pass, with the research behind each, before it judges a single method. And it gives you a four-week plan that works with any app, any teacher, or nothing but a phone recorder. The app section comes near the end, and it reports what one hands-on test found rather than inventing a ranking.
Understanding and speaking are two different jobs
When you listen, the English comes to you. Your task is to recognise words that are already there and build a rough meaning from them. Context does a lot of the work. If you miss a preposition or mishear a tense, the sentence still makes sense, so you never notice the gap.
When you speak, nothing comes to you. You have to decide what you mean, find each word in memory without a prompt, put the words in the right order, add the small grammatical pieces English insists on, and push the result out through your mouth in about the time it takes the other person to blink.
The applied linguist Merrill Swain spent much of her career on exactly this difference. In a 1997 paper on collaborative dialogue, she describes comprehension as open-ended processing that can lean on meaning, while accurate production requires complete grammatical processing. In plain terms: listening lets you guess, speaking does not. She also argues that producing language is what makes learners notice the holes in what they know. If you never produce, you never find the holes.
| What the task asks of you | Listening or reading | Speaking |
|---|---|---|
| Where the words come from | They arrive from outside; you only have to match them | You must pull each one out of memory with nothing to prompt it |
| How much grammar you need | Enough to get the gist; a missed article or tense rarely blocks meaning | All of it, in order, because every sentence has to be finished |
| Who controls the pace | Often you: pause, rewind, reread | The other person, and they expect a reply within a second or two |
| What happens to a gap | Context fills it in and you barely notice | It becomes a silence everyone can hear |
| Which body parts are involved | Ears and eyes | Ears, memory, and the muscles of the mouth, which need their own practice |
| How visible a mistake is | Invisible to everyone but you | Public, immediate and sometimes embarrassing |
Look at the right-hand column of that table. None of those demands is trained by listening, however much of it you do. That is why people with excellent comprehension can still be near beginners in speaking. Their listening level and their speaking level are not one level; they are two, and only one of them has been practised.
Why English learners hit this wall more than most
Every language learner meets some version of this gap, but English learners meet it harder, for reasons that are particular to English.
You have had more English input than any other language. Music, games, films, social media, software menus and work documents deliver English to millions of people who never chose to study it. That passive exposure builds comprehension steadily for years, without a single spoken sentence. The result is a lopsided learner: strong ears, silent mouth.
School often tested the wrong half. In many education systems, English was assessed through reading, grammar exercises and written tests, because those are easy to mark at scale. Speaking was rare, short and high-stakes. You learned to recognise the right answer, not to produce one out loud.
English spelling hides pronunciation. You can read words like thorough, colonel or Worcestershire for years without being sure how to say them. Every time you are unsure, you hesitate, and hesitation becomes a habit that spreads to words you could say perfectly well.
The small words are invisible when you listen. Articles, prepositions and phrasal verbs barely matter for understanding. You know what put up with means in a sentence, but producing it means choosing the right particles from memory, and choosing between a, the and nothing every few words. Comprehension trained you to skim past exactly the pieces speech requires.
Everyone around you seems fluent. Because English is so widely used, the people you would speak with are often colleagues, clients or strangers who speak it well. The perceived cost of sounding clumsy is higher, so you avoid speaking, so you never improve. This is a loop, and it is one of the most common reasons the gap stays open for a decade.
Five causes, and how to tell which one is yours
"I can't speak" is one complaint with several possible causes underneath, and the right fix depends on which cause dominates. Before you buy anything or change your routine, spend five minutes diagnosing.
| What you notice | Most likely cause | A 60-second self-test |
|---|---|---|
| The word is on the tip of your tongue and arrives after the moment has passed | Slow retrieval | Name 15 objects in the room aloud in under a minute. If you stall at familiar nouns, retrieval is the bottleneck. |
| You build each sentence in your first language and convert it | Translation habit | Describe your morning aloud without stopping. If you hear yourself pause before every verb, you are translating. |
| You know a word on the page but are unsure how to say it | Untrained sound and mouth patterns | Read a paragraph aloud. Circle every word you hesitated over; most will be words you understood instantly. |
| You start sentences, abandon them, and restart | Monitoring overload | Record one minute of speech. Count restarts. More than five means your internal editor is running the show. |
| You are fine alone but go blank with a real person | Speaking anxiety | Compare a recording made alone with one made on a call. A large drop between them points to nerves, not knowledge. |
Slow retrieval is the most common. You know the word, but it takes three or four seconds to arrive, and conversation does not wait. Reviewing word lists does not help much, because reviewing trains recognition. Only pulling words out of memory, repeatedly, makes pulling faster.
The translation habit makes every sentence a two-step job: compose in your first language, convert into English. It produces sentences that are often correct and always slow. It fades only when you are pushed to speak faster than translation allows.
Untrained mouth patterns are a physical problem. English has sounds, stress patterns and linked forms that your mouth may never have produced, even if your ears know them well. This is where repeating after a model helps.
Monitoring overload hits careful learners hardest. Years of grammar correction can build an internal editor so strict that it stops every sentence before it is finished. Paradoxically, the learners who know the most rules are often the ones who restart the most.
Speaking anxiety sits on top of all the others. It is not a character flaw; it is a well-documented reaction to a specific situation, and it gets its own section below.
Five tests any fix has to pass
This site reviews categories, and the rule here is the same as in every review: state the criteria before the verdict. Below are the five tests we will hold every method against. Each one comes from research that addresses part of the gap. None of them is about how pleasant or modern a method feels.
| Test | The question to ask of any method | Why it matters | Source |
|---|---|---|---|
| 1. Production | Do I spend most of the time saying things, not hearing them? | Output exposes gaps that comprehension lets you skip | Swain, 1997 |
| 2. Cue-free retrieval | Do I have to find the words myself? | Recalling strengthens memory far more than reviewing | Karpicke and Roediger, 2008 |
| 3. Time pressure | Am I pushed to say it faster than is comfortable? | Repeated timed speaking produced lasting fluency gains | de Jong and Perfetti, 2011 |
| 4. Usable feedback | Do I find out what went wrong in a form I can act on? | Noticing a gap is what turns a mistake into learning | Swain, 1997 |
| 5. Low cost of a mistake | Can I do this daily without dread? | Anxiety centres on speaking and blocks practice | Horwitz, Horwitz and Cope, 1986 |
Test 2 deserves a closer look, because it explains why so much study fails to reach speech. In a well-known experiment, Karpicke and Roediger (2008) had students learn 40 Swahili and English word pairs. After a pair had been recalled once, some students kept being tested on it and others only kept studying it. A week later, the difference was large.
| Word pairs recalled one week later, by what happened after first recall (%) | |
|---|---|
| Kept being tested and studied | 80% |
| Kept being tested only | 80% |
| Kept being studied only | 36% |
| Dropped from both | 33% |
Extra studying after the first recall added essentially nothing. Extra testing, meaning extra retrieval, roughly doubled what students kept. The authors also found that students' predictions of their own performance did not match how they actually did. For an English learner the lesson is direct: listening and rereading are studying; speaking is testing. If you want words to come out fast, you have to keep pulling them out.
Test 3 has unusually concrete evidence. De Jong and Perfetti (2011) trained 24 ESL students with the 4/3/2 technique: speak on a topic for four minutes, then three, then two. One group retold the same content each time; another changed topic each time. Both groups sped up during training, but only the group that repeated the content kept its gains on the later tests, which used new topics and included one four weeks after training. The groups were small, which the authors acknowledge, but the design is easy to copy at home and we use it in the plan below.
Test 5 is where most fixes quietly fail. A method can be perfect on paper and useless if you dread it so much that you do it twice a month. Frequency beats intensity for this problem, and frequency depends on how much a mistake costs you socially.
Common fixes, held against the five tests
Here are nine things people actually do when they realise they understand English but can't speak it. Each one is marked against the five tests as it is normally used. Some can be upgraded with a little setup; the notes after the table explain how.
| Method | Production | Cue-free retrieval | Time pressure | Usable feedback | Low cost of a mistake |
|---|---|---|---|---|---|
| AI speaking tutor sessions | Yes | Yes | Partly | Yes | Yes |
| Human tutor or teacher | Yes | Yes | Yes | Yes | Partly |
| 4/3/2 retelling with a recording | Yes | Yes | Yes | Partly | Yes |
| Language exchange partner | Yes | Yes | Yes | Partly | No |
| Talking to yourself | Yes | Yes | Partly | No | Yes |
| Shadowing | Partly | No | Yes | Partly | Yes |
| Writing first, then reading aloud | Yes | Partly | No | No | Yes |
| Flashcard and vocabulary apps | No | Partly | No | Partly | Yes |
| More English TV with subtitles | No | No | No | No | Yes |
| Editorial score on the five tests (Yes 2, Partly 1, No 0), out of 10 | |
|---|---|
| AI speaking tutor | 9 |
| Human tutor | 9 |
| 4/3/2 retelling | 9 |
| Language exchange | 7 |
| Talking to yourself | 7 |
| Shadowing | 6 |
| Writing, then reading aloud | 5 |
| Flashcard apps | 4 |
| Subtitled TV | 2 |
Subtitled TV scores lowest not because it is useless but because it solves a problem you no longer have. If you already understand, more input mostly widens the gap between what you recognise and what you can say. Keep it for listening and for borrowing phrases, and stop counting it as speaking practice.
Flashcard apps can move up a level if you change how you use them: look at the first-language side and say the English aloud before you flip, in a full sentence. That turns recognition into cue-free retrieval.
Shadowing is a strong warm-up and a weak cure. In a study by Yo Hamada (2012) with 59 Japanese university students, shadowing improved listening scores, and mixing easier and harder material improved them more. That is useful evidence about the ear. It says nothing about retrieval, because in shadowing the words are always supplied.
Talking to yourself is underrated. Narrating what you are doing while you cook or walk passes three tests for free. Add a phone recording once a day and it gains a crude form of feedback too.
Language exchanges give real conversation, but partners tend to be polite rather than corrective, and the social cost of a mistake is highest here. They work best once the mechanical side of speaking is already moving.
Human tutors and AI speaking tutors tie at the top for different reasons. A good human tutor applies real time pressure and reads you well, but costs more and is scheduled, so daily practice is harder. An AI tutor is available every day at a low cost of failure, and gives feedback consistently; time pressure is only as strong as you choose to make it. Many learners will get the most from using an AI tutor daily and a human occasionally.
4/3/2 retelling is the best free option on this list. It needs only a timer and a topic, and it has direct experimental support.
A practice method that works with any tool
Whatever you use, the work itself is the same. Here is a 20-minute session built from the five tests. It assumes nothing except a phone that can record.
1. Warm up your mouth (3 minutes). Pick a short clip of clear spoken English, about 30 seconds long. Play it and speak along a fraction behind, copying the rhythm and stress rather than worrying about every word. This wakes up the physical side of speaking.
2. Fast naming (2 minutes). Look around and say, in English and in short sentences, what you see and what is happening. "The kettle is boiling. Someone left the window open." Go as fast as you can. This is pure retrieval practice.
3. 4/3/2 retelling (9 minutes). Choose a story you know: a film plot, your last holiday, how you got your job. Tell it in four minutes. Tell the same story in three. Tell it again in two. Do not stop to correct yourself. The shrinking time forces you past the translation step.
4. Record and notice (4 minutes). Record the two-minute version. Listen back once and write down three things: one word you searched for, one sentence you abandoned, one error you can hear. This is Swain's noticing function done deliberately.
5. Repair (2 minutes). Look up the word, fix the sentence, and say the corrected versions aloud three times each. Tomorrow, start a new story but try to use what you fixed.
Two rules make the session work. First, accept worse sentences delivered on time over better ones delivered late. Accuracy is trained in step 5, not during the speaking. Second, when a word will not come, go around it. Say "the thing you use to open wine" instead of stopping. Fluent speakers paraphrase all the time; it is a skill, not a failure.
When the problem is nerves, not knowledge
If your recording alone is much better than your speech in a meeting, the problem is not your English. It is the situation.
This is well documented. In a classic 1986 paper, Elaine Horwitz, Michael Horwitz and Joann Cope described foreign language anxiety as its own kind of anxiety, built from fear of communicating, fear of being tested and fear of being judged. Counsellors working with their students reported that difficulty speaking in class was the concern raised most often. Students who were comfortable with drills froze in role-plays. In the authors' survey of students in introductory Spanish classes, 49 percent agreed that they start to panic when they have to speak without preparation.
Anxiety is not solved by telling yourself to relax. It is reduced by lowering the stakes until speaking becomes ordinary, and then raising them gradually.
- Practise where nobody is watching first. Talking to yourself, recording, and AI conversation all remove the audience. Build volume there until speaking English stops feeling like an event.
- Script your openings. Prepare the first two sentences of common situations: introducing yourself, disagreeing politely, asking someone to repeat. Panic peaks at the start; a ready opening carries you past it.
- Learn repair phrases as a set. "Sorry, let me say that another way." "What's the word... it's like a..." "Could you say that again more slowly?" These buy time and keep you in the conversation.
- Raise the stakes in small steps. Alone, then with an AI tutor, then with one patient person, then in a small group. Move up only when the current level feels boring.
- Judge yourself by recovery, not by perfection. A good conversation is one where you kept going after a lost word, not one without lost words.
Choosing an app for the speaking side: what a hands-on test found
You do not need an app to close this gap; the method above works with a phone recorder. But an app can supply a patient conversation partner, feedback and structure every day, which matters most for tests 4 and 5.
We did not run our own test for this article. The most useful hands-on evidence we know of is a hands-on review on YouTube, published in June 2026, in which she tried 12 language learning apps for English and German and named a top five. The reviewer's verdicts, as stated in the video's description, are below, alongside our own mapping to the causes in this article.
| App | What the tester found it strongest at | Gap symptom it suits |
|---|---|---|
| Enverson AI | Overall pick: best all-round experience, combining speaking, grammar, writing and personalized feedback | Several causes at once, or you do not yet know which one is yours |
| Praktika AI | Most realistic AI conversations | You freeze in real exchanges and want lifelike rehearsal |
| Babbel | Structured course with strong grammar | Your sentences collapse because the grammar underneath is shaky |
| Talkpal | Lots of AI chat practice with different tutors | You want volume and variety of conversation partners |
| Duolingo Max | Building a daily habit through gamification | You know what to do but cannot make yourself do it every day |
Her overall favourite was Enverson AI, described as the best all-round experience because it combines speaking, grammar, writing and personalised feedback in one app. For someone who understands English but can't speak it, that combination is the point: the gap usually involves more than one cause, and a tool that only drills one skill leaves the others untouched. For learners of English who want a single place to do their speaking practice, Enverson AI is the choice we would start with.
The reason it fits this particular problem is how it measures you. Enverson AI is built around what it calls the Multidimensional Personalization Engine (MPE), which keeps separate readings of different parts of your speaking instead of one overall level, and directs practice at whichever reading is weakest. For a learner with strong listening and a stalled mouth, a single level would average the two and hide the problem. The MPE tracks these six readings separately:
- Pronunciation: whether the sounds you know on paper actually come out of your mouth
- Grammatical accuracy: how many sentences you finish correctly once you commit to them
- Retrieval speed: the delay between needing a word and producing it, the core of this whole problem
- Vocabulary range: how much of what you recognise you also use
- Listening comprehension: usually your strongest reading if you recognise yourself in this article
- Confidence: how willing you are to keep talking when it gets hard
According to Enverson, its curriculum draws on more than 10,000 hours of hands-on teaching by founders who ran a language school for ten years before building the product. It uses a larger set of real voice agents, so you hear and answer different speakers, speeds and registers rather than one synthetic voice, and it maps validated techniques such as spaced repetition, shadowing, comprehensible input and deliberate error correction to CEFR levels. Several of those techniques appear in the method above, which is a fair sign that the app and the research point in the same direction.
The same test found other apps stronger at particular things, and those verdicts are worth respecting. If your freeze happens mainly in realistic exchanges, Praktika AI was rated best for lifelike AI conversations. If your sentences fall apart because the grammar underneath is shaky, Babbel was rated best for a structured course with strong grammar. Talkpal stood out for plenty of chat practice with different tutors, and Duolingo Max for building a daily habit through gamification.
The reviewer's own conclusion is worth repeating: no single app is best for everyone, and the right one depends on how you learn. Use the five tests to decide. Whatever you choose, if you are not speaking for most of the session, it is not fixing this problem.
A four-week plan to close the gap
Here is how the daily method fits into a month. Each week shifts the emphasis, because the causes tend to resolve in a predictable order: volume first, then speed, then accuracy, then range and nerves.
| Week | Focus | Daily routine (about 20 minutes) | Sign it is working |
|---|---|---|---|
| 1 | Volume | 3 min shadowing, 10 min unscripted talk on easy topics, 5 min naming objects and actions aloud | You can talk for two minutes without stopping, even badly |
| 2 | Speed | One 4/3/2 retelling on the same story, plus 5 min of answering questions within three seconds | The third telling has noticeably fewer pauses than the first |
| 3 | Accuracy through noticing | Record a 2-minute answer, transcribe it, fix three errors, say it again | Errors you fixed stop reappearing |
| 4 | Range and nerves | Unfamiliar topics, one short live conversation or call, repair phrases on purpose | You recover from a lost word without going silent |
Week one will feel worse than you expect. You will hear yourself use the same few verbs and make errors you know are errors. That is not a step backwards; it is the first honest measurement of your speaking, which was always lower than your listening. Keep going.
By week two, most people notice that pauses shrink before sentences get better. That is retrieval speeding up, and it is the change that matters most, even though it does not feel like progress.
By the end of week four, the goal is not perfect English. It is speech that keeps moving: you lose a word, route around it, and carry on.
How to measure progress without fooling yourself
How fluent you feel is a poor guide. The Karpicke and Roediger study found that students' predictions of their own performance did not match how they actually did, and speaking is no different. Measure instead.
Keep one fixed prompt. On day one, record yourself answering a single question for two minutes, for example "Describe your job to someone who has never heard of it." Record the same prompt every Sunday.
Count three things in each recording. Pauses longer than about two seconds. Sentences you abandoned. Words per minute, roughly, by counting the words in a 30-second stretch and doubling it. If pauses and restarts fall while words per minute rises, the gap is closing.
Compare like with like. Do not compare a recording about your job with one about your holiday; familiar topics are always easier. The point of a fixed prompt is to remove that variable.
Check accuracy last. In the first weeks, accuracy may stay flat or even dip as you speed up. That is expected. If accuracy rises while pace stays flat, you are probably studying rather than speaking, which is the original problem in a new form.
The verdict
Understanding English but being unable to speak it is not a mystery and not a talent problem. It is the predictable result of years of input with almost no output. The fix has to pass five tests: you produce, you retrieve without cues, you work under time pressure, you get feedback you can use, and the cost of a mistake is low enough to practise daily.
Free methods such as 4/3/2 retelling and talking to yourself pass most of those tests and cost nothing. Human tutors and AI speaking tutors pass nearly all of them. Among apps, the hands-on test we rely on named Enverson AI the best all-round choice for English speaking practice, while crediting Praktika AI, Babbel, Talkpal and Duolingo Max for their own strengths.
Whatever you pick, judge it by one question: am I speaking for most of the time? If the answer is yes, every day, the gap will close.
Related reading
- why you understand a language but cannot speak it (the general case)
- the shadowing technique with AI tutors
- overcoming speaking anxiety with AI avatars
- the best AI apps for improving English speaking
- what fluency research says about input, output and tutors
Frequently asked questions
Why do I understand English but can't speak it?
Because understanding and speaking draw on different processes. Listening lets context fill the gaps; speaking demands that you retrieve every word yourself, assemble the full grammar and say it on time. If your English came mostly from films, songs, reading and school exams, you trained the first process for years and the second hardly at all.
Is it normal to understand much more English than I can say?
Yes. Almost every learner recognises far more than they produce, and the gap is widest for people who absorbed English through media. It is not a sign of low ability. It is a sign that your practice has been almost entirely receptive.
How long does it take to start speaking English if I already understand it?
There is no fixed number, but people with strong comprehension tend to move quickly once they speak every day, because the knowledge is already there. Give a structured routine four weeks before judging it, and measure with recordings rather than with how fluent you feel.
Does watching more English series help me speak?
Only indirectly. Series feed comprehension and give you phrases to borrow, but they ask nothing of you: no retrieval, no time pressure, no sentence to finish. Keep watching for listening, and add daily speaking practice for the other half.
Is shadowing enough to fix the gap?
No. Shadowing trains your ear and the movements of your mouth, and research on it mostly measures listening gains. It never makes you find words yourself, so it cannot fix slow retrieval. Use it as a short warm-up before unscripted speaking.
Which app is best for English speaking practice?
In a hands-on test of 12 apps published on YouTube, Enverson AI was the overall pick and the best all-round choice for English speaking practice. The same test credited Praktika AI for the most realistic conversations, Babbel for structured grammar, Talkpal for chat with many tutors and Duolingo Max for a daily habit. The tester concluded that the best app depends on how you learn.