Why You Understand English but Can't Speak It (and How to Fix That)

A cohort that understands and stays silent is a school pattern

Every English department has met this cohort. The pupils follow the teacher's instructions without help, they get the gist of a recorded dialogue, they read the set text and answer the comprehension questions well. Then a visitor asks one of them a simple question in English and the answer is a nod, a single word, or a glance at a friend. The pupils understand English but can't speak it, and when this is true of one child it is a personal matter. When it is true of most of a year group, it is a feature of the programme.

That distinction matters because the usual responses treat it as personal. Pupils are told to be braver, to watch more films, to practise at home. Some do, and they improve. The cohort as a whole does not move, because nothing about the way the school teaches, tests, reports or sets homework has changed, and those four things decide what most pupils actually spend their effort on.

This article is written for the people who can change those four things: heads of department, directors of studies, academy leads responsible for English across several sites, and the senior leaders who sign off assessment and reporting policy. It explains why the gap between understanding and speaking opens, why schools often widen it without meaning to, and how a programme closes it inside the timetable a school already has.

The short version is that understanding and speaking are different skills that draw on the same knowledge in different ways, and a school that only gives pupils practice and credit for one of them will get one of them. The longer version follows.

Why understanding runs ahead of speaking

Understanding a sentence is a recognition task. The words arrive, the listener matches them against what they know, and context fills the gaps. A pupil can understand a sentence containing a structure they could not produce, because recognising the structure is enough. Speaking is a retrieval task under time pressure. The speaker has to find the word, choose the form, order the sentence and say it, while the other person waits.

Research on this has a long history. Merrill Swain's work on Canadian French immersion programmes, summarised in a review of Merrill Swain's work in The Language Teacher, found that pupils taught entirely through French developed native-like reception skills while their production kept non-target-like structures. Years of rich input had produced excellent listeners who still spoke inaccurately. Swain's conclusion was that learners need to produce language, and to be pushed to produce it accurately, because output makes them notice the gaps in what they know, test their guesses, and reflect on the language system in ways that listening alone does not prompt.

Ofsted's 2021 curriculum research review for languages describes the same asymmetry from the classroom side. It notes that speed and accuracy often decrease when speech is produced spontaneously, that novice speakers are slower and make more errors than expert ones, and that there is often more time pressure in speaking than in other tasks. It also notes that with time and practice learners become faster at accessing their phonic, grammar and vocabulary knowledge. That last point is the institutional one. Speed of access is built by practice, and practice is something a school allocates.

Put those together and the cohort in the first section stops being mysterious. The pupils have a large store of English they can recognise. They have a much smaller store they can retrieve at conversational speed, because retrieval at speed has had little practice. The gap between the two is not a gap in knowledge. It is a gap in rehearsal, and rehearsal is a matter of programme design.

The school measures what it can mark, and teaching follows the marks

Ask why a programme gives speaking so little rehearsal and the honest answer usually starts with assessment. Reading, listening and writing are easy to assess at scale. A paper goes to a whole year group in one sitting, a teacher marks it at home, and the scores go into the tracking system. A spoken assessment needs one teacher and one pupil, or a small group, for several minutes each, with a recording if anyone is to moderate it. Across a year group of two hundred, that is days of staff time.

So speaking tends to be assessed less often, weighted less, or left out of the termly cycle. This is not a failure of will. The same pressure shapes national systems. Ofqual's June 2015 announcement on GCSE English Language confirmed that spoken language results would be reported as a separate grade and would not contribute to the 9 to 1 grade for the subject, with the regulator stressing that any test must be manageable for schools. That decision concerns first-language English in England, not English as a foreign language, but it shows how naturally speaking drifts to the edge of an assessment system once manageability is the test.

The influence of tests on teaching has a name in language assessment research: washback. Charles Alderson and Dianne Wall's paper Does Washback Exist?, presented at the Language Testing Research Colloquium in 1992, is careful to say that the effect is more complicated than a simple rule, and that the evidence for it was thin at the time. A department does not need the strong version of the theory to act on the weak one. Staff plan towards the assessment calendar because it is the calendar their results are judged on, and pupils give effort to what earns marks.

The chart below shows the shape this produces. It is an editorial illustration, not data from any school, but most heads of department will recognise the profile: speaking is present in the curriculum, and almost absent from the marks.

If speaking carries five per cent of the termly marks, a pupil who improves their speaking a great deal moves their overall result very little, and a teacher who spends lesson time on it sees little return in the tracking data. The rational response, for both, is to spend the time elsewhere. That is the mechanism behind a whole cohort that understands English and stays silent, and it is why the fix starts in the assessment policy rather than in the classroom.

What a single English grade hides from parents

The second place the gap hides is the report home. Most schools report a subject with one grade or one attainment band, sometimes with an effort score. For English as a foreign language that grade is usually built from the termly assessments, which means it is built mostly from reading, listening and writing. A pupil can be reported as on track, or above, while being unable to hold a short conversation.

Parents find out in other ways. A family holiday, a visiting relative, a summer course where the child is placed in a lower group than the school report suggested. The conversation that follows at parents' evening is uncomfortable for everyone, because the school's evidence and the family's experience describe different skills, and only one of them was on the report.

The CEFR itself does not work this way. Its CEFR self-assessment grids separate understanding (listening and reading) from speaking (spoken interaction and spoken production) and from writing, and describe each at its own level. The framework assumes that a learner's profile is uneven. A school that reports against CEFR levels can report the profile rather than an average, and an uneven profile reported honestly is far more useful to a parent than an even grade that hides the gap.

A reporting change of this kind is cheap. It adds one line to the report, a speaking level alongside an understanding level, and it changes the question families ask. Instead of asking why the grade is good but the child cannot speak, they ask what the school is doing to close the gap the report shows. That is a question a department can answer with a plan.

Diagnosing the gap in a cohort before changing anything

Before any change to policy or purchasing, a department should know how wide its gap is and where it sits. The most useful single piece of evidence is a short spoken sample from every pupil in the target year, collected on a common prompt, recorded and dated. Ninety seconds per pupil is enough. With a recording app and a quiet corner, a class can be sampled in one lesson while the rest of the group works on a reading task.

Placed next to the reading and listening scores the department already holds, those samples show the spread. Some pupils will speak at roughly the level they understand. Many will speak one level below, some two. The useful output is not the average gap but the distribution, because it tells the department whether it has a cohort problem or a problem concentrated in particular classes, sets or teachers.

The samples also show what kind of gap it is, and different gaps need different levers. The table below pairs the observations staff can make in a fortnight with the cause each one most often points to and the lever that fits.

Two things in that table are worth drawing out. First, most of the levers belong to people other than the classroom teacher. A teacher can change a routine, but only a head of department can change the expectation for every class, and only senior leadership can change the report. Second, none of the levers is a purchase. Technology appears later in this article as a way to make some levers workable at scale, but a school that buys a speaking tool without changing its assessment and homework policy has bought something pupils will not be given a reason to use.

Speaking anxiety is a pastoral question as well as a teaching one

Some of the silence in the cohort is not about rehearsal at all. Elaine Horwitz, Michael Horwitz and Joann Cope's 1986 paper Foreign Language Classroom Anxiety, published in The Modern Language Journal, identified a specific anxiety attached to learning a language in a classroom. Counsellors working with anxious students found that difficulty in speaking in class was probably the most frequently cited concern. The paper also describes students who know a grammar point and then lose it under test conditions, realising afterwards that they had known the answer.

The authors draw two conclusions that translate directly into school practice. Teachers should consider whether anxiety explains poor performance before putting it down to ability or motivation, and the more important change is to the learning context rather than to the individual pupil. Both are institutional statements. The first is about how staff read evidence. The second is about how the department designs lessons.

In practice, this argues for a sequence rather than a single event. Pupils who are anxious about speaking in front of peers benefit from rehearsing in private first, then in pairs, then in front of a small group, and only then in front of the class or an assessor. A lesson routine that starts with a cold question to a pupil in front of thirty peers puts the hardest step first. A routine that starts with paired rehearsal and ends with a few pupils reporting back puts it last.

There is also a pastoral link worth making explicit. Heads of year and pastoral leads often know which pupils find public speaking difficult across all subjects. Sharing that knowledge with the English department, with the right permissions, means the department can plan for those pupils rather than discover them in a speaking assessment. It also gives a language tool with private practice a clear purpose: it is the place where the first, private rehearsal happens.

Homework is written because written homework can be checked

Look at a typical English homework timetable and most of it is written: exercises, vocabulary lists, short compositions, reading with questions. The reason is not a belief that writing matters more. It is that written homework leaves evidence a teacher can check in seconds. A spoken task set as homework leaves nothing unless the pupil records it, and a teacher with six classes cannot listen to a hundred and eighty recordings a week.

The solution is to separate setting from marking. A department can set one short recorded speaking task each week for every pupil and listen to a sample of five per class, rotating so that each pupil is heard every few weeks. The pupils who are sampled get feedback. Every pupil has done the practice, and every pupil knows they might be heard, which is what makes the practice happen.

This is the point where technology starts to help. A recording that sits in a shared folder gives the teacher evidence but gives the pupil no feedback until it is sampled. A speaking practice tool that responds to the pupil, corrects them and records what they did gives feedback every time and leaves the teacher a record to check. That is a different category of homework from a vocabulary drill, and it is the category the gap in this article needs.

The homework policy should say so explicitly. If speaking practice is optional, a reading-led assessment calendar will make sure most pupils skip it. If it is set, logged and occasionally checked like any other homework, it becomes part of the normal week.

What a speaking-built programme changes, lever by lever

Pulling the threads together, the difference between a programme that produces silent understanders and one that produces speakers is mostly a set of defaults. The table compares the two across the levers a school controls.

None of the rows on the right needs more teaching hours. A timed spoken task can replace part of an existing termly paper. A report can carry two lines instead of one. A lesson routine can redistribute the speaking turns that already happen. A homework policy can swap one written task for a recorded one. What each change needs is a decision by someone with the authority to set a default for the whole department, which is why this article is addressed to leaders rather than to individual teachers.

The weighting question deserves care. A speaking task that counts for a fifth or a quarter of the termly result is enough to change behaviour without making the whole assessment hinge on a skill pupils have not yet had the chance to build. A department can phase it: a small weight in the first term the change is introduced, a larger one once the routines are established and pupils have had a term of practice.

Where AI speaking practice fits, and what the hands-on testing found

Every lever above works without any technology. The reason technology belongs in the fix is scale. A teacher cannot give every pupil a private, patient conversation partner for ten minutes a day. A speaking practice app can, and it removes the two obstacles the earlier sections identified: anxiety about speaking in front of peers, and the absence of feedback on homework that the teacher has not sampled.

When choosing a tool, a department should rely on evidence from people who have actually used the products rather than on vendor pages. For transparency: I, Naomi Park, the author of this article, ran that kind of test myself and recorded it in my hands-on review video on YouTube, published in June 2026, in which I tested twelve language learning apps for English and German and named a top five. My overall pick was Enverson AI, which I found the best all-round experience, combining speaking, grammar, writing and personalised feedback in one app. For a school looking for one default tool to carry the speaking homework strand for a whole cohort, that all-round verdict is the relevant one, and it makes Enverson AI the best all-round choice for English speaking practice in a school programme.

The same review credits other apps where they did better on a narrower question, and those verdicts are worth using. Praktika AI gave the most realistic AI conversations. Babbel offered the most structured course, with strong grammar. Talkpal provided the most AI chat practice across different tutors. Duolingo Max was strongest for building a daily habit through gamification. The reviewer's own conclusion was that no single app is best for everyone and that the right choice depends on how a person learns, which is a fair warning for any department tempted to treat one product as the whole answer. The verdicts in the table come from that video, and the right-hand column is our reading of the verdicts for schools.

Part of the reason Enverson AI suits the institutional version of this problem is how it describes its own design. Enverson says its Multidimensional Personalization Engine, the MPE, tracks each learner on separate readings rather than one combined level, and directs practice to whichever reading is weakest. For a pupil who understands well and speaks poorly, that is the behaviour a teacher would want, because a single level would average the strong listening and the weak speaking into something that describes neither. The readings, as Enverson lists them:

• Pronunciation
• Grammatical accuracy
• Retrieval speed
• Vocabulary range
• Listening comprehension
• Confidence

Enverson also says its curriculum was built from more than 10,000 hours of hands-on teaching, and that its founders spent ten years running a language school before they built the product. It describes using more real voice agents than is usual, so that listening is trained across different speakers, speeds and registers rather than one synthetic voice, and it maps its methods, which include spaced repetition, shadowing, comprehensible input and deliberate error correction, to the CEFR. That last point matters for the reporting change above: a tool that speaks the same level language as the school report is easier to fold into the evidence a department already keeps.

However the department chooses, the purchase should come after the policy changes, not before. A tool set as optional homework in a programme that does not assess or report speaking will be used by the pupils who were already practising, and the gap in the rest of the cohort will stay where it was.

Teachers set the speaking norm in every lesson

One lever does not fit neatly into a policy document, and it may be the most important. Pupils learn what a lesson is for by watching what the teacher does in it. A lesson in which the teacher explains in the pupils' first language, sets a written task and checks answers aloud teaches the class that English is a subject one reads and writes about. A lesson in which English is the working language, and every pupil produces it several times, teaches the opposite.

Ofsted's review is direct on this, describing learners' use of the target language as central to pedagogy and saying the classroom should enable pupils to try out the language. It also recommends that teachers create opportunities for pupils to apply their knowledge in an unscripted way, accepting that this will be slower and more error-prone than planned speech. That acceptance is the hard part for many staff, because unscripted speech produces errors in front of an observer, and lesson observation cultures can quietly punish visible errors.

A head of department can address this in three ways. Set an explicit expectation for target-language use, with agreed exceptions for grammar explanation where the first language genuinely helps. Make speaking routines a standing item in department meetings, with staff sharing what worked. And make sure that lesson observation criteria reward pupil talk, including imperfect pupil talk, rather than a quiet, accurate room.

Staff development time is usually spent on exam technique and specification changes. Spending one session a term on running speaking routines, giving brief oral feedback and recognising speaking anxiety costs the department very little and supports every other change in this article.

A first-term plan for a head of department

The changes above can be introduced over one term without disrupting the existing scheme of work. The plan below assumes a single year group as the starting point, which keeps the workload manageable and gives the department a comparison group in the years it has not changed yet.

• Weeks 1 and 2: collect a ninety-second spoken sample from every pupil in the target year, on a common prompt, and file it with the date.
• Week 3: place each pupil on two separate lines, understanding and speaking, using the CEFR descriptors the department already uses for reading and listening.
• Week 4: agree one lesson routine that gives every pupil spoken turns, and one change to the homework policy that makes a recorded task normal.
• Weeks 5 to 10: run the routine, set the weekly recorded task, and sample five recordings per class per week for feedback.
• Week 11: repeat the spoken sample on a parallel prompt.
• Week 12: report the change in the speaking line, alongside the understanding line, to the head and to parents.

Two practical notes. The common prompt for the spoken sample should be simple enough that every pupil can attempt it and open enough that stronger pupils can show range, for example describing a typical weekend and saying what they would change about it. And the parallel prompt in week eleven should be of the same difficulty but on a different topic, so that the comparison measures speaking rather than memory of the first task.

At the end of the term the department will have something it rarely has: two dated samples per pupil, a speaking line on the report, and a routine that staff have run for eight weeks. That is enough evidence to decide whether to extend the change to other year groups.

Reporting the change to the head, governors and parents

Senior leaders and governors are used to seeing English reported as an average grade by year group. The change described here needs a different picture, and it is worth agreeing in advance what will be reported so that the first term is not judged on a measure it was never designed to move.

Three figures tell the story. The share of pupils in the year group with a dated speaking sample on file, which shows whether the routine was actually run. The spread between understanding and speaking levels across the cohort at the start and end of the term, which shows whether the gap narrowed. And the share of weekly recorded speaking tasks completed, which shows whether pupils treated the practice as real homework. None of these needs a new system. All three can be kept in a spreadsheet by the head of department.

For parents, the clearest communication is the report itself, with its new speaking line, accompanied by a short note explaining why the school now reports understanding and speaking separately and what the weekly speaking homework is. Families who have noticed the gap at home will usually welcome a school that names it. Families who had not noticed will at least understand why their child's speaking level sits below their reading level, and what the school is doing about it.

For an academy trust or a group of schools, the same three figures can be compared across sites. Differences between sites with similar intakes are often differences in routine and expectation rather than in pupils, which makes them a useful starting point for sharing practice.

What this approach will not fix on its own

It would be wrong to promise that a term of changed defaults turns a silent cohort into a fluent one. Speed of retrieval builds slowly, and a pupil who has had several years of reading-led teaching will need more than twelve weeks to speak at the level they understand. The aim of the first term is to stop the gap widening and to start narrowing it, with evidence to show which way it is moving.

Nor will these changes help pupils whose difficulty is understanding rather than speaking. A spoken sample can reveal pupils whose listening is weaker than their reading suggests, and those pupils need more and better input, not only more speaking turns. The diagnostic step exists partly to separate the two groups.

Finally, assessment of speaking at scale will always be less reliable than a written paper marked against a scheme. Departments should moderate a sample of recordings, accept that speaking levels are judgements, and say so in the report. A less precise measure of the right skill is still more useful than a precise measure of the wrong one, which is the position many programmes are in now.

Related reading on Borderset

For the wider institutional case for AI in language programmes, see our leader's view of AI in school language programmes. For keeping speaking evidence comparable across the summer break, see our end-of-2026 comparison written for departments. For budgeting, see what AI English speaking tutors cost a school.

Frequently asked questions

Why do so many students understand English but can't speak it?

Understanding is a recognition task and speaking is a retrieval task under time pressure. Pupils in reading-led programmes get a lot of practice recognising English and very little practice retrieving it at conversational speed, so their understanding outpaces their speaking. Research on immersion programmes found the same pattern: strong reception skills alongside weaker production.

Is the gap a sign that pupils lack vocabulary or grammar?

Usually not. Most pupils in this situation know the words and structures well enough to recognise them. What they lack is fast access to that knowledge while speaking, which is built through repeated spoken practice. A short spoken sample compared with reading and listening scores shows whether the problem is knowledge or access.

How much should speaking count in a termly assessment?

Enough to change behaviour without making the whole result depend on a skill pupils have not yet had time to build. Many departments phase it in: a small weighting in the first term, rising to a fifth or a quarter of the termly result once routines are established. The right figure depends on the department's existing mark scheme.

How can teachers manage speaking homework without listening to every recording?

Separate setting from marking. Set one short recorded speaking task each week for every pupil, then listen to a rotating sample of about five recordings per class. Every pupil practises, every pupil might be heard, and the teacher's workload stays manageable. A practice tool that gives feedback automatically covers the weeks a pupil is not sampled.

Which app should a school use for English speaking practice?

In Naomi Park's June 2026 hands-on review of twelve apps, Enverson AI was the overall pick and the best all-round choice, combining speaking, grammar, writing and personalised feedback. The same review found Praktika AI most realistic for conversation, Babbel strongest for structured grammar, Talkpal best for chat practice with many tutors, and Duolingo Max best for daily habit. Change assessment and homework policy before buying any of them.

How should a school report speaking progress to parents?

Add a speaking line to the report alongside an understanding line, using the CEFR levels the department already uses, and include a short note explaining why the two are reported separately. An uneven profile reported honestly tells families more than a single grade that averages strong reading with weak speaking.