Speaking

Speaking Sample

One prompt, thirty seconds to think, up to three minutes to speak. The longest continuous English you will produce on the test — and part of what institutions actually watch.

  • Up to 3 min on the clock, Once per full test
  • Feeds your Speaking subscore
  • Built for the current July 2025 DET format, including the new Interactive Speaking task
Speak for up to 3 minutes Describe a decision you made that turned out better than you expected. 1:47 / 3:00
Up to 3 minon the clock
Onceper full test
Speakingsubscores fed
Harddifficulty to fix

The Speaking Sample gives you a prompt, about thirty seconds to prepare, and up to three minutes to answer. It appears once, near the end of the test.

Three minutes is a long time to speak without notes and without a conversation partner to give you something to react to. Most candidates have never done it in a second language, and the difference between people who have practised it and people who have not is immediately audible.

It is also different from the rest of the test in one important way: alongside the Writing Sample, it forms part of what an institution can view with your results. Your other responses generate a score; this one is heard.

The screen

What you actually see

A written prompt, a preparation countdown, then a recording indicator and a timer counting up to three minutes.

  • An open-ended prompt inviting an extended personal answer — an experience, an opinion, a description, a decision.
  • The prompt stays visible while you speak.
  • About thirty seconds of preparation.
  • Up to three minutes of recording. You may finish earlier.
  • One appearance in a full test.
Worked example

A structure that survives three minutes

Prompt: "Describe a decision you made that turned out better than you expected."

Three minutes is roughly 350–450 words of natural speech. Improvising that much without a shape is how answers collapse into repetition at ninety seconds. This structure does not:

  1. The decision, stated plainly (20s). What it was and when. No preamble — "About two years ago I turned down a job in my home city and moved."
  2. The context that made it hard (40s). Why it was not obvious. This is where you get to use past perfect, concession, and comparison naturally.
  3. What actually happened (60s). The narrative core. Concrete detail, in sequence. Most of your fluency marks are earned here simply because you have plenty to say.
  4. Why it exceeded expectations (40s). The reflective part — conditionals and counterfactuals live here. "If I'd stayed, I'd probably still be…"
  5. What it changed about how you decide now (20s). A generalisation that lands the answer instead of letting it fade out.

Five sections, none longer than a minute. Each one has a job, so you always know what comes next, and the grammatical range distributes itself naturally across them rather than needing to be deployed deliberately.

Scoring

How Speaking Sample is scored

The Speaking Sample feeds Speaking, and so Conversation and Production. Because it is the longest speaking response, it provides more evidence than any other single speaking item.

Assessment covers fluency and coherence, the range and accuracy of your grammar and vocabulary, intelligibility, and whether the answer actually addresses the prompt. That last one is not a formality. In our own grading engine, a response that drifts off the prompt is capped regardless of how good the English is, because rewarding fluent irrelevance would make the score meaningless — and the published DET criteria treat task relevance the same way.

Length interacts with this. Speaking for the full three minutes is not required and padding is not rewarded, but a forty-second answer to a three-minute prompt gives very little to assess and tends to score below the speaker's actual level.

What the scorer looks atWhy it matters
Relevance to the promptGates everything else. Drifting off-topic caps the response however fluent it is.
Sustained fluencyThree minutes without collapse. Long silences, repetition and restarts are what the length exposes.
Range of grammar and vocabularyThe longer format is an opportunity: narrative, conditional and comparative structures all become available.
Coherence and organisationWhether the answer has a shape a listener can follow, or is a sequence of loosely related remarks.
IntelligibilityClear, followable speech. Accent is not the criterion.
Mistakes

The mistakes that cost the most

The characteristic failure is not linguistic. It is running out of material at ninety seconds and spending the next ninety saying the same thing again in different words.

Do this

  • Use the preparation time to choose a structure with sections, not to draft sentences.
  • Pick a concrete, specific experience over a general opinion. Specifics generate detail; abstractions run dry.
  • Signpost as you go — the thing that surprised me was, looking back. It buys thinking time and demonstrates discourse control.
  • Let the reflective section carry your complex grammar. Conditionals and counterfactuals arrive naturally there.
  • Finish deliberately with a closing generalisation rather than trailing off.

Not this

  • Do not answer a prompt you would rather have been asked. Relevance is assessed and drifting is the most costly error available.
  • Do not recite a prepared answer. Rehearsed material rarely fits the prompt, and the mismatch is audible.
  • Do not stop at forty seconds. There is not enough language there to score your actual level.
  • Do not repeat yourself to fill time. Padding dilutes the average quality of everything you said.
  • Do not apologise or narrate your difficulty. It spends seconds and demonstrates nothing.
Practice

How to actually get better at Speaking Sample

Three minutes of continuous monologue is a specific physical skill. Practising two-minute answers does not prepare you for it, because the collapse happens after two minutes.

  1. Speak for a full three minutes, twice a week, without stopping

    On any prompt. The point is time under load. The last minute is the part that improves, and you only reach it by going the distance.

  2. Learn one flexible structure and reuse it

    The five-section shape above adapts to almost any personal prompt. Not having to invent a structure under thirty seconds of pressure is most of the benefit.

  3. Prepare experiences, never scripts

    Six or seven real events you can describe in detail will cover most prompts. They adapt; memorised answers do not, and the mismatch costs more than the preparation saved.

  4. Record and count your dead air

    Silences over two seconds, and repeated phrases. These are countable, they correlate with the fluency judgement, and they improve when tracked.

  5. Practise the reflective section deliberately

    It is where the highest-value grammar lives and it is the section most people rush. Drill answers to "why did that matter" and "what would have happened otherwise".

Leverage

How much this one task moves your score

One appearance, but the longest speaking response on the test, feeding Speaking, Conversation and Production.

Its weight is larger than its frequency suggests, for two reasons. It supplies more evidence than any other speaking item, and it is one of the two responses an institution can actually review. A score is an abstraction; a recording is not.

If your Speaking subscore is behind your Reading one — a very common pattern, because reading is what most language study consists of — this task and Listen, Then Speak are where the difference is made. Start with Read Aloud to build delivery, then bring it here.

FAQ

Speaking Sample: questions people ask

Do I have to speak for the full three minutes?

No, and padding is not rewarded. But a very short answer gives too little language to assess and typically scores below your real level. Develop the answer properly — most people stop early because they ran out of structure, not because they had finished.

Can institutions listen to my Speaking Sample?

Yes. The Speaking Sample and the Writing Sample are shared with the institutions you send your results to, alongside your scores. It is worth writing and speaking with the knowledge that a person may review them.

What happens if I go off-topic?

It costs you heavily. Relevance to the prompt is assessed, and a fluent answer to the wrong question scores as a poor response. If you realise mid-answer that you have drifted, steer back explicitly rather than continuing.

Should I prepare answers in advance?

Prepare experiences and a structure, not scripts. A memorised answer rarely fits the prompt you get, and forcing it produces exactly the off-topic drift that caps the score. Six or seven real experiences you can describe in detail will cover most prompts.

How do I stop running out of things to say?

Use a sectioned structure decided during preparation — the decision, the context, what happened, why it surprised you, what it changed. Knowing what comes next is what prevents the ninety-second collapse, and it distributes your grammatical range across the answer automatically.

What kind of prompts does the Speaking Sample ask?

Open questions about your own experience and opinions — a decision you made, something you would change, a view you hold — on everyday topics. None require specialist knowledge, so you can always answer from your own life rather than searching for facts you do not have.

Am I allowed to make notes during the preparation time?

No scratch notes — the thirty seconds is for choosing an experience and a rough structure in your head, not for writing anything down. Deciding your three or four sections in that window is what keeps the answer from collapsing halfway through.

Is the Speaking Sample scored, or only shared with institutions?

Both. It is assessed as part of your Speaking performance like the other spoken tasks, and it is also sent to the institutions you choose for a person to review. Treating it as a low-stakes afterthought at the end of the test is a mistake on both counts.

How long is the Speaking Sample?

Up to three minutes of recording, with about thirty seconds of preparation beforehand. It appears once, near the end of the test.

Can I see the prompt while I speak?

Yes, it stays on screen. That removes any memory burden and means drifting off-topic is a structural failure rather than an understandable slip.

Is three minutes of speaking really expected?

Not required, but the length is the point of the task. Most people run out of material at around ninety seconds, and a sectioned structure decided during preparation is what carries you through the second half.

What if I have no experience matching the prompt?

Adapt one you do have, or answer hypothetically — "I have not faced that exactly, but something close was…". Nobody verifies your life. What is assessed is the English and whether it addresses the prompt.

Should I use advanced vocabulary to score higher?

Use precise vocabulary, which is not the same thing. A misused rare word costs more than a correctly used common one, and strained vocabulary usually comes with the hesitation that damages fluency.

Does hesitation matter across three minutes?

Yes, and it is measurable. Silences over about two seconds and repeated phrases are the concrete things that separate a fluent-sounding answer from a struggling one, and counting them in your own recordings is the fastest way to reduce them.

Who actually listens to the Speaking Sample?

Institutions receiving your results can review it alongside your scores. It is one of only two responses on the whole test that a person may hear or read rather than simply see a number for.

In context

Where Speaking Sample sits in a full test

No task on the DET is scored in isolation. The adaptive engine is building one picture of your English out of every response, so it helps to know how much of the hour this particular task accounts for and which of the others are drawing on the same ability.

The table below is the whole test. Your current page is highlighted.

Timings and frequencies follow the published Duolingo English Test format. Exact counts vary between sittings because the test adapts.
Question typeTimeHow oftenSubscores it feeds
Read & Complete3 minutes3–6 timesReading, Writing
Read & Select5 seconds15–18 timesReading
Fill in the Blanks20 seconds6–9 timesReading, Writing
Listen & Type1 minute6–9 timesListening, Writing
Speak About the Photo90 secondsat least onceSpeaking
Interactive Speaking35 seconds each6–7 questionsSpeaking, Listening
Write About the Photo1 minuteat least twiceWriting
Interactive Reading7 minutes2 setsReading
Interactive Listening6 minutes2 setsListening, Speaking
Speaking Sample — you are here3 minutesonceSpeaking
Writing Sample5 minutesonceWriting

Two things are worth reading off that table. The first is that the tasks feeding Speaking are not only this one — improvement transfers, so time spent here shows up elsewhere. The second is that the tasks taking the largest share of the hour are not the ones most people practise most.

Start a practice test

An hour under real conditions, all eight subscores, and feedback on every written and spoken answer.

Take a test
SCORE REPORT 155 C2 of 160 Reading 160 Listening 160 Writing 150 Speaking 150 FEEDBACK 8 SUBSCORES