TOEFL Speaking Exercises: 150 Drills You Answer Out Loud

Three tracks of TOEFL speaking exercises

Two of them mirror the real TOEFL speaking tasks question for question. The third exists because you cannot fix your pace and your pausing while you are also inventing what to say. Pick one and you go straight into it.

You are probably already practising with a general AI. Here is what it cannot do

Plenty of students rehearse speaking with a general assistant, and it is genuinely useful for that. It will happily generate questions and tell you your answer sounded good. What it will not do is measure.

The last 15 exercises are harder than the real test

That is the last 15 in every track, so 45 exercises in total. Each track climbs through four bands. The first three stay inside real exam parameters. Hard Mode is harder than the exam on purpose, so that test day feels like a step down rather than a step up.

What 59,123 timed answers say about a TOEFL speaking score

Word count came out as the strongest predictor in the whole set, at a correlation of 0.49 with the score the answer received. The curve is steep and it is close to monotonic. Answers of 0 to 30 words average 0.99 out of 5. Answers of 90 to 110 words average 2.82. Answers of 110 to 130 average 2.97. Answers of 130 words and up average 3.20. Nothing else in the data moves a score that far. If you stop talking at second 30 because you have run out of things to say, that is the first thing to fix, ahead of grammar and ahead of vocabulary, because everything else you might improve is being applied to half an answer.

Pace follows word count closely, at a correlation of 0.35. High scoring answers run at about 137 words per minute. Low scoring answers run at about 112. The gap is 25 words a minute, which over a 45 second window is roughly 19 words, and 19 words is often the difference between a claim with an example attached and a claim left hanging. Note what 137 words per minute actually means: in a 45 second answer it produces about 103 words, which lands inside the target band our own data says scores best. It is a brisk conversational speed, not a race.

Silence is the finding that surprises most students, because it does not behave the way the advice on the internet says it does. It is U-shaped, not linear. Answers that were 15 to 25 percent silence scored best, at 2.83. Answers with almost no silence at all, 0 to 15 percent, scored lower at 2.67. Answers that were 45 percent silence or more collapsed to 1.78. Speaking without a break does not read as fluent. It reads as rushed, and it usually means you are reciting rather than thinking. The goal is not to eliminate pausing. It is to move your pauses to the ends of clauses, where a listener expects them, and out of the middle of phrases, where they leave the listener holding an unfinished sentence.

Filler words barely register. The correlation between hesitation sounds and the score is -0.10, which is close to nothing. Answers with one or two fillers averaged 2.60, very slightly higher than answers with none at all, which averaged 2.57. Only at six or more does anything measurable start to cost you. A student who spends a week trying to stamp out every um is optimizing a variable with almost no weight while the one variable that carries half the signal, how much you said, goes untouched.

The failure rate tells you where to spend your time. Fifty-five percent of interview answers in the corpus score 2 out of 5 or below, and the most common reason is simply not saying enough. On the repetition task, forty percent of attempts score 2 or below, and almost all of that is the difficulty of holding a whole sentence in memory rather than anything to do with pronunciation. Those two numbers are why this product exists, and why its report screen is built around length, pace and pausing instead of a single band.

Three tracks of TOEFL speaking exercises, and why one of them is not a TOEFL task

The 2026 TOEFL iBT Speaking section runs about 8 minutes and contains 11 items worth 55 raw points: 7 Listen and Repeat items, then 4 interview questions at 45 seconds each. Two of the three tracks here map onto those task families one for one. The third does not, and it is deliberate.

Listen and Repeat is 50 exercises of 10 items each. You hear a sentence once and repeat it back, with the text hidden until after your attempt, exactly as the real task works. Sentences start at 6 to 10 words in the early exercises and climb to 17 to 26 words at the top, with the recording window moving from 8 seconds to 15 as the sentences grow. What the drill actually trains is working memory under a single hearing, which is where the failures are. Students who assume their repetition score is an accent problem usually find, once they see which words dropped out, that they lost the back half of a sentence they never fully held.

Interview is 50 exercises of 10 questions each, always on a 45 second clock, because shortening the answer window would make the drill easier rather than harder and the whole problem this product exists to fix is students not filling the window. The difficulty ladder moves thinking time instead. Preparation runs 10 seconds in the first eight exercises, 5 seconds through the middle, 3 seconds at test pace, and 0 seconds at the top, which is what the real section gives you anyway. Each exercise carries a target word range taken straight off the measured curve: 70 to 100 words while you are warming up, 90 to 130 once you are building, and 110 to 140 in the hardest band.

Read Aloud is 50 exercises and it is not a TOEFL task at all. A sentence appears on screen and you read it out loud, with windows of 13 to 22 seconds depending on length, and sentences climbing from 8 to 14 words up to 20 to 34. It is here because of what the data above says. Word count and pace are the two things that carry the score, and a student cannot fix pace and pausing while simultaneously inventing an argument against a running clock. Read Aloud removes the composition load so the delivery habit can be built on its own. Interview then puts the load back. If you only have time for one track before a test date, do Interview, since that is where the marks are lost. If you have three weeks, start on Read Aloud for the first few days and you will arrive at Interview already speaking at a usable rate.

The four interview question types, and what a full answer looks like

Every interview exercise rotates through all four question shapes so you are never drilled on just one: personal recall, preference, agree or disagree, and prediction. That rotation is a design decision backed by the corpus. Average scores across the four types run from 2.21 to 2.43, a spread small enough that no single type is anyone's real weakness. Students who believe they are bad at opinion questions are usually just as short on personal recall questions. The problem is length, and length is type independent.

A full answer has the same skeleton whatever the shape of the question. Say your position in one sentence so the listener knows where you are going. Give one reason for it. Then give a specific example from your own life, with a name, a place, a number or a date in it. A bare opinion sits at the bottom of the scale. An opinion with a reason sits in the middle. An opinion, a reason and a concrete example is what the top looks like, and it is also the easiest honest way to fill 45 seconds, because a real example has detail in it and detail is what generates words.

If you want a number to aim at, 103 words is what 45 seconds at 137 words per minute produces, and that sits comfortably inside the 90 to 130 band the data says scores best. Count it once on a written answer so you know what 100 words feels like out loud. Most students are shocked at how little it is. Roughly four sentences of normal length.

The difficulty ladder, and why the last 15 exercises are harder than the real test

Each track climbs through four named bands. Exercises 1 to 8 are Warm Up: short, concrete, everyday language, built to establish the habit of speaking without stopping. Exercises 9 to 20 are Building: longer sentences and less familiar topics, which is where pace starts to matter. Exercises 21 to 35 are Test Pace: real exam length and topics on the real clock. Exercises 36 to 50 are Hard Mode, which is past what the exam uses, on purpose. That is 15 Hard Mode exercises in every track, 45 in total.

Hard Mode raises three levers at once. Sentence length goes beyond the range the real section uses. The prompt voice speeds up, by 12 percent on Listen and Repeat and 10 percent on Interview, which on a repetition task is the single biggest difficulty lever there is. And thinking time drops to zero. The interview questions get more abstract at the same time, and the target answer length rises to 110 to 140 words. The point is not to be unfair. The point is that if your practice ceiling is the exam, then on test day you are operating at your ceiling with adrenaline on top. If your practice ceiling is above the exam, test day is a step down.

Scoring on each attempt is a percentage, and the star cuts are fixed and public so a retake means something: 85 and up is three stars, 70 is two, 50 is one. Only your best attempt at an exercise is kept, so a bad retake can never cost you anything. Progress accumulates as experience points across eight levels, from Getting Started up to Command, weighted by both exercise difficulty and score so that repeating exercise 1 forever is worth far less than clearing exercise 40 once.

What gets measured, and what we refuse to pretend we measure

Measured, on every attempt: how much you said, your pace in words per minute against the 137 that strong answers average, how much of the answer was silence, how long your longest single stop ran, how long you took to start speaking, and on the two sentence tracks, how much of the sentence actually landed, word by word. Where your pauses fell is also reported, split by whether they came at a clause boundary or inside a phrase, since published research is clear that fluent speakers pause at clause ends about as often as learners do and that the extra pausing a learner does is almost all mid sentence. That one is shown to you but kept out of the score, because the thresholds behind it come from research rather than from our own measured corpus, and we hold scored components to the corpus.

Not measured: pronunciation and accent. Doing that honestly needs sound by sound analysis that our speech recognition does not produce, so any accent score we displayed would be invented. On the two sentence tracks we do check which words the recognition heard, so speech that is very hard to make out can still cost you there, but that is a word match and not a verdict on how you sound.

Also not treated as faults: a handful of filler words, and silence itself. The data above is the reason. One or two hesitation sounds score marginally higher than none, and 15 to 25 percent silence scores better than near continuous speech. Most speaking advice online tells you the opposite on both counts. We checked our own numbers and they disagree, so we score what the numbers support and we tell you which parts we are leaving alone.

How to practice speaking for the TOEFL iBT in the weeks before test day

Give it weeks rather than days. The research on speech recognition based speaking practice is blunt about the timeline: programs running under four weeks produced almost no lasting effect, while programs running five to eight weeks produced a large one. Twenty minutes on most days beats one long session the night before by a margin that is not close. A single exercise is ten questions, which is about two minutes on Read Aloud and about ten on Interview once you include the clock, so a twenty minute session is two or three exercises.

Do not grind one exercise. Lambert, Kormos and Minn (2017, Studies in Second Language Acquisition) had 32 learners perform three task types six times each and found speech rate gains were largest across the first three performances, continued to about the fifth, then flattened. The learners themselves judged three to four repetitions optimal. Shiki and colleagues report the same plateau at about five for repeating the same script. Retakes here are unlimited and always will be, but once you have run one exercise five times the next exercise is worth more to you than a sixth attempt at this one, and the report screen says so.

Spread the repetitions out rather than stacking them. Suzuki and Hanzawa (2022, Studies in Second Language Acquisition) compared massed against spaced repetition and found massed repetition cut pausing the most in the moment, but produced slower articulation and more verbatim repetition, and no scheduling effect survived to a delayed post test a week later. In practice: three exercises today and three tomorrow will serve you better than six in one sitting, even though six in one sitting will feel more productive while you are doing it.

A workable four week shape looks like this. Week one, Read Aloud exercises 1 to 8 plus Interview 1 to 4, so the delivery habit starts forming while you meet the question types. Week two, Listen and Repeat 1 to 12 and Interview 5 to 12, and start watching your word count rather than your star rating. Week three, Test Pace on all three tracks, exercises 21 to 35, which is where the prep time drops to 3 seconds and the sentences reach real exam length. Week four, Hard Mode on whichever track your reports are weakest on, and one full timed speaking test at the end of the week to see whether it moved.

Where these exercises sit next to a TOEFL speaking practice test

They do different jobs and you want both. A speaking practice test is one timed run through the whole section, sat once, producing a band and an expert evaluation of what you produced. It is a measurement. These exercises are the drilling: ten items on a single task type, repeatable as often as you like, with only your best attempt kept and a report about delivery rather than a band. A measurement tells you where you stand. A drill is what changes it.

The loop that works is straightforward. Sit a speaking practice test to find out which task family is costing you, drill that family here for two or three weeks against your own word count and pace numbers, then sit a different test and compare. If your interview word count has moved from the 60s into the 100s and your pace has come up toward 137, the band will usually have followed. If you want the underlying theory instead of the reps, the TOEFL Speaking course cover how the section is scored, the interview framework, and what to do when you lose your thread halfway through an answer.

These are the speaking drills. Reading, Listening and Writing have their own, built the same way and covering all ten of the task types the 2026 test uses: the full set of TOEFL practice exercises runs one task type at a time, twenty levels each, with every answer explained. Level 1 of every task type is free, the same as exercise 1 here.