TOEFL Speaking Practice: Take an Interview, Four Questions at Forty-Five Seconds Each

As of September 2026, TOEFL iBT Speaking has two task types: Listen and Repeat, where you hear a sentence and repeat it back, and Take an Interview, where you answer a set of short questions about an everyday or campus topic. ETS's own 2026 test blueprint puts Take an Interview at four items out of the Speaking section's 11 total, with Listen and Repeat making up the other seven, and lists the whole section at about 8 minutes of testing. The full TOEFL iBT format and timing breakdown shows how those 8 minutes sit alongside Reading, Listening, and Writing.

The blueprint describes the skill being measured as the ability to "speak spontaneously and meaningfully in response to questions in an interview format," scored on "clear, coherent elaboration using accurate grammar, varied vocabulary, and intelligible prosody." The word spontaneously is doing real work there: this task is not scored on a rehearsed explanation, it is scored on what you can produce the instant a question lands.

That is a different skill from the pre-2026 Speaking tasks, which are gone as of the January 21, 2026 test change. The old independent question gave you 15 seconds to plan before 45 seconds to speak. The old integrated tasks gave you a reading passage or a lecture to take notes on before you spoke. Take an Interview gives you neither. The question plays or appears once, and the response window starts immediately; there is no scratch-pad phase to organize your thoughts before you open your mouth. Practicing with old TOEFL speaking material trains the wrong reflex: it trains you to plan first and speak second, when the current task rewards speaking and organizing at the same time. Writing changed in the same rollout: the TOEFL Write an Email task replaced the old 20-minute Integrated Writing essay with a short, register-sensitive response.

Test-prep resources that have documented the January 2026 rollout consistently report a 45-second response window per question, with no preparation time before each one starts. ETS's public pages confirm the four-question item count and the zero-second preparation framing directly; the 45-second figure specifically is not published in ETS's own blueprint document, so this drill treats it as a well-corroborated working estimate rather than an official number, and says so here rather than presenting it as confirmed.

Each set combines one topic frame with four of six possible question shapes, chosen at random but reproducibly: the whole set is built from a single numeric seed shown at the top of the drill, so entering that same seed into the "seed to replay" box and pressing Load Seed regenerates the identical four questions, in the same order, every time. Nothing is looked up from a stored bank of real TOEFL items; every question is assembled on your device, in your browser, from the two lists below.

Length and pace are two views of the same raw number, because the window is fixed at 45 seconds: 85 to 115 words in 45 seconds works out to 113 to 153 words per minute, which is why the two bands line up closely (the small gap comes from rounding each band to a clean, easy-to-remember number rather than a decimal). Structure checks for three specific marker phrases: an opening commitment in your first eight words ("I think," "I would," "the best," "for me," "definitely," "my choice"), a reason marker anywhere in the answer ("because," "since," "the main reason," "that is why"), and an example marker anywhere in the answer ("for example," "for instance," "last year," "when I," "at my school"). Specificity looks for one concrete anchor: a number, a capitalized word that is not the first word of your answer (a name, a place, a course code), or a time marker like "last semester" or "every morning." Fillers counts six specific tokens; it will not catch every disfluency, and it will catch "like" even when you meant it as a verb ("I like my roommate"), which is a known limitation, not a bug.

Two worked transcripts from testing this page, typed at exactly 45-second-window length, show how the meters interact:

The structure meter only checks your first eight words for a commitment marker, and that is a fair reflection of how a listener processes speech: the first sentence sets the frame for everything after it. An opening like "so, um, this is kind of a hard question" burns three or four seconds of your 45 and tells the listener nothing about where you stand. An opening that commits immediately, "the best part of my routine is X, because," gives the listener a position to track for the rest of your answer, and gives you a reason clause to fall into next without having to invent a new sentence structure mid-response.

The filler meter counts six tokens: um, uh, standalone like, you know, basically, and actually. It will undercount real disfluency, because people who type up what they said tend to quietly delete their own "ums" without noticing, which is one reason to record audio at least once a week (see the last section on this page).

Restarting a sentence costs more than finishing an imperfect one. A false start, "the best part is, well, actually the best part is," burns three or four words and resets your listener's attention without adding any content; an imperfect but complete sentence, "the best part is, I think, the schedule," keeps moving and still lands on a reason. Because pace and length are locked together in a 45-second window, every restart is words you cannot get back: there is no partial credit for the discarded half-sentence.

To buy half a second without a filled pause, use a real word that also stalls: repeat the last content word ("the best part… the best part of the schedule is…"), or drop in a genuine transition phrase that still moves the sentence forward ("the thing is," "what I mean is"). Both read as fluent hesitation rather than a filler token, and both keep your mouth producing intelligible speech instead of a filled "um" while your brain catches up.

Natural conversational English runs roughly 110 to 150 words per minute. Translate that into a fixed 45-second window (0.75 minutes) and you get a word target directly: 110 × 0.75 = 82.5 words at the low end, 150 × 0.75 = 112.5 words at the high end. This drill rounds that to a clean 85-to-115-word band, which is the same range translated into a number you can count while typing rather than a decimal you would have to compute mid-drill.

Racing to fit more in past that band costs more than it earns. Once your pace crosses roughly 165 words per minute, which works out to about 124 words in a 45-second window, the drill flags "too fast to be intelligible," echoing the same "intelligible prosody" language ETS uses in its own subskill description for this task. Extra words past 115 rarely add new content anyway, since a 45-second interview answer only has room for one opening, one reason, and one example; words beyond that are usually restating the opening in different phrasing, which a rater or a listener notices as padding, not depth. If you consistently overshoot 115 words, the fix is not to talk faster, it is to cut the answer down to one clean example instead of two rushed ones.

Three sets a day is 12 questions, roughly 12 minutes of speaking practice, which is enough repetition to build the answer-first reflex without burning out on a drill that deliberately has no variety in format. Record one full set a week on your phone and play it back specifically listening for three things a transcript cannot show you: dead air after the question before you start talking, fillers you typed out of the transcript without noticing you said them, and whether your voice actually commits on the opening line or trails off even when the words on the page look committed. If your recorded self sounds less confident than your typed transcript reads, that gap is the real thing to practice closing. To slot these four weeks into a wider plan alongside Reading, Listening, and Writing, use the TOEFL study plan builder.

A text-based check has three specific blind spots. First, it cannot hear pronunciation, intonation, or prosody at all, and ETS's own subskill description for this task explicitly includes "intelligible prosody" as something that gets scored; a transcript that reads perfectly can still come from a delivery that a rater would mark down. Second, typing up what you said is not the same act as saying it, and people reliably clean up their own disfluencies when transcribing from memory, so your typed filler count is very likely lower than your spoken filler count. Third, the pace meter assumes you spoke continuously across the entire 45-second window; if you actually paused for eight seconds partway through and then talked fast to finish, the meter has no way to see that, it only sees the total word count.

Record actual audio at least once a week and compare it honestly against what you typed. The gap between the two, not either number alone, is what tells you whether the meters on this page are still describing your real speaking, or just your memory of what you meant to say.