Why You Keep Scoring 3 or 4 on TOEFL Speaking (Even When Your English Is Good)

The 2026 format changed everything. Here’s what’s actually costing you points.

https://www.pexels.com/id-id/pencarian/toefl/

There’s a pattern that shows up constantly in TOEFL prep communities.

Someone holds full conversations in English without breaking a sweat. Their grammar is decent. Their vocabulary is fine. They’ve been consuming English-language content for years. Then their TOEFL Speaking score comes back as a 3 or a 4, and they have absolutely no idea why.

This is not a rare situation. It is, honestly, one of the most common frustrations among test takers right now. And the reason it keeps happening has a lot more to do with how TOEFL Speaking works than with how good your English actually is.

Let’s unpack it.

First, a Quick Format Check

If you’re still preparing with materials from 2025 or earlier, stop immediately. The Speaking section changed completely on January 21, 2026, and old prep resources will actively work against you.

The entire previous four-task structure is gone. No more independent opinion task. No more campus situation listening. No more integrated reading-plus-listening tasks. All of it replaced with just two task types across 11 questions:

Listen and Repeat (7 questions): You hear a sentence, then repeat it exactly once. No text on screen. No preparation time. The sentences get progressively harder from items 1 through 7, starting with short simple phrases and building up to complex clauses with heavier information loads. You have 8 to 12 seconds to respond after each one.

Take an Interview (4 questions): A prerecorded interviewer asks you spontaneous questions on a single everyday or academic topic. You get 45 seconds to answer each one. Zero preparation time.

Neither task gives you any prep time. The whole section takes around 8 minutes. Every response is scored by an automated scoring engine (SpeechRater, developed by ETS), with human raters providing quality assurance oversight for flagged responses. Since January 2026, the AI scores first — a human only reviews your response if the system flags it as unusual.

Your final Speaking score comes back as a band from 1.0 to 6.0 in half-band increments, aligned to the CEFR scale, with 6.0 mapping to C2. A 3.0 sits around B1. A 4.0 is roughly B2. Most university targets sit at 4.5 or above.

So What Is TOEFL Speaking Actually Measuring?

This is where things get counterintuitive.

The scoring rubric for Take an Interview evaluates four dimensions: Fluency, Intelligibility, Language Use, and Organisation. Listen and Repeat adds a fifth — Repeat Accuracy — in place of Organisation.

Look at those dimensions carefully. Fluency is not vocabulary size. Intelligibility is not grammar accuracy. Language Use does include grammar and vocabulary, but notice what’s missing: there is no dimension called “how much English do you know.”

You can know a lot of English and still score a 3. You can have B2-level grammar and still score a 3. Because the things that actually tank scores are mostly behavioural habits, not knowledge gaps.

The Real Score Killers

1. Filler Words Are Getting Counted Against You

SpeechRater is an acoustic and linguistic engine. It listens to your response as audio, generates a transcript, and extracts measurable features from it. One of those features is filler word and disfluency frequency.

Every “um,” “uh,” “like,” “you know,” mid-sentence restart, and long mid-sentence pause gets tracked. The engine reads a high filler rate as someone who is retrieving language word by word rather than someone who has internalised it — regardless of how accurate the actual words are.

The fix sounds deceptively simple: replace fillers with brief silence. A clean half-second pause scores better than “uhhhh.” In practice this takes deliberate training, because most people’s instinct under pressure is to fill the silence. But if you can develop the habit of pausing quietly instead of reaching for a filler, your Fluency dimension score climbs fast.

2. You’re Using Maybe Half the Available Time

This one is widespread. In the Take an Interview task, test takers routinely burn through their response in 15 to 20 seconds and then… stop. Sitting in silence for the remaining 25 seconds doesn’t help you at all.

The scoring rubric explicitly rewards elaboration and development. One clear direct answer, supported by a specific reason and a concrete example, spoken for the full 45 seconds, will outscore three shallow points rushed through in 20 seconds every time. The structure doesn’t need to be complex — direct answer, develop it, give one example, wrap it up. That’s the entire framework. The goal is to fill the time with substance, not padding.

Think of the 45 seconds less as a time limit and more as a target.

3. Listen and Repeat Is Not a Warm-Up

Seven of your eleven questions are Listen and Repeat. That’s the majority of your raw score. And it’s the task that catches people the most off guard.

The instinct when you hear a sentence is to immediately repeat it as fast as possible while the audio is still fresh in your memory. This is the wrong move. Speed does not help your score. Precision does.

The scoring looks at pronunciation accuracy, rhythm, stress patterns, and overall intelligibility. Speaking faster than you can comfortably pronounce words clearly costs more points than a slightly measured pace. The task is genuinely trainable — but you have to drill it specifically. General English conversation practice does not prepare you for it. You need to practice hearing sentences and reproducing them accurately with controlled speech.

The progressive difficulty also trips people up. Items 1 and 2 feel easy, items 6 and 7 are genuinely hard. Don’t let the early softballs lull you into a pace that falls apart when the sentences get complex.

4. Repetitive Sentence Structures Cap Your Language Use Score

Using the same four or five connector phrases in every answer — “in my opinion,” “for example,” “on the other hand,” “that is why” — across multiple responses is something the scoring engine flags. The Language Use dimension rewards range and variety, not just accuracy.

Templates are genuinely useful for structure, especially early in prep. But if your answer to every question sounds structurally identical, you’re limiting yourself. Varying sentence structures and mixing in different linking expressions goes a long way on this dimension.

The Part Nobody Wants to Hear

Almost every case of “good English, low score” comes down to the same underlying issue: the test taker prepared for a general English challenge and showed up to a very specific performance test.

TOEFL Speaking in 2026 is not asking “how good is your English.” It is asking “can you produce fluent, organised, elaborated spoken English spontaneously under time pressure, and can you accurately reproduce spoken sentences with controlled delivery.” Those are learnable skills. They are also different from the skills you built by watching TV, holding conversations, or reading English content every day.

The good news is that because the score killers are specific and measurable, they’re also fixable. Filler words, short answers, rushed Listen and Repeat, repetitive structures — these are training problems, not knowledge problems. And training problems respond to training.

Platforms that offer SpeechRater-calibrated feedback (PrepEx, Magoosh, and a few others that specifically support the 2026 format) let you hear your own score through the same lens the actual test uses, which is the fastest way to identify exactly which dimension is dragging you down.

If you’re stuck between a 3 and a 5, the gap is almost certainly not about knowing more English.

Have you taken the new 2026 format? Did Listen and Repeat catch you off guard?