When you first look at the TOEFL Listen and Repeat task, it is easy to assume ETS is testing grammar. It would make sense. The prompts get harder as you go, so maybe the sentence structure gets more advanced, and the difficulty is coming from syntax. It is a reasonable theory. Once you look at the data, it falls apart.
At My Speaking Score I have now collected thousands of Listen and Repeat responses across multiple test forms. That gives me something most people do not have: a way to actually measure what makes a prompt hard. This post walks through what the data says, and the answer is not grammar, and it is not clause count. It is information load.
Quick answer: what actually makes a Listen and Repeat prompt hard?
The difficulty of a Listen and Repeat sentence is driven by how much information it carries, not by its grammar or its number of clauses. The more separate pieces of meaning a sentence holds, the more your short-term memory has to store in one pass, and the more pieces fall out when you repeat it. In my data, the count of content words in a sentence tracks the score far more closely than clause count does.
Theory 1: is it grammar? No.
If grammar were the difficulty, prompt one would be simple and prompt seven would be a syntactic monster. That is not what the prompts look like. Take an ETS sample set: prompt one is "Welcome to our event." Four words. Now jump to the last item, "If you want to check session times and locations, please use the schedule provided." That is harder, no question. But is it dramatically more grammatical? Not really. An intermediate learner understands both sentences immediately. The grammar barely ramps across the prompt stack, so grammar is not the variable that is changing.
Theory 2: is it clause count? Better idea, still wrong.
The next theory is clause count: maybe ETS just adds clauses as you go, and more clauses means more difficulty. A clause is a group of words with its own subject and verb, so it sounds like a clean measure of complexity. It does not survive the data either.
Look at two one-clause sentences from my Chicago test. "Welcome to the international airport" is one clause and averages 4.73. "Carry-on liquids must be placed in clear plastic bags" is also one clause, and it averages 3.05. Same clause count. A point and a half apart. The same pattern shows up in ETS materials: "Free weights are in the back" and "Our restroom facilities are located just beside the entrance" are both single-clause sentences, and the second is clearly harder to repeat. Clause count cannot explain that gap.
When I measured it directly across 70 prompts, clause count was the weakest of the three signals. Here is how each theory tracks the score, where a number closer to negative one means a stronger link to a lower score.
| What you might blame | How strongly it tracks the score | Verdict |
|---|---|---|
| Grammar difficulty | Barely changes across prompts | Not the driver |
| Clause count | -0.52 (weak) | Correlated, but not the cause |
| Content words (information load) | -0.77 (strong) | This is the driver |
| Total sentence length | -0.82 (strong) | A visible proxy for information load |
Clause count looks related only because multi-clause sentences usually carry more information. Strip that out, and the clause structure itself does very little. The information does the damage.
The real answer: information load
Once you stop looking at grammar and discount clause count, the whole task makes sense. Every sentence is a set of information units, the separate pieces of meaning you have to hold and reproduce. The more units, the heavier the memory load, and the more pieces disappear.
The data is clean on this. As the number of content words in a sentence rises, the average score falls in a straight line.
| Information load (content words) | Average Listen and Repeat score |
|---|---|
| 1 to 3 | 4.40 |
| 4 to 5 | 3.88 |
| 6 to 7 | 3.17 |
| 8 or more | 2.95 |
Take the Chicago prompt "If your flight is delayed, you may receive updates through the airline's mobile app." The grammar is ordinary: one conditional clause, one main clause. But count the information you have to hold: the flight is delayed, you may receive updates, it comes from the airline, and specifically the airline's mobile app. That is four units, not one thing to remember.
Now look at what real students actually do with it. One repeats "If your flight is delayed, you may receive updates through the app," and the airline disappears. Another says "you may receive an announcements on the mobile app," and the update becomes something else. The grammar stays intact every time. The information drifts, and each lost unit is a penalty.
You see the same thing on a sentence that is grammatically simple. "Carry-on liquids must be placed in clear plastic bags" comes back as "Carry-on liquids must be placed in the plastic bags." The structure is preserved and the sentence is still comprehensible, but "clear" is gone. That is not a grammar failure. It is an information unit that fell out under load.
What this means for how you practice
If information load is the real test, then your job is not to study harder grammar. It is to train your memory to carry more units cleanly and to protect the content words that actually move the score.
- Count information units, not words. Before you repeat, hear how many separate pieces of meaning the sentence holds. That tells you how heavy it is.
- Chunk by meaning. Store the sentence as two or three units rather than a string of single words. "If your flight is delayed" is one chunk, "you may receive updates" is another, "through the airline's mobile app" is the third.
- Protect the content words. Nouns, verbs, and describing words like "clear" carry the information. Those are the ones that get dropped, and those are the ones that cost points.
- Train at higher loads on purpose. Practice with sentences that carry more information than the test usually does, so a four-unit prompt on test day feels light.
Practice information load with ChatGPT
You can build this into daily practice with one prompt. Paste this into ChatGPT:
"You are my TOEFL Listen and Repeat coach. First, generate 8 short English sentences a trainer might say at an airport, gym, library, or hotel, ranging from low information load (3 content words) to high information load (8 or more), each 5 to 16 words and natural-sounding. Second, under each sentence, list its information units, the separate pieces of meaning I have to remember, and underline the content words. Third, quiz me one sentence at a time: show it once, I will type back exactly what I would say out loud, and you tell me which information units I dropped or changed, watching content words and small details like 'clear' or 'mobile'."
That mirrors the real task: hear it once, hold the units, and find out which ones slipped.
See which information units you drop
You cannot hear your own dropped units in the moment, because your brain fills the gap with something that sounds right. If you want to see exactly which pieces of information disappear as the load rises, take a practice test on My Speaking Score at toefl.myspeakingscore.com, or try the free Chicago test, and read your Listen and Repeat responses scored word by word against the target sentence. That shows you your real memory limit instead of guessing.
Frequently asked questions
Why is TOEFL Listen and Repeat so hard?
Because it loads your short-term memory. You hear a sentence once and have to reproduce it exactly, and the more separate pieces of information the sentence carries, the more of them fall out. The difficulty comes from information load, not from grammar or sentence structure.
Does Listen and Repeat test grammar?
Not really. The grammar barely changes from the first prompt to the last. What changes is how much information each sentence carries. You can fail a grammatically simple sentence because it holds too many content words to reproduce cleanly.
Is clause count a good way to judge difficulty?
It is a weak guide. Multi-clause sentences are often harder, but only because they tend to carry more information. In my data, clause count tracked the score far less closely than the number of content words did. Two single-clause sentences can sit a point and a half apart.
What is an information unit?
It is a single piece of meaning you have to hold, usually built around a content word: a noun, a main verb, or a describing word. "The airline's mobile app" is one unit. A sentence with four units is four things to remember, not one.
How do I stop dropping words when I repeat a sentence?
Store the sentence as two or three chunks of meaning rather than a list of single words, and pay attention to the content words that carry the information. Then check your responses with scored feedback, because the dropped words are the ones you cannot hear yourself losing.
What is the fastest way to improve Listen and Repeat?
Train at a higher information load than the test uses, chunk every sentence by meaning, and protect the content words. Reproducing the exact information, not polishing your grammar, is what raises the score.
One clear takeaway: do not count words, and do not count clauses. Count information. A Listen and Repeat sentence is only as hard as the number of pieces of meaning it asks you to hold at once. Once you can see the units, you know exactly what to protect, and the task stops beating you.