How fast should you speak English?
Ask how to sound more fluent and you get advice about intonation, linking, and filler words. All of it is true and none of it is measurable on a Tuesday night by yourself. Speaking rate is measurable, it needs nothing but a timer and a transcript, and it turns out to be the single most useful number an advanced learner can watch.
This post is about what the number means, where the useful thresholds sit, and the two ways it misleads you.
The bands
Conversational English between native speakers sits somewhere around 140 words per minute. Prepared speech — a talk, a lecture, an exam answer — runs slower, usually 110 to 150, because the speaker is choosing rather than reacting.
For a learner speaking continuously on an unfamiliar prompt, these are the thresholds worth knowing:
| Words per minute | What it usually means |
|---|---|
| under 90 | Retrieval is the bottleneck. You are composing sentences, not delivering them. |
| 90–110 | Working, but effortful. Long pauses at clause boundaries. |
| 110–130 | Comfortable. This is where most successful exam answers live. |
| 130–150 | Fluent delivery. You are choosing words rather than finding them. |
| over 150 | Fast. Fine if it is clear; often a sign of nerves. |
These are the bands Reframit uses, and they are calibration starting points rather than measurements of any population. The useful thing is not which band you are in today — it is that the band stops moving once your retrieval improves, which is a change you cannot feel from the inside.
Why a low rate is the first thing to fix
If you are under 100 wpm on a topic you know something about, almost nothing else you practice will stick. Intonation drills do not help someone who is still assembling the sentence. Vocabulary lists do not help someone who cannot reach the words they already have under time pressure.
What moves it is volume of speech under a clock. Not conversation — conversation halves your talking time and lets you wait for the other person to fill the silence. Uninterrupted monologue on a prompt you did not choose, repeatedly, is the thing that moves retrieval speed, and it is unpleasant enough that almost nobody does it without a structure forcing them to.
The two ways the number lies
It counts words, not thought. Someone who has memorized four template openings will post a high rate while saying nothing. Rate measures delivery, not content, and it is perfectly possible to be fast and empty. This is why it should never be the only number you look at.
It punishes the wrong kind of pause. A two-second pause before a genuinely better word is a good pause. A two-second pause because you have lost the thread is not. Rate cannot tell them apart. If your rate drops while your answers get sharper, the rate is wrong and you are right.
The number that goes with it
Rate pairs badly with a plain vocabulary count and well with a windowed one.
The obvious measure of range is the type-token ratio: unique words divided by total words. It has a fatal flaw for this purpose — it falls as you say more. Speak for thirty seconds and you might score 0.75; speak for three minutes on the same topic with the same vocabulary and you will score 0.55. The measure is reading length, not range, and the speaker who practices harder scores worse.
The fix is to take the ratio inside a fixed window — say fifty words — and slide it across the whole transcript, averaging as you go. That figure stops depending on how long you spoke, which makes it the first vocabulary number you can actually compare across two different sessions. It is called a moving-average type-token ratio, and it is the one Reframit reports.
Together they answer a question neither can answer alone: are you getting faster because you are more fluent, or faster because you have narrowed down to the words you can reach quickly?
What to do with it
Watch the number monthly, not per session. A single session's rate is mostly a reading of how well you slept and whether the prompt happened to land near something you know. Across a month it is a real signal, and it is one of very few that an independent learner can generate without paying anyone.
And when it stops moving, stop watching it. A measure that has done its job is a measure you can put down.
---
Reframit is an offline speaking drill for advanced learners: a prompt, a structure, a timer, and then the next one. What it is.