Word counter with a CEFR reading level

How to use the level estimate

The number at the top is only useful if you know what question it answers. It answers this one: at what level does a learner know enough of these words to get through this text?

If you are a teacher, this is the fastest way to check that a text you found is at the level of the class you found it for. Paste it in, look at the coverage bars, and read the off-list words. Those off-list words are your pre-teaching list — they are the ones your learners will not have met on any of our lists at any level, which usually means a handful of names, a few technical terms, and two or three words genuinely worth putting on the board.

If you are a learner, paste something you have written. If it comes back below your own level, you are writing under your ceiling — you know more words than you are using, which is extremely common and easy to fix. If it comes back above your level, look at which words did it. Half the time it is one unnecessary word where a shorter one was available, which is the definition of writing that is harder to read than it needs to be.

Either way, read the sentence length too. It is reported separately for a reason, and that reason is the whole limitation of the method, explained below.

What the CEFR estimate actually measures

This is a lexical coverage measure, and it is worth being precise about that, because "readability" is a word that gets used for a dozen different things.

Coverage is the share of the running words in a text that a reader already knows. Vocabulary research has converged on two useful figures: at roughly 95% coverage a reader can get through a text with occasional help, and at roughly 98% they can read it comfortably and unaided. Both numbers are reported here rather than one, because they answer different questions and because presenting a single threshold as if it were a law of nature would be overstating what is a well-supported convention.

So the tool takes every word in your text, finds the lowest level at which this site teaches it, and adds up the running total. The level it reports is the first one at which the total passes 95%. It is a count, not a model, and you can check it by hand if you disagree with it.

• Function words count as A1. The, of, and, was, which — the grammar words. They appear on no topic vocabulary list anywhere because they are grammar rather than vocabulary, and a learner meets all of them in their first term. Roughly 40% of any English text is made of them, which is why the chart on this page starts at 40% rather than zero.
• Endings are undone before giving up. Walked is looked up, missed, stripped to walk, and found. Plurals, -ing forms, comparatives, adverbs in -ly, the common contractions and the irregular verb table are all handled, so an inflected word is not counted as unknown.
• Names and numbers are set aside. A capitalised word in the middle of a sentence is treated as a name and left out of the calculation entirely. It is not vocabulary in either direction, and counting Barcelona as an unknown C1 word would make every travel text unreadable.
• Everything left over is "off-list". That means: not in the words this site teaches. It does not mean the word is hard.

The word list behind all of this is not a general dictionary. It is 5,290 words from our levelled vocabulary catalogue, plus 4,005 more taken from the levelled stories and exercises we publish — 9,278 words in total, each at the lowest level our own material uses it. A word in an A1 story is a word we hand to an A1 learner and expect them to read, which is as good a definition of "an A1 word" as this site can honestly offer.

How wrong this can be, and in which direction

It is an estimate. Not a measurement, not a score, and not something to put in a report. Here is what limits it, stated as plainly as we can manage, because a tool that hides its error bars is worse than no tool.

• It is a picture of our syllabus, not of English. The word list behind it is what we teach, at the level we teach it. Another course would place many of the same words a level up or down, and an ordinary word we happen never to have used comes out as off-list however easy it is.
• It cannot see grammar at all. A third conditional built entirely from A1 words scores A1. This is the biggest single limitation of any coverage measure and it is why sentence length is reported next to the level rather than folded into it — when the sentences are long and the vocabulary is easy, the tool says so out loud instead of quietly changing the number.
• It takes the easiest sense of a word. Board is an A1 school word and a B2 business word; the lookup takes A1. Every estimate is therefore biased slightly downward — a text is a little more likely to be harder than the level shown than easier.
• Short texts are noisy. Below about forty countable words, one unusual word moves the whole result by a level. The tool says when your sample is that small rather than reporting a number as if it were solid.
• It says nothing about content. A text can be linguistically A2 and entirely unsuitable for the reader in front of you. That judgement is yours and no counter is going to make it.

Used within those limits it is genuinely useful, and it is checkable: every word it levelled is a word you can look up on the vocabulary pages, on the page it links to in the results. If you think it has a word at the wrong level, you can see exactly where that level came from — which is more than most readability scores will tell you.

About the reading times

Three figures are given, because one number would be misleading. Silent reading assumes about 200 words per minute, which is a normal adult reading in their first language. Learner reading assumes about 120, which is closer to a B1 reader working through a text in English. Reading aloud assumes about 130, which is the figure to use if you are timing a presentation or a speaking exam answer.

All three are averages over a lot of people and none of them is your speed. If you are planning a two-minute IELTS speaking turn, the aloud figure is close enough to be useful; if you are estimating how long a class will take over a text, add half again and you will be nearer.

A worked example

Here is the tool's complete output for a short paragraph, produced by the same code that runs in the box above and printed here exactly as it came out.

Words                96
Sentences            5
Average sentence     19.2 words (long)
Reading time         29 sec silent · 48 sec for a B1 learner

Known by A1            79%
Known by A2            88%
Known by B1            94%
Known by B2            97%   ← crosses 95% here
Known by C1            97%   ← crosses 95% here

Names and numbers    1
Off our lists        3   (atlas, librarian, historical)

ESTIMATE             B2

The coverage curve above is drawn from this same paragraph — the line is the running total, and the level it reports is simply where the line crosses the 95% rule. If you paste this text into the box yourself you will get these numbers back, because it is the same calculation.

It is worth reading that off-list line rather than skipping it, because it is the most honest part of the output. Those 3 words are not in anything we publish — which says something about our word lists, and nothing at all about how hard the word is. A tool that reported them as "difficult" would be making a claim it cannot support.

Frequently asked questions