Vocabulary coverage is the share of the running words in a text that you already know. At 98%, about one word in 50 is unfamiliar, the level researchers link to comfortable reading without help; at 95% it is one in 20. The gap looks small, but in one large study the last few points brought the biggest gains in comprehension.
What exactly does coverage count?
Coverage counts running words, often called tokens: every word on the page, each time it appears. Paul Nation defines it as the percentage of running words in a text that the reader knows. That is different from counting different words, or types, and the gap between the two is huge. In Nation's analysis of the novel Lady Chatterley's Lover, 121,099 running words were made up of 8,757 types, and the 1,000 most frequent word families on their own covered 80.88% of the running words.
Researchers usually count in word families: a headword plus its inflections and common derived forms, so that nation, national and nationally go together. A learner who knows a family is assumed to recognise its members, which is generous at the start and fair later on.
Names are normally counted as known, because a reader works them out on first sight. Nation found that proper nouns made up between 0.5% and 2.12% of the running words in the novels he studied and 4.55–6.12% in newspapers. Short learner stories lean on them more. By a rough count in the Contos library (October 2026), capitalised names make up about 7% of the running words in the free Spanish chapters, and between 6% and 7% in French, Italian and Portuguese. That is a few points of coverage you get almost for free.
Why do the last few percent matter so much?
Because comprehension does not rise in step with coverage. The clearest evidence comes from Batia Laufer and Geke Ravenhorst-Kalovski, who tested 745 learners of English in Israel on vocabulary size and on the reading section of a national university entrance exam.
| Vocabulary size (word families) | Coverage of the exam texts, excluding names | Mean reading score (out of 150) |
|---|---|---|
| 1,000 | 78.6% | 83 |
| 2,000 | 87.7% | 90 |
| 3,000 | 90.6% | 102 |
| 4,000 | 92.8% | 111 |
| 5,000 | 94.0% | 122 |
| 6,000–8,000 | 94.8–96.3% | 138 |
Going from 1,000 to 2,000 word families added about nine points of coverage but only 7 points of reading score. At the top, about one more point of coverage went with the largest gain of all, around 17%. The authors propose 95% coverage, about 4,000–5,000 families, as a minimal threshold for academic reading in English, and 98%, about 8,000 families including names, as the optimal one. They add that reasonable comprehension below the threshold is possible.
Ronald Carver approached it from the other side in 1994. In his studies, easy material contained close to 0% unknown words, hard material 2% or more, and material matched to the reader around 1%. Our page on extensive reading sets out the other studies behind the 95% and 98% figures.
Is 98% the target for listening, films and TV too?
Possibly not. Van Zeeland and Schmitt tested 36 native and 40 non-native listeners on informal spoken stories whose coverage had been controlled. Most of both groups understood the stories adequately at 90% coverage, and the non-native listeners showed relatively good comprehension at 95%. That points to a lower vocabulary bar for following everyday speech than for reading, although speech brings other problems, as our page on reading or listening first explains.
Stuart Webb and Michael Rodgers measured the vocabulary of screen English:
| Material | 95% coverage | 98% coverage |
|---|---|---|
| 318 film scripts | 3,000 word families (95.76%) | 6,000 word families (98.15%) |
| 88 television programmes | 3,000 word families (95.45%) | 7,000 word families (98.27%) |
Both studies found large differences between genres and between episodes, so a single number is only a starting point.
How much of the next level do you already know?
A useful way to see coverage in action is to ask how much of a harder text a learner has already met. We took every free chapter in the Contos library at each level and checked what share of the running words at the next level had already appeared, in exactly the same form, in the chapters below it:
| Running words already met at lower levels | Spanish | French | German | Italian | Portuguese |
|---|---|---|---|---|---|
| A2 chapters, after all A1 chapters | 76% | 79% | 76% | 78% | 73% |
| B1 chapters, after A1 and A2 | 79% | 76% | 78% | 75% | 80% |
| B2 chapters, after A1 to B1 | 84% | 83% | 84% | 84% | 85% |
This is a strict test. A new ending on a familiar verb counts as unknown, and a reader also learns words from flashcards, other reading and life, so real coverage is higher. But it shows why each step up a level feels harder than the last chapter did: every level brings a fresh layer of vocabulary, and the gaps have to be filled by tapping, reviewing and reading more. That is why Contos gates each level by known words: 500 for A2, 1,500 for B1 and 4,000 for B2.
To estimate your own coverage, take a passage of about 100 words and count the words you would need to look up. Two means about 98%; five means 95%; ten or more means the text is better saved for later. Our page on choosing your reading level turns that into a quick test.
Read the first 100 words of a free chapter and count the words you need to tap.
Read free in your browserFrequently asked questions
Does 98% coverage mean understanding 98% of the text?
No. It means knowing 98% of the running words. Comprehension depends on grammar, background knowledge and which words are missing too. In Schmitt and colleagues' study, discussed on our extensive reading page, 98% coverage went with about 70% comprehension of academic texts.
Do names count as known words?
In most coverage research, yes, because readers identify them on sight. They are worth a few percentage points: between 0.5% and 2.12% of running words in the novels Nation analysed, and roughly 6–7% in Contos's short free chapters.
Is 95% coverage enough?
For some purposes. Laufer and Ravenhorst-Kalovski call 95% a minimal threshold for reading, and van Zeeland and Schmitt's listeners coped well at 95% with informal stories. For relaxed reading without help, 98% is the safer target.
How many words do I need for 98% coverage?
It depends on the text. Laufer and Ravenhorst-Kalovski put academic English at about 8,000 word families with names, and Webb and Rodgers found 6,000 for films. Texts written for learners reach 98% with far fewer, which is the point of graded material.
Why does a new level feel so much harder?
Because coverage drops when you move up. In the Contos library, only about three-quarters to four-fifths of the running words at each level have already appeared, in the same form, in the free chapters below it.