Every country reads English better than it speaks it

The EF English Proficiency Index publishes a single national score for each of 123 countries, and that number is what gets quoted. Underneath it sit four skill scores that almost nobody reads together. Put seven countries side by side and the same shape appears in all of them, including the country ranked first in the world.

Do people really read English better than they speak it?

Yes, and the size of the difference is the surprise. The Netherlands is ranked first of 123 countries and regions on the EF English Proficiency Index 2025 with a national score of 624, and its four skill scores are 634 in reading, 628 in listening, 554 in writing and 514 in speaking. That is a gap of 120 points between the strongest skill and speaking, in the country that tops the table. Germany, ranked fourth with 615, scores 623 in reading and 521 in speaking, a gap of 102 points.

The pattern holds in every country checked for this piece. In all seven, reading is the highest of the four EF EPI 2025 skill scores, and speaking sits between 48 and 120 points below it. The countries span most of the length of the ranking, from first to ninety-sixth, and cover three continents and five unrelated language families: Germanic, Romance, Semitic, Koreanic and Japonic. That rules out the easy explanations. This is not a Dutch quirk, not an Asian one, and not a property of any single school system.

How big is the gap, country by country?

The table below is the whole dataset behind this page. Ranks and scores are EF's own, taken from each country's EF EPI page, and the final column is simply reading minus speaking. Japan scored 446 on the EF EPI 2025 and ranked 96th of 123 countries, with reading at 454 and speaking at 393. Israel scored 524, ranked 46th, with reading at 533 and speaking at 485.

One column deserves a warning label. Speaking is the lowest of the four skills in five of the seven countries, but not in Israel or Brazil, where writing scores lower still. That is why every claim on this page is phrased as reading minus speaking rather than as speaking being the worst skill: the productive skills, speaking and writing, both trail the receptive ones, and which of the two comes last varies by country.

Does the gap close as a country gets better at English?

The intuitive answer is yes, and in this sample the data says the opposite. The three highest ranked countries here, the Netherlands at first, Germany at fourth and Spain at thirty-sixth, have the three widest reading-to-speaking gaps, at 120, 102 and 96 points. The four narrowest gaps are the four countries below them in the table: Israel at 48 points, South Korea at 51, Brazil at 52 and Japan at 61. The split is clean: every gap above 90 points belongs to a country in the top forty of the ranking, and every gap below 65 belongs to a country outside it.

Read carefully, that is not a claim that Dutch people speak English badly. They speak it far better than almost anyone: 514 is a high absolute speaking score, well above the 488 global average for all four skills combined. It is a claim about where the gains went. As national English improves, reading improves faster and further than speaking does, so the distance between what a population can understand and what it can produce grows rather than shrinks. Whatever raises reading, and school hours, subtitles, the internet and English-language media all plausibly do, does not raise speaking at the same rate.

Getting better at English, at national scale, does not close the speaking gap. In the EF EPI 2025 sample here, the three countries that rank highest are the three with the widest distance between reading and speaking.

Why does speaking lag behind reading?

Reading and listening are recognition. The words arrive, and the job is to match them against what is already stored. Speaking is production under time pressure: retrieving a word before the pause gets awkward, choosing a tense while the sentence is already moving, and repairing a mistake out loud without stopping. These draw on the same vocabulary and the same grammar, which is why the skills correlate, but they are not the same operation, and practising one does not automatically train the other.

The exposure most learners get is overwhelmingly receptive. Subtitled television, English-language social media, technical documentation, song lyrics and school reading comprehension all feed recognition. Most of it can be consumed alone, at any hour, at no cost. Speaking needs a second person who will listen, wait, and say when something came out wrong, and that has always been the scarce and expensive half. The skill scores are what that asymmetry looks like when a few million people are measured on both.

What actually closes a speaking gap?

The uncomfortable implication of the table is that more input will not do it. A learner who already reads at 634 and speaks at 514 does not have a comprehension problem to solve, and another year of English television will move the number that is already high. Closing the distance takes the thing the numbers say is missing: unscripted production, often enough to become automatic, with correction arriving while the sentence is still fresh rather than in a report a week later.

That is the case for talking rather than studying, whoever you talk to. A conversation class, a language exchange partner, a tutor or a voice AI all satisfy the same requirement, and the constraint that decides which one people actually use is availability: a skill that needs daily repetition has to be schedulable on the days nobody else is free. SpeakBase exists for that specific gap, and we should say plainly that it is ours: Kate is a voice tutor who talks with you in English and switches to your own language to explain the moment you get stuck.

The same gap, one country at a time

Four of the countries in the table have their own edition of this research, written in that country's language and built on its own exam and ministry data rather than on EF alone: Israel, Spain, Brazil and Japan. For what the production half costs in study time, the US State Department hour budgets count their target in speaking and listening, not in reading.