Count the words you understand when you read them, then count the words that arrive when you need one, and the second number is a fraction of the first. That gap is not shyness and more words will not close it, because the words that fail you are already on the first list. What separates the two lists is retrieval: a word becomes usable by being pulled out under time pressure, not by being met again.
So the work has two halves and they need different weeks. Get the first list to a size where the gaps stop mattering, which is a question of frequency and schedule and is mostly solved. Then move words across, which no flashcard app does for you and which is where nearly all of the disappointment in this subject lives.
The six this cluster hangs on
- the first two thousand words cover most speech, which is why a small, well chosen list beats a large scattered one.
- a word is known when you can use it, not when you recognise it. Recognition arrives years before production and the gap is the whole problem.
- make a decision, not do a decision. Which verb goes with which noun is arbitrary and has to be learned as a pair.
- heavy rain, not strong rain. The same is true of adjectives, and getting it wrong is grammatical and still wrong.
- collocation is what fluency sounds like. Native speakers are recognised by their word partnerships far more than by their grammar.
- phrasal verbs are the everyday register, so avoiding them produces English that is correct and noticeably formal.
How much is enough, and enough for what
The sizes are known and they are larger than a course implies. Paul Nation's 2006 count put comfortable reading of a novel or a newspaper at 9,000 word families and following ordinary spoken English at about 7,000, and it is worth reading those two figures together: speech asks for less because speakers repeat themselves, slow down and point at things, while a page does none of that and cannot be asked. The first 2,000 families are about 80 percent of the words on any page, a proportion that has not moved since Michael West published his list in 1953, and it sounds close to finished, which is exactly why the middle of this subject catches people out.
A family is one root and everything grown from it, so decide, decision, decisive and undecided are one entry rather than four. That is not a bookkeeping convenience. It is a claim about what you have to learn once, and it is why one root, and the nouns, verbs and adjectives around it is worth an afternoon of deliberate attention: taking the endings seriously turns a list of nine thousand into a much shorter list of things to meet.
The third thousand is where the free supply stops
Everyone gets to roughly three thousand families the same way, by watching things and reading things and looking up what blocks them, and everyone assumes the next six thousand will arrive the same way. They will not, and the reason is arithmetic rather than effort. Words in the fourth thousand and above turn up rarely enough that months can pass between two encounters with the same one, and nothing survives a gap that size on its own. The learner keeps watching, the input keeps arriving, and the count stops moving.
This is the point where a list stops being a crutch and becomes the only instrument that works, because a list is the one thing that can put a rare word in front of you twice in a week. That is the argument for the core word list, worked through one book at a time, whose 4,000 headwords are ordered by frequency rather than by topic, and for the vocabulary that carries essays, papers and lectures if the target has an exam behind it: both are frequency-ordered sets built precisely for the band where meeting words by accident has stopped paying.
The schedule that makes a list survivable is its own subject, and it is on why words vanish, and the schedule that keeps them rather than here.
Two tools, and what each one is for
Anki is still the spaced repetition program in 2026, having outlasted every startup that set out to replace it. It replaced the paper card box, and specifically the part of the card box nobody could do by hand: showing you a card at the moment you were about to forget it rather than in the order the cards happen to sit. Free on desktop and on Android, paid once on iOS, and the iOS price is the honest thing to mention rather than to leave out. The setting that matters is not the interval. It is which way the card faces: English on the front and your own language on the back tests recognition, which you already have, and the reverse tests retrieval, which is the thing you came for.
The other one is a corpus. english-corpora.org has held a set of large searchable collections of real English for two decades, and in 2026 it is still the fastest way to settle whether a phrase is actually said and how often. It replaced asking a native speaker, who gives you an opinion where you wanted a count and who is unreliable on exactly the pairings this matters for. Free for the amount of looking up a learner does. Search a verb and read the words that turn up beside it: that list is which words English puts together, and which it never does, and there is no way to derive it from the meanings.
The mistake: collecting the interesting ones
Vivid words feel like progress. Rare ones feel like proof. So a notebook fills up with meticulous and serendipity and the sixteen ways to say very tired, while sort out, spare, bring forward and put up with, which are in every second conversation, are skipped because they look like words you already know. The result is a speaker with an ornamental vocabulary and holes in the middle of it, and the holes are the part that gets noticed.
The same trap sits under phrases that do not mean what their words say and what people actually say, and when not to say it. Both are worth a small deliberate budget and both are the worst possible place for a large one: their frequency is low, their register is narrow, and an idiom deployed slightly wrong lands harder than a plain sentence ever could. get up, put off, run into: verbs a preposition rewrites is the opposite case and deserves the hours, because it is high frequency, unavoidable, and the one part of the lexicon where a learner's careful Latin-rooted alternative marks them out in a sentence.
What the count does not tell you
A number of families is a claim about breadth and says nothing about how deep any one entry goes, which is why two people with the same score sound nothing alike. It also says nothing about speed. Everything on this page is about getting words onto the list and further into it; whether they arrive inside the half second a conversation allows is a separate mechanism with a separate fix, and that is getting from correct sentences to an actual conversation. The supply problem, meanwhile, has one honest answer above the third thousand, and it is volume: getting through more of it without losing the meaning is where the middle-frequency words live in any quantity, and the world's second language, from its grammar to the job interview is where the question of how big your list needs to be gets decided by what you want the English for.