Why Learning 50 Words a Day Made My Student Worse at English
Tomas learned fifty words a day and his English got worse. Not slower, worse. The story has a shape I have now seen dozens of times, and a good ending.

Plate IAfter three months of cramming, Tomas sits down in the classroom and says his English is getting worse.
I want to tell you about Tomas, because his story has a shape I've now seen dozens of times, and because he gave me permission, and because the ending is good. But mostly because the middle of his story contains a sentence I've never forgotten. Three months into the most intense vocabulary project of his life, having "learned" over four thousand words, he sat down in my classroom and said: "I think I'm getting worse."
He wasn't imagining it. He was right. And understanding why he was right, mechanically, specifically, is worth more than any study tip I could give you, because Tomas wasn't lazy, wasn't careless, and wasn't doing anything the internet hadn't enthusiastically told him to do. He was doing everything, at maximum volume, and the volume was the disease. If you've read my post on how many words to learn a day, you know my position and the arithmetic behind it. This post is the other half: not the math, but what the failure actually looks like from inside a life. Because the math persuades some people, and stories persuade the rest.
01 · The Ascent (Weeks One to Three)
The Ascent (Weeks One to Three)
Tomas was an engineer, mid-thirties, disciplined in a way most learners only pretend to be. He'd found a frequency list, done the calculation every ambitious learner does (fifty words a day, five thousand words in a hundred days, fluency by summer), and built the machine: flashcard app, list loaded, ninety minutes daily, alarms set. When he told me the plan, I made my usual objections. He listened politely, the way engineers listen to warnings about over-engineering, and carried on. I understood completely. The plan had numbers in it, and numbers feel like guarantees.
And for three weeks, the dashboard agreed with him. Words learned: climbing beautifully. Daily streak: intact. Accuracy on reviews: high, because the reviews were still mostly words from the last few days, still warm. He messaged me a screenshot in week two, and I want to be honest: it was impressive. Fifty a day, every day, is a genuine feat of discipline. The problem was never his discipline. The problem was that the dashboard was measuring encounters, and his brain was quietly running a different accounting system altogether.

02 · The Collapse (Weeks Four to Ten)
The Collapse (Weeks Four to Ten)
The first crack was the review pile, exactly on schedule. By week four, each of those fifty-word days from the past was sending its review demands to the present, and the demands were stacking: three hundred due cards became five hundred became a number the app displayed in a smaller font, possibly out of embarrassment. Tomas responded like a disciplined person: longer sessions. Then like a tired disciplined person: faster clicking. And there, in the faster clicking, the real damage began, because speed-reviewing five hundred half-known words isn't review at all. It's a slideshow. Recognition flickered, "known" got clicked, and the words stayed exactly as half-known as before, now with an official stamp saying otherwise. The system's honesty died weeks before Tomas noticed anything.
The second crack was stranger, and it's the one that made him say "worse": his speaking began to degrade. Words he'd owned solidly for years suddenly came with hesitations, because their neighbourhoods had become crowded. He'd learned raise, rise, and arise in the same week, from adjacent list entries, and now all three flickered together whenever he needed any one of them. Same with economic and economical, sensible and sensitive, lie and lay. Fifty words a day off a frequency list means constantly learning lookalikes side by side, with no separate contexts to keep them apart, and interference (regular readers met this villain in the word-lists post) doesn't politely limit itself to the new words. It reaches back and smudges the old ones. Tomas hadn't just failed to add four thousand words. He'd installed static over vocabulary that used to be clean.

And the third crack was the simplest: the ninety minutes had to come from somewhere, and they came from everything that actually builds speaking. His podcast listening stopped. His weekly conversation exchange stopped ("no time, the reviews"). The one-minute retells I'd taught him stopped. The vocabulary project had eaten the language practice, the way a garden fence project might somehow consume the garden. By week ten he was spending more time on English than ever in his life, doing less English than ever in his life, and feeling every bit of that paradox without being able to name it.
03 · The Diagnosis (One Sentence)
The Diagnosis (One Sentence)
When he finally sat in my classroom with his "I think I'm getting worse," we opened the app and did an honest audit, and I'll give you the whole diagnosis in one sentence, because it's the sentence this post exists for:
That's what an oversized daily number actually is: not an achievement, a liability. Every word you meet, you either service (with spaced, honest, spoken retrievals) or default on. At five to ten a day, servicing costs ten pleasant minutes. At fifty, the debt compounds past any human's capacity within a month, and then the defaults begin: the fake "known" clicks, the abandoned reviews, the interference smudge, the cannibalized speaking time. The cruel elegance of it is that the dashboard keeps celebrating the borrowing while the defaulting happens off-screen. Tomas's numbers had never looked better than in the exact weeks his English was genuinely getting worse.
04 · The Repair (What We Actually Did)
The Repair (What We Actually Did)
The rebuild took ten minutes to design and a small act of courage to execute. First, the courage: we deleted about three thousand cards. Not suspended. Deleted. Everything he'd never once needed in real life, every lookalike cluster, every naked single word (you know my feelings; the survivors were rebuilt as chunks in personal sentences). Tomas watched three months of dashboard glory disappear and described the feeling, accurately, as "losing a game score I'd been cheating at anyway."

Then the new constitution, which regular readers could write themselves by now: seven new chunks a day, ceiling, chosen from his actual input rather than a list. Reviews first, spoken, production-direction, capped at fifteen minutes. The reclaimed seventy-five minutes went back to English itself: the podcasts, the exchange partner, the retells. And, because he was an engineer and needed a metric, we replaced "words learned" with a better dashboard: chunks retrieved successfully after 30+ days, which is the only number that was ever measuring anything real.
The recovery took about six weeks to feel and three months to trust. The old solid words came back first, once the static around them stopped being refreshed daily. Then the new chunks started surfacing in his conversation exchange, unforced, which had literally never happened with the list words, not once in four thousand attempts. A year later he passed the C1 exam he'd originally built the fifty-a-day machine for. Total new items in that year: roughly two thousand, half his old monthly pace. All of them, in his word, "load-bearing."
05 · Common Mistakes (The Ones Inside This Story)
Common Mistakes (The Ones Inside This Story)
Mistake 1: Trusting the dashboard over your mouth. Every metric in Tomas's app improved while his English declined. The only honest test was always the same one: can the word come out, spoken, in a sentence, a month later? Measure that, and you can't be fooled. (The vocabulary notebook post is about this same illusion in paper form.)
Mistake 2: Responding to overload with discipline. When the pile grows, the disciplined instinct is longer sessions and faster clicking, which converts review into slideshow. The correct response is always smaller intake, never bigger effort. You cannot out-discipline arithmetic.

Mistake 3: Learning lookalikes together. Frequency lists serve them up adjacent; brains smudge them together. If two words could be confused, they need separate weeks and separate contexts, or they'll spend years flickering.
Mistake 4: Letting the meta-work eat the work. When managing your learning system takes more time than using the language, the ratio has inverted, and no system survives inversion. Vocabulary exists to be spoken, read, heard. The moment it stops feeding those, it's fence-building in a garden you no longer visit.
06 · Try This Today
Try This Today
The Tomas audit, five honest minutes. Open whatever holds your vocabulary and answer three questions. How many items are waiting for review right now, and is that number growing week over week? Of the last twenty words you "learned," how many have you spoken aloud in a real sentence, ever? And how much of your weekly English time goes to managing words versus using English?
If the answers sting, do the courageous ten minutes: cull hard, cap the intake at a single digit, and give the reclaimed time back to input and speaking. It will feel like losing a score. It's actually paying off a debt, and debt-free, as Tomas can now tell you in fluent English, is a much better feeling than a high number.
07 · The Good Ending
The Good Ending
One last detail from the story, because it's my favourite. Months after the rebuild, I asked Tomas what the fifty-a-day period had actually given him, expecting something rueful. He thought about it properly, engineer-style, and said: "The proof. I needed to watch it fail with my own numbers. You told me in week one, but it wasn't mine yet."
So maybe that's the honest purpose of this post. Some of you will read it, adjust your daily number tonight, and skip the three-month detour. Others need the detour, and that's human, and the door back is always the same ten courageous minutes away. But at least now, somewhere around week four, when the pile grows strange and the old words start to flicker, you'll hear a small voice saying: this has a name, and a shape, and an ending. Tomas says hello, by the way. He's the one who suggested I write it all down.

WordCat caps the intake, spaces the reviews, and feeds you chunks from real contexts, which is the whole repair, built in from day one. Try it at wordcat.app.