Which English Words Are Worth Learning? A Three-Part Filter

A recurring question in language-learning forums goes something like this: is there a tool that tells you how well known a word is among native speakers? If nine out of ten native speakers know it, learn it. If it's obscure, don't bother. The wording varies; the instinct is right.

It's also a question almost no product answers. Vocabulary tools compete on memory (schedulers, streaks, decks) and quietly hand the choice of *what* to memorize to a frequency list someone else wrote, or to whatever you happened to pause at. Usual disclosure: Piccard is mine, one of those tools, and selection is the part of it this post is about. The filter comes first; the product is an example at the end, not the point.

The question every learner skips

Method gets the attention. Anki or not, FSRS or not, twenty new cards a day or fifty. Those are real decisions — and they are all the second decision.

The first decision is what goes in, and it compounds. Twenty new cards a day is more than seven thousand cards a year. Each card comes back for review perhaps half a dozen times over its life. Choose badly and you have pre-paid tens of thousands of future review sessions for words that were never going to cross your path again.

Selection errors are also invisible when you make them. Saving a word feels like progress regardless of whether the word deserves it. The bill arrives months later, as a deck you can't face.

Two ways to pick badly

Capture everything. Every unknown word goes in. A 40-minute video can hand you 30 new words; you keep all 30, every video, for a month. Now the list has nine hundred entries, review takes half an hour, and opening the app feels like visiting a debt. This is the graveyard list: it dies of size. Nothing in it tells you which twenty of the nine hundred mattered.

Capture at random. The opposite failure is quieter. You save whatever surfaced: one rare idiom from a comedy sketch, an ornate adjective from a documentary, the name of a tool you don't own. Each save feels productive; the selection was made by accident of what you watched, and most of those words you will never meet again. This is how someone studies for two years and still can't follow an ordinary conversation: the hours went into a private collection of the language's least-used words.

Three inputs a real filter needs

The forum questioner wanted one input: how well known a word is among native speakers. That's the right first input, and it isn't the whole filter.

How common the word is. Word frequency is steeply unequal: a small core of words covers a striking share of ordinary text, a pattern stable enough to have a name — Zipf's law. A word nearly every native speaker knows pays you back on contact; a word one native speaker in a hundred knows almost never does. Commonness is the floor: learn what the language actually runs on, the same judgment behind curated core lists like the Oxford 3000.

Whether your goal needs it. Frequency is measured across a population, and you are not a population. "Oblique" is rare in general English and routine in design talk. Exam vocabulary is its own country. If your goal is following cooking videos, cooking words should clear the bar even when the general table ranks them low.

Whether you nearly know it. The cheapest word to learn is one you've already half-met — you recognize it reading but can't produce it, or you've met it three times this week. One or two reviews close that gap; a brand-new word with no hook costs several times more. Good selection is biased toward words you're already circling, not words that flatter you with their obscurity.

What YouTube does to the question

When your input is YouTube, you meet words in the order your watching serves them — not in frequency order.

That's a weakness. One niche video can hand you ten words you'll never see again: the capture-at-random failure, pre-loaded. Frequency lists exist precisely because raw exposure is skewed.

It's also why words from video stick when list-words don't. The words that keep appearing in what you actually watch have been filtered by the only judge that matters for you — your own consumption. A word that has turned up in three videos you chose isn't a frequency guess anymore; it's a fact about your life in English. (The other half of why video words stick is keeping the sentence and the moment, and that's a separate piece: How to Save Words from YouTube Subtitles.)

What YouTube doesn't do is make the keep-or-skip decision for you. Every line you pause at still asks it. That decision point is where a filter either exists or doesn't.

Where a tool can carry part of the filter

Here's how Piccard handles selection, described as it ships.

While watching. On YouTube videos that have captions, click a subtitle line and candidate cards appear: the word, its meaning, an example from that line. Keep or discard, one click each. You're still the filter; the tool removes the ceremony. Off YouTube, double-clicking a word on any page brings its definition and the same save.

On longer material. Paste a long text or upload a file (PDF, photo, video, audio) and you get a candidate checklist. Each candidate gets a 0–100 score for how much it is worth learning, a reason of a dozen words or fewer, and a CEFR estimate, so the why is visible before anything is ticked. Anything scoring 50 or above shows up already ticked, best first, and one confirm turns at most 20 candidates into cards. The score is an ordering, not a verdict: the ranking won't argue when you uncheck.

Your goal, weighted in. At setup you type your goal in your own words: "IELTS 7.5", "understand my teammates' jokes", "cook along without subtitles". That sentence is stored and fed to the ranking, so lines that serve your goal score higher. That's the second filter input, the one no frequency list can have.

Review runs on FSRS-6; recent Anki versions moved to this same scheduler family. A card saved from a video plays the video again from that exact moment. The free tier is 150 energy (about 150 cards), then prepaid top-ups at $5 / $10 / $20; no subscription, and topped-up energy never expires. To try the click-a-line version: Piccard on the Chrome Web Store.

If you're comparing the whole field of Migaku, Trancy, Anki, and Language Reactor, I compared them honestly elsewhere: Language Reactor Alternatives in 2026. And if you currently spend more effort moving words between two apps than choosing them, the export problem has its own write-up: Exporting Language Reactor to Anki.

The honest limits

YouTube only, captions required. No Netflix, no other players. Within YouTube, the video needs a caption track; without captions there's nothing to click. Uploaded files are the separate flow for material you own.

Built for English learners. The dictionary and card side focus on English, with Korean, Chinese, and Japanese as the translation side (the web app also offers a few European language pairs). Learning Spanish through German is not this tool.

The score is a first draft. It knows general commonness and it knows your stated goal. It doesn't know you half-know "albeit", or that you quit cooking. The final filter is you, so the override is one click.

The short version

"Which words?" decides more than the method does, and it has three inputs: how common the word is, whether your goal uses it, and how close to knowing it you already are. Frequency tables give you the first. Only you have the other two. What a tool can fairly do is put all three on the table at the moment of decision (scored, reasons visible) and then get out of the way. Skip the obscure word without guilt. Keep the word that keeps finding you.

All guides: Piccard Blog.