NYT Spelling Bee Strategy: How to Find Every Word

To find every word in the NYT Spelling Bee, stop free-associating and run a sweep instead: pair the center letter with each outer letter in both positions, work through the common endings, then extend every short word you already found by one letter at a time. Random recall plateaus fast. A fixed order doesn't, because it forces you past the words your brain reaches for first and into the ones it skips. Here's the method, and the letter-frequency data that decides what order to check things in.

Why is the Spelling Bee so hard to finish?

Because the puzzle sits right on top of the densest part of English, then removes your best shortcut.

Consider the word-length data. Peter Norvig, Director of Research at Google, analysed the Google Books Ngrams corpus in English Letter Frequency Counts: Mayzner Revisited, covering 97,565 distinct word types appearing 743.8 billion times across 3.56 trillion letters. He found the average English word runs 4.79 letters, with 80% of all word mentions falling between two and seven letters.

Spelling Bee requires four letters minimum and gives you seven tiles. So the answer list lives almost exactly where English is thickest. There are a lot of valid words on any given day, which is why finishing feels endless.

Then there's the missing S. The puzzle never includes it, and that single omission removes the pluralisation trick that would otherwise let you double your list without knowing a single extra word. What's left has to come from actual vocabulary.

Why do you stop finding words halfway through?

Because free recall doesn't run smoothly. It runs in clumps, and it stalls when the clump you're in runs dry. Understanding that is what makes the sweep in the next section worth the effort, so it's worth a minute.

The research on this comes from verbal fluency testing, where people are asked to produce as many words as they can under a constraint. Angela Troyer, Morris Moscovitch and Gordon Winocur set out the standard model in Neuropsychology in 1997 (volume 11, issue 1, pages 138 to 146), a paper indexed by the US National Library of Medicine. They argued that fluency performance breaks into two separate components: clustering, meaning the production of words within a subcategory, and switching, meaning the ability to shift between clusters. You dig in one seam, exhaust it, then jump to a new one.

Which of those two actually drives your score depends on the type of task, and this is the part that matters for the Bee. In their first experiment, with 54 older and 41 younger adults, both components correlated highly with word count on semantic fluency, the "name all the animals" version. But on phonemic fluency, the version constrained by letters rather than meaning, switching was more highly correlated than clustering with the number of words produced.

Spelling Bee is a phonemic fluency task. You're not searching by meaning, you're searching under a letter constraint. So the finding maps directly: how many words you end up with depends less on how deep you dig in any one direction and more on how often you successfully jump to a new one. Staring at the honeycomb waiting for inspiration is digging. The sweep is switching, on purpose, on a schedule.

Troyer's team also ran a second experiment that has an annoyingly practical implication. Among 22 young adults, dividing attention reduced both the number of words generated and the amount of switching, and it did so on phonemic fluency only. Semantic fluency held up. If you play the Bee with something else running, the thing distraction takes from you is precisely the switching, which is the thing your score depends on most. Their overall read was that switching is tied to frontal-lobe functioning, which is the same machinery that handles deliberate task-shifting generally.

One honest caveat. These were 60-second clinical fluency tests, not a puzzle you can leave open on your phone all day. The mechanism transfers. The numbers don't, and nobody has run this study on Spelling Bee players. Treat it as an explanation of why the checklist beats staring, not as a measurement of how much.

Does staring at your found words make it harder?

Probably, yes. And this one is worth knowing because the instinct is so strong and so wrong.

You stall, so you scroll back through the words you've already found, hoping one of them jogs something loose. Feels productive. Memory research says it's closer to the opposite.

The finding is called the part-list cuing effect, and Norman Slamecka reported it in the Journal of Experimental Psychology in 1968. Give people a list to learn, then at test hand some of those items back as helpful cues, and their recall of the remaining items gets worse. Not better. Worse than people given no cues at all.

It has held up for decades. In a recent open-access replication indexed on PubMed Central, participants shown part of the list recalled 37.33 percent of the rest against 49.48 percent for the uncued control in the first experiment, a moderate effect at d = 0.59. The second experiment showed the same pattern, 42.36 percent against 51.90 percent, at d = 0.45.

The leading explanations all point the same way. Reading the cues pulls you into covertly retrieving them again, and that retrieval crowds out or actively suppresses the items you haven't got yet. There's a related account where the cues wreck the retrieval plan you'd built. Either way, the words in front of you are competing with the words you want.

Which describes the Spelling Bee stall almost exactly. Your found-words list is a part-list cue, permanently on screen, and every time you re-read it you strengthen the words you already have at the expense of the ones you don't.

Three things follow, and they cost nothing to try:

• Stop re-reading the list when you stall. Look at the seven letters instead. The letters are the search space; your found words are the part of it you've already spent.
• Collapse or cover the word list if the interface lets you. Out of sight genuinely helps here, which is not something you can say about most productivity advice.
• Walk away for ten minutes. Coming back cold is the cheapest way to clear the interference, and it's why the puzzle you abandoned at lunchtime often cracks open in the evening.

But there's a wrinkle in that same study, and it's the most useful thing in it. Looking at your found words isn't automatically the problem. How you look at them is.

Wallner and Bäuml ran a third condition alongside the two above, where people moved through the cues at their own pace instead of being marched through them. The impairment nearly vanished. Self-paced cuing scored 46.87 percent against the 49.48 percent control in the first experiment, an effect size of d = 0.13, next to d = 0.59 for the standard version. The second experiment went further: 52.47 percent against a 51.90 percent control, d = 0.03. No measurable cost at all.

So the damage isn't in the cues themselves. It's in being swept past them with no control over the tempo, which is roughly what happens when you flick your thumb up the found-words list scanning for a spark.

That sharpens the advice rather than reversing it:

• Scrolling fast is the version that costs you. A quick anxious skim is closest to the condition that produced the biggest impairment.
• Deliberate beats frantic. If you do want to review what you've found, go slowly and on purpose. Read the words rather than flicking past them.
• Better still, review with a question. Look for the shapes you haven't tried yet rather than rereading for its own sake. "Have I done anything with that middle letter doubled?" is a plan. Scrolling isn't.

Same honest caveat as the section above. These were list-learning experiments in a lab, not people playing a puzzle on their phone across a whole day. The mechanism transfers cleanly enough to act on. The percentages are not a prediction about your score.

Does rearranging the letters actually help?

Yes, and there's eye-tracking evidence for why. The section above told you to stop staring at your found words and look at the seven letters instead. This is the follow-up nobody mentions: the arrangement those seven letters happen to be sitting in is not neutral, and it can be actively working against you.

The honeycomb holds one fixed layout until you press shuffle. Most players treat that button as decoration.

What happens when letters sit next to each other

Jessica Ellis and Eyal Reingold ran the experiment that makes this concrete, published as "The Einstellung effect in anagram problem solving: evidence from eye movements" in Frontiers in Psychology in 2014. Sixty undergraduates at the University of Toronto Mississauga worked through 72 anagram trials with their eye movements tracked throughout.

The design is the clever bit. Every anagram had six letters: a central three-letter string plus three loose letters. Sometimes that central string spelled a familiar word. Sometimes it was a meaningless nonword. Same task, same number of letters, one difference in how they were arranged.

People were slower when the chunk was a real word. Solving took 15.3 seconds on word trials against 13.4 seconds on nonword trials, t(59) = 3.25, p < 0.01. Overall accuracy across the whole study ran at 52.7 percent, and the word trials were, if anything, slightly worse rather than better.

Two seconds doesn't sound like much until you notice what the eyes were doing, because the pattern is the opposite of what you'd guess. Participants took in the familiar chunk faster, with much shorter initial fixations on the central string in word trials, F(1,59) = 29.78, p < 0.001. Then they spent significantly longer dwelling on the individual letters, F(1,59) = 25.31, p < 0.001.

So the familiar arrangement was easy to read and hard to escape. Recognising it cost nothing. Breaking it apart to build something else cost real time. That's the Einstellung effect, and it's the same shape as the trap in the previous section: the thing that feels like a foothold is the thing holding you still.

Now look at your honeycomb. If three of your letters have landed side by side spelling something familiar, you're in the word condition of that experiment, and you'll keep reading them as that chunk every time your eye passes over them. Shuffling breaks the adjacency. It doesn't give you new letters. It stops the old grouping from being handed to you for free.

The usual caveat holds. Their anagrams had a fixed centre string and loose outer letters, which isn't quite the Bee's fixed centre and rotating ring, and these were single trials in a lab rather than a puzzle you pick at all day. The mechanism transfers. The seconds don't.

And when shuffling isn't enough?

Then you leave it, which turns out to be a real technique rather than an admission of defeat. Ut Na Sio and Thomas Ormerod pooled the evidence in "Does Incubation Enhance Problem Solving? A Meta-Analytic Review", published in Psychological Bulletin in 2009 and indexed by the US Department of Education's ERIC database as EJ827110. They report that across the literature "the authors identified a positive incubation effect."

Two of their moderators change how you should actually take a break. The first: "Longer preparation periods gave a greater incubation effect." Stepping away only pays once you've genuinely worked the puzzle. A break isn't a substitute for effort, it's what you do with the effort afterwards.

The second is the one worth changing your habits over. Filling the gap "with high cognitive demand tasks gave a smaller incubation effect", and, as they put it, "low cognitive demand tasks yielded a stronger incubation effect than did rest during an incubation period when solving linguistic insight problems." Spelling Bee is a linguistic task. So doing something mildly occupying beats sitting and thinking about it, and it also beats picking up something absorbing. Wash up. Walk somewhere. Don't open another puzzle, and don't scroll.

What to do with all that

• Shuffle when you stall, not constantly. Reshuffling every few seconds just adds noise. Use it as the first move when you feel the plateau, before you reach for anything else.
• Shuffle before you scroll your found words. It's the cheaper intervention and it doesn't carry the part-list cuing cost from the section above.
• Write the letters out if one chunk won't let go. A different order on paper, or in a circle, breaks a grouping the interface keeps rebuilding.
• Earn the break, then make it a boring one. Work hard first. Then go and do something undemanding, which the meta-analysis says beats both resting and getting absorbed in something else.

None of this manufactures words you don't know. It clears the interference sitting between you and the ones you do, which is what most of the gap between your score and Genius turns out to be. If you'd rather see the shapes you're missing than keep circling them, the Spelling Bee helper is there, and the section below on when to use it is worth reading first.

How do you sweep the search space systematically?

Six outer letters and one center letter give you a small enough space to cover properly. The trick is covering it in a fixed order so you never lose track of what you have checked.

1. Run the center-letter pairs. Take the center letter and pair it with each outer letter, in both orders. With six outer letters that's twelve two-letter openings. Say the center is T and the outer letters include A: check TA and AT. Most valid words in the puzzle start with one of these twelve pairs.
2. Clear the four-letter words. They're worth 1 point each, which makes them look skippable. They aren't. They're the densest part of the list and they show you which letter combinations actually work today.
3. Extend everything. Every four-letter word you found is a stem. Add one letter, then two. TACO becomes TACOT? No. But COAT becomes COATED if D and E are present. This step converts 1-point words into 6-point words.
4. Sweep the endings. Work a fixed list: -ING, -ED, -ER, -LY, -ION, -TION, -MENT, -ANT, -ENT, -ATE, -ITY, -OUS, -ABLE, -ANCE. Only some will be possible with today's letters, and eliminating the impossible ones takes seconds.
5. Sweep the prefixes. RE-, UN-, IN-, DE-, EN-, OUT-, OVER-, PRE-, NON-. These stack on words you already have.

Do it in that order every day and the puzzle stops being a memory test. It becomes a checklist.

Which letter patterns should you check first?

Not all letters deserve equal attention, and Norvig's corpus numbers show why. E is the most common letter in English at 12.49% of all letters, followed by T at 9.28%, A at 8.04%, O at 7.64% and I at 7.57%.

But raw frequency isn't the useful part. Position is. Norvig found that E is roughly four times less common as a first letter than it is elsewhere in a word. That's directly actionable: if E is one of your seven letters, don't waste time hunting words that start with E. Hunt endings instead, because that's where E actually lives. Norvig puts it plainly: E "makes a comeback as the most common last letter," and it's very common in the third and fifth positions too. N behaves the same way and gets ignored more often, running about three times rarer as an opener than it does overall. So when either turns up in your seven, spend your time on what a word ends with rather than what it starts with.

Letter	Share of English letters	What it means for your sweep
E	12.49%	Check endings first, not beginnings
T	9.28%	Strong in -ION, -ATE, -ITY endings
A	8.04%	Works anywhere, good stem vowel
O	7.64%	Pairs well into -OUS, -ION
I	7.57%	Feeds -ING, -ITY, -ION
Q, Z, J	0.12%, 0.09%, 0.16%	Rare, so expect a shorter answer list

Frequencies from Peter Norvig's analysis of the Google Books Ngrams corpus.

The vowel-to-consonant split in your seven letters tells you what kind of day it is. Four or five consonants against two vowels means short, blunt words and probably a lower total. Three or four vowels means longer words and a bigger list, so budget more time.

Does the order you check letter pairs matter?

It does, and it's the one gap in the sweep as described above. Step one tells you to run twelve center-letter pairs. It doesn't tell you which to run first. On a puzzle where you're going to stall eventually, that order decides how much you've banked before you do.

Norvig's corpus answers it, because the same analysis that produces single-letter frequencies also counts pairs. Across 2,819,662,855,499 two-letter sequences he found 669 distinct bigrams that actually occur, out of 676 possible combinations. Only seven never showed up anywhere in the corpus: JQ, QG, QK, QY, QZ, WQ and WZ.

The distribution is steep, which is what makes it useful. Here are the pairs worth knowing:

Bigram	Share of all letter pairs	Bigram	Share of all letter pairs
TH	3.56%	EN	1.45%
HE	3.07%	ND	1.35%
IN	2.43%	TI	1.34%
ER	2.05%	ES	1.34%
AN	1.99%	OR	1.28%
RE	1.85%	TE	1.20%
ON	1.76%	OF	1.17%
AT	1.49%

Bigram frequencies from Peter Norvig's analysis of the Google Books Ngrams corpus. Read down the left column, then the right: the table runs in descending frequency order.

Notice how fast it falls away. TH on its own is three times as common as OF, and TH and HE together outweigh the bottom five pairs on the list combined. So the order isn't a tiebreaker. If you only ever get through the first few buildable pairs before you stall, you've still covered the heaviest part of the distribution.

One deletion before you use that list. ES is dead on arrival here, because the Bee never includes S. So you get fourteen usable pairs out of the top fifteen, not fifteen. It's the same reason you can cross -ES and -IES off your suffix sweep before you start, which saves a few seconds every single day.

How to actually use it takes about ten seconds:

• Cross the list against your seven letters. Most days only three or four of the top fourteen are even buildable, which is a much shorter list than twelve.
• Run those first, in frequency order. TH and HE before OR and TE, every time.
• Remember these are pairs anywhere in a word, not just openings. IN and ER earn most of their frequency in the middle and at the end, so check them as chunks to build around rather than as starting blocks.

If T, H and E all turn up in your seven, pay attention. Those are the two highest-frequency pairs in English sitting next to each other, and they chain. THE itself is too short to score, but THEN, THEME, ETHER, TETHER and anything ending -THE come out of the same three letters.

The caveat, since this article has been honest about the others. Norvig's percentages count how often pairs appear in running text, and running text is dominated by a handful of very common words. What you want for a puzzle is which pairs appear across many different words, which is a slightly different question nobody has published an answer to for this corpus. The ranking is close enough to act on. Don't treat the decimal places as meaningful.

How do you find the pangram faster?

Every puzzle has at least one pangram, a word using all seven letters, and it carries a 7-point bonus on top of its normal score. A seven-letter pangram is worth 14 points. On a small puzzle that single word can be most of the gap between Amazing and Genius.

The fast method is subtraction rather than addition. Write the seven letters out and separate the consonants from the vowels. Now look at the ugliest one or two letters, the ones you can't easily place. The pangram almost always exists to use those. Ask what suffix or prefix would absorb them.

If you have G, I and N together, the answer very likely ends in -ING. If you have T, I and O, try -TION. If you have M, E, N and T, try -MENT. These aren't guesses, they're the highest-frequency multi-letter chunks in English, and the puzzle constructor has to build around the same constraints you do.

One more thing worth knowing: letters can repeat. A pangram uses all seven letters at least once, not exactly once, so a five-letter set can still stretch into a nine-letter word.

Is a perfect pangram worth hunting for?

Yes, but not for the reason most players assume. It isn't worth more points. It's worth hunting because it's the only part of the puzzle you can actually solve instead of search.

Start with what the term means, because it gets used loosely. A perfect pangram uses each of the seven letters exactly once. No repeats at all. And that constraint fixes the length for you: seven tiles used once each is a seven-letter word, every time.

Which gives you a small piece of logic worth carrying around. Any seven-letter pangram is automatically a perfect pangram. It can't be anything else, because there's no room for a repeat inside seven slots when all seven letters have to show up. So "perfect pangram" and "seven-letter pangram" describe the same thing. And every pangram of eight letters or longer repeats at least one letter, without exception.

Why the points argument gets it backwards

A perfect pangram scores 14. Seven points for its seven letters, plus the 7-point pangram bonus. That's exactly what the section above said a seven-letter pangram is worth, which follows, since they're the same word.

There is no bonus for perfection. The Bee doesn't know or care that you used each letter once. An eight-letter pangram scores 15 and beats it. A nine-letter pangram scores 16. So if you're chasing perfect pangrams for the score, you're chasing the cheapest pangram on the board.

The value is somewhere else entirely.

It's the one closed problem in the puzzle

Everything else in the Bee is an open search. You don't know how many words exist, you don't know how long they are, and you don't know when you've finished. That's what makes the plateau in the section above feel so bad. There's no edge to the space you're searching.

A perfect pangram is different. You know the exact letters. You know the exact length. Nothing is hidden. That isn't a word hunt any more, it's an anagram, and anagrams are a solved problem in the sense that you can work them methodically to the end.

And because it's an anagram, actual research applies to it, which is not something you can say about most puzzle advice. Laura Novick and Steven Sherman, in "On the nature of insight solutions: evidence from skill differences in anagram solution" in the Quarterly Journal of Experimental Psychology, 2003, 56(2), pages 351 to 382, ran two experiments on how people crack anagrams. Experiment 1, in their words, "documented that anagrams yield pop-out solutions, especially among highly skilled solvers."

The mechanism they propose is the useful part. They suggest "pop-out solutions may result from parallel processing of the constraints on the rearranged order of the anagram letters, whereas search solutions may result from a serial hypothesis-testing procedure."

Two different modes, in other words. One where the answer arrives whole, and one where you grind through candidates in order. Their Experiment 2 found both modes still depend on gradually accumulating partial information, so the pop-out isn't magic. But the practical read for a Bee player is straightforward:

• Give it a genuine stare first. Look at the seven letters as a set and wait a beat. For an anagram, that's a real method rather than stalling, and the study found the pop-out mode shows up most in skilled solvers. It gets better with practice.
• Then switch deliberately. If nothing arrives, stop staring. Staring longer is not the serial mode, it's just staring. Move to testing shapes one at a time.

How to tell if one exists before you spend time on it

Here's the catch nobody mentions: the puzzle guarantees at least one pangram, but it guarantees nothing at all about a perfect one. Plenty of days simply don't have one. So the first move is a feasibility check, and it takes about five seconds.

Count the vowels in your seven letters.

You're looking for two or three. A seven-letter English word built from a single vowel is a genuine rarity, and one built from five vowels and two consonants is worse. Two or three vowels against four or five consonants is the ordinary shape of a seven-letter English word, and if your tiles are in that range a perfect pangram is at least possible.

To be straight about what that is: it's a rule of thumb from the shape of the language, not a published number. Peter Norvig's corpus counts, which this article leans on elsewhere, put E at 12.49% of all letters, A at 8.04%, O at 7.64% and I at 7.57%. Those describe letters in running text, not how many vowels a seven-letter word needs, and nobody has published the second figure for this corpus. Use the vowel count as a filter for where to spend your attention, not as a rule.

Working it when the check passes

The thing that makes a perfect pangram hard is the thing that makes it findable: you can't repeat anything. Half the shapes you'd normally reach for are gone. No -LL-, no -EE-, no -OO-, no -TT-. If your instinct is a doubled letter, drop it and move on.

So work the ends instead, and work them in order of how much they commit:

• -TION spends four of your seven letters and leaves three in front. That's a very short list to check.
• -ING spends three and leaves four, but only if I, N and G are all present and unrepeated elsewhere in your candidate.
• -MENT, -ANCE, -ABLE are worth a glance, though each needs an E and each eats four of your seven.
• UN-, RE-, OUT- work from the front and pair well with a three or four letter tail.

Then place your ugliest letter first, the same trick the pangram section recommends, except now you're placing it into a fixed seven-slot frame rather than an open word. That's a much smaller question. If a rare letter genuinely won't sit anywhere in seven slots, you've learned something useful: the day's pangram is longer and repeats something, and you can stop hunting for the perfect one and go back to the normal method.

When to do it

At the very start, before you've found much. And this isn't a preference, it follows from the found-list section above.

That section explains why a long list of found words makes the remaining ones harder to reach. The same interference applies here, and an anagram is exactly the kind of task it damages, because you're trying to let seven letters rearrange themselves freely in your head while a screen full of words you already found pulls your thinking toward the patterns you've already used.

So run the vowel check and give the anagram ninety seconds while your head is still clear. If it lands, you've got 14 points and a strong sense of which letters actually combine that day. If it doesn't, you've spent ninety seconds and ruled out a whole category of answer. Both are good outcomes, which is not something you can say about most of the time spent staring at this puzzle.

Why do rare letters change the whole puzzle?

Because they cap how many answers can exist. Norvig's corpus puts Q at 0.12% of English letters, Z at 0.09% and J at 0.16%. Compare that to E at 12.49% and the scale of the difference is obvious: E appears more than a hundred times as often as Z.

So when one of those rare letters is your center letter, every single valid word has to contain it, and the answer list shrinks hard. That's actually good news. A short list means Genius arrives at a lower raw score, and it means you can afford to check exhaustively rather than broadly.

When a rare letter sits on the outside instead, treat it as optional. Plenty of the day's words won't use it at all. But it's almost certainly in the pangram, because that's usually the only place it fits.

Why does the Bee reject a word you know is real?

Because the Bee isn't checking a dictionary. It's checking a curated list that a human editor maintains, and the test for getting on that list isn't whether a word exists. It's whether the word feels fair to a very wide audience.

Sam Ezersky, who edits the puzzle, described his process to Slate in 2021. When he's unsure about a word, he said, "I'll see if the word is listed in the major dictionaries that I have at my disposal, which primarily are Merriam-Webster and the Mac dictionary, which I believe riffs off New Oxford American." So dictionaries inform the call. They don't make it.

The actual standard is editorial, and he stated it plainly: "What feels fair to our wide-ranging audience? I don't want to snub those where it's a word that is so common to their background or lifestyle or culture, but I also don't want to include something that will truly mystify the vast majority of our solving audience."

Read that twice, because it changes what a rejection means. Your word getting refused isn't a ruling that you were wrong. It's one person's judgment call about a readership of millions, and judgment calls move over time as solvers push on them. That's why a word that bounced last year sometimes lands this year.

And the same logic runs the other way. The list contains words you've never seen, which is not a gap in your reading so much as a fact about how vocabularies work.

Here's the research on that. In a study published in Frontiers in Psychology, Brysbaert, Stevens, Mandera and Keuleers (2016) estimated vocabulary size across a large crowdsourced sample.

Group	Lemmas known	What it means at the Bee
Average 20-year-old	42,000	The baseline the puzzle is pitched at
Lowest 5%	27,000	More rejections that feel unfair
Highest 5%	52,000	More accepted words you'd never guess
Gain from 20 to 60	6,000 more	About one new lemma every 2 days

Look at the spread. Two people can both be fluent adult native speakers and still be 25,000 lemmas apart. So when a friend says the puzzle was easy and you ground to a halt at Amazing, you weren't necessarily playing worse. You were partly playing a different puzzle, because the overlap between the day's answer list and your own vocabulary isn't the same overlap as theirs.

The practical takeaway is about where your time goes. Arguing with a rejection costs you the one thing the puzzle actually rewards, which is sweeping the letter pairs methodically. Log the word, move on, and check it after you're done.

But do check it afterwards. That same study found the average person picks up about 6,000 lemmas between 20 and 60, roughly one new word every two days. A puzzle that hands you an unfamiliar accepted word has just done that job for you, on schedule.

Can a hint tell you what's missing without spoiling it?

Yes, and this is the option most players skip straight past on their way from unaided play to a full solver. The NYT publishes its own hints page every day, and it hands you counts rather than words.

Two things sit on it. A grid with starting letters down the side and word lengths across the top, where each cell holds the number of words of that length starting with that letter. And below it, a two-letter list: every valid opening pair in today's puzzle with a count next to it. If it says GL and a 3, there are exactly three words beginning GL and no more.

Nothing there is a word. It's a map of where the words are.

Why counts unlock words you already know

Because the problem the section above describes isn't a vocabulary problem. It's a retrieval problem, and there's a classic experiment that separates the two cleanly.

Endel Tulving and Zena Pearlstone published "Availability versus accessibility of information in memory for words" in the Journal of Verbal Learning and Verbal Behavior (1966, vol. 5, pp. 381 to 391). Participants learned lists of categorised words in a single pass, with lists of 12, 24 and 48 words. Then they recalled them either with the category names supplied as cues or without.

Cued recall beat uncued recall, and the gap grew with list length. But the part that matters here is what happened next.

Everyone took a second recall test, this time with cues available to all. For the groups already cued the first time, nothing moved: mean recall was 21.17 on the first test and 21.20 on the second, which the authors describe as neither forgetting nor reminiscence. For the previously uncued groups, second-test scores were significantly higher on every list but one.

Same people. Same words. Minutes apart. Nobody learned anything in between. The only thing that changed was that a retrieval structure appeared, and words that had been sitting there came out. Tulving and Pearlstone's conclusion is the sentence to keep: those words were "available in the memory storage, but not accessible for retrieval."

That is exactly your position at 40 words with the grid unopened. The remaining words are almost all words you know.

How to use it without wrecking the puzzle

The trick is to take the smallest cue that unsticks you rather than the biggest one available.

• Start with the totals, not the grid. Just knowing there are 34 words and you have 22 tells you whether to keep going or stop. Sometimes that's all you needed.
• Open one row, not the whole thing. Check the letter you've found fewest words under. One row is a nudge. The full grid is close to a solution.
• Use the two-letter list last. It's the strongest cue on the page, because a pair plus a length usually leaves very few candidates.
• Chase the cells with a gap, not the ones you've cleared. The grid's real value is negative information: it tells you where to stop looking, which is what frees the attention the plateau was eating.
• Look before you're exhausted, not after. Once you've given up you tend to read the grid as an answer key. Ten minutes earlier you'll read it as a prompt.

And it reframes the question in the next section. The choice isn't unaided play versus looking up the answers. A count is a cue, a cue is not an answer, and the research above says a cue is enough to reach most of what you were missing anyway.

How many words does a typical Spelling Bee actually have?

Somewhere between 25 and 75 on an average day. That range is probably the most useful number on this page for calibrating what you should expect of yourself.

The section just below works the Genius gap out using an invented forty-word puzzle, which is fine for showing the arithmetic. But it's worth seeing the real spread first, because it's wider than most players assume.

WordsRated, which has catalogued 2,832 puzzles as of 3 February 2026, puts the everyday range at 25 to 75 valid answers, and notes that most puzzles carry one or two pangrams. So one perfectly normal Bee can be three times the size of another perfectly normal Bee.

The extremes stretch further. The biggest puzzle in that dataset ran to 81 words and 371 points, set on 8 June 2019. The smallest was 16 words and 47 points, on 27 March 2023. Same game, same seven letters, and one carries nearly eight times the scoring weight of the other.

Now sit with what that does to the rank names. Genius is 70 percent of the day's points, so on the 371-point puzzle you'd need roughly 260 to reach it, and on the 47-point puzzle about 33. Those are not remotely the same achievement and they get handed the same label. A Genius on a thin Tuesday and a Genius on a fat Saturday tell you almost nothing comparable about how well you played.

Two things follow that are worth carrying into tomorrow's puzzle.

• Stop reading your rank as a score against yesterday. It's a percentage of a total that moves underneath you. A run of Genius days says at least as much about puzzle size as it does about your form.
• Judge a stall against the puzzle, not against a number in your head. Stopping at 20 words feels like failure if you're assuming every puzzle holds 50. On a 16-word day, 20 would have been the entire list and then some.

And that's really why the sweep earlier on this page is written as a procedure rather than a target. You can't know at the start whether today holds 25 words or 75. The only part you actually control is whether you searched the space properly.

Why does Genius still leave words on the board?

Because Genius is 70 percent of the points, not 70 percent of the words. Those two numbers are nowhere near each other.

The scoring is what pulls them apart. A four-letter word scores 1 point. Every longer word scores a point per letter. And a pangram takes a 7-point bonus on top, which is why a seven-letter pangram is worth 14. So one pangram outscores fourteen four-letter words on its own.

Now put that next to something this page said earlier: four-letter words are the densest part of the answer list. The points sit in the long words. The word count sits in the short ones. Chase points and you are deliberately skipping most of the list.

Here's a made-up puzzle to show the size of the gap. The shape is typical even though the numbers are invented, and you can check the arithmetic yourself.

Word length	How many	Points each	Points total
Four letters	20	1	20
Five letters	12	5	60
Six letters	6	6	36
Seven letters	1	7	7
Seven-letter pangram	1	14	14
Whole puzzle	40 words		137 points

Genius lands at 70 percent of 137, so 96 points. Now find the cheapest route there. Take the pangram (14), the other seven-letter word (7) and all six of the six-letter words (36). That's 57 points from 8 words. Add eight five-letter words at 5 apiece and you're on 97.

Sixteen words. Out of forty. You just hit Genius with 24 answers still sitting on the board, and every one of the twenty four-letter words untouched.

That's the honest reason a Genius badge can sit next to a half-empty grid. The rank was never measuring what you think it measures.

So what is finding every word called?

Queen Bee. It's the level past Genius, and it means exactly what the title of this article promises: the complete list, nothing left.

And it is not a slightly harder Genius. It's a different job. Genius rewards a small number of high-value finds, which is what all the pangram hunting and pair sweeping above is built for. Queen Bee gives you no credit for cleverness, because it needs every single one-point four-letter word too. High-yield strategy stops helping the moment the target becomes exhaustive.

Why do the last few words take longer than all the rest?

Because that's the shape memory retrieval has always had, and it was measured long before anyone built a puzzle out of it.

Bousfield and Sedgewick ran the experiment in 1944. They gave people a category, animals or cities, and 18 minutes to name as many members as they could, marking where they'd got to every 2 minutes. The plot of cumulative answers against time bends. Most of your words arrive early, then the rate keeps falling.

They derived a function for it that researchers still use. Richard Hass and Roger Beaty set it out in Frontiers in Psychology in 2018, describing a cumulative response curve that "flattens out (reaches an asymptote) as time grows," in the form R(t) = a(1 - e^(-t/τ)), where a is "the 'asymptotic level of responding' or the total number of items available for retrieval."

Their account of why is the useful part. They describe it as repeated sampling of semantic space during memory retrieval, which is then depleted, leading to more false retrievals and an exponential slowing of retrieval rate.

Read that against your own last twenty minutes with the Bee. The slowdown isn't fatigue and it isn't you getting worse. It's the curve doing what the curve does, and it would happen to a well-rested expert too.

Which is exactly why the systematic sweeping earlier on this page earns its keep. Free recall is the thing that gets depleted. A structured pass through letter pairs isn't recall at all, it's enumeration, and enumeration doesn't flatten the same way. When the words stop arriving on their own, that's the signal to switch methods rather than to try harder at the method that's run dry.

So decide which game you're playing before you lose an hour to it. If you want Genius, hunt the pangram and the long words and accept that plenty of short ones will survive you. If you want Queen Bee, put the clever stuff down and start enumerating, because from there it's a completeness problem and nothing else.

Should you solve the Bee with someone else?

Yes, but not the way most couples and group chats do it. Shouting words across the kitchen feels productive and finds fewer of them. Working alone first and then pooling your lists finds more. That's one of the most replicated results in the psychology of groups, and the Bee is almost a lab version of it.

The name for it is collaborative inhibition. Put people together to recall a list and the group remembers less than the same people would if they'd each worked alone and you'd combined their answers afterwards. Marion and Thorley pulled the research together in a meta-analysis for Psychological Bulletin (2016), covering 75 effect sizes from 64 studies. They called collaborative inhibition "a robust effect", and it got stronger in exactly the conditions a shared Bee creates: bigger groups, free-flowing turn-taking, and answers called out in any order.

Their explanation lines up with something earlier on this page. Everyone builds their own route through a search, and hearing someone else's words knocks you off yours. It's the same trap as the found-words list staring back at you, except now the cues arrive out loud from another person, on their schedule rather than yours.

The brainstorming literature says the same thing about generating ideas rather than remembering them, which is closer to what the Bee actually asks of you. Diehl and Stroebe ran four experiments for the Journal of Personality and Social Psychology (1987) to work out why groups underperform. It wasn't mainly people slacking off or feeling judged. Their fourth experiment found that production blocking accounted for most of the loss: while one person talks, everyone else has to hold their idea and wait, and some of those ideas don't survive the wait. A meta-analysis by Mullen, Johnson and Salas in Basic and Applied Social Psychology (1991) added that the loss was stronger in larger groups and when contributions were spoken rather than written down.

So the fix isn't to stop playing together. It's to change the order.

• Split first. Each person works the puzzle on their own for ten or fifteen minutes. No reading lists out, no "ooh, what about".
• Write, don't call out. Speaking your words is the condition that made the loss worse. Typing them into your own copy of the game doesn't block anyone.
• Then merge. Compare lists at the end. The words only one of you found are the whole point, and they tend to be the ones that show you a pattern the other person's search skipped.
• Keep it small. Two people pooling lists beats five people in a chat, and the research says the gap gets worse as the group grows.

And there's a reason to merge rather than just compare scores. Marion and Thorley's separate analysis of what happens after a group session, based on 27 effect sizes, found that remembering together generally helps people's later recall on their own. In Bee terms, the words your partner found are worth reading properly once you've done your own sweep. They can nudge your next solo session, even if they'd have got in your way during this one.

One honest caveat. These are lab studies of recall lists and brainstorming tasks, not of word puzzles, and none of them tested the Bee. But the mechanism is general, and it costs nothing to try: play one day shouting across the room and one day split then merge, and see which gets you further. If you like head-to-head word games more than shared ones, our Connections strategy guide covers a puzzle where comparing notes works differently, since there's a fixed set of answers to find.

When should you stop and use a helper?

The honest answer is that it depends on what you're playing for, and there's no single right call.

If Spelling Bee is your daily unaided test, a helper ruins it and you already know that. But if you play to widen your vocabulary, there's a real argument for checking after you're genuinely out of ideas. The words you'd never have produced are exactly the ones worth learning, and seeing them attached to today's letters makes them stick better than reading a list.

A reasonable middle path: play until you stop finding anything for five straight minutes, then look. You keep the challenge and still learn the words you were missing. If you like this kind of structured puzzle thinking, our NYT Connections strategy guide applies the same logic to a different daily game, and the best Wordle starting words covers how letter frequency drives opening choices there.

Does playing word puzzles actually help your brain?

There's decent evidence, though it needs reading carefully.

A randomised controlled trial led by D.P. Devanand of Columbia University with Murali Doraiswamy of Duke University, published in NEJM Evidence on 27 October 2022, compared web-based crossword puzzles against computerised cognitive games in 107 adults with mild cognitive impairment, average age 71. As Columbia University's Department of Psychiatry reports, the crossword group showed greater improvement on the ADAS-Cog measure at both 12 and 78 weeks, better daily functioning at 78 weeks, and less brain shrinkage on MRI.

Separately, the PROTECT study run by Dr Anne Corbett at the University of Exeter Medical School surveyed more than 19,000 participants aged 50 and over and published its findings in the International Journal of Geriatric Psychiatry in May 2019. Regular word-puzzle players performed on grammatical reasoning tests at a level equivalent to someone ten years younger, and on short-term memory tests equivalent to eight years younger.

One detail is worth holding onto before you take any of that to heart. Both studies looked at people aged 50 and over, and the Columbia trial recruited only people who already had mild cognitive impairment. Neither tells you what a daily Spelling Bee does for a healthy 30 year old. That question hasn't really been answered, and anyone claiming otherwise is filling in a blank the research left empty.

Corbett was careful about what that does and doesn't show, stating that researchers "can't say that playing these puzzles necessarily reduces the risk of dementia in later life". The Exeter work shows association, not cause. The Columbia trial is the stronger design because it randomised people. Together they suggest word puzzles are a reasonable thing to do with fifteen minutes, not a medical intervention. Our piece on word game brain benefits goes further into what the research supports.

What do players ask about Spelling Bee strategy?

How many points do you need for Genius in Spelling Bee?

Genius sits at 70% of the day's total available points, so the raw number changes with every puzzle. A puzzle worth 200 points needs 140. Because the threshold is a percentage rather than a fixed score, finding the pangram matters more on small puzzles, where its bonus is a larger slice of the total you need.

What is the fastest way to find the pangram?

Write the seven letters out and sort them into consonants and vowels, then look for a familiar stem that already uses the awkward ones. Pangrams usually hide behind a prefix or suffix that absorbs leftover letters, so test the outer letters against endings like -ING, -TION, -MENT, -ABLE and -OUS before you try anything exotic.

Why does Spelling Bee never use the letter S?

The puzzle excludes S by design. With S available you could pluralise almost any valid word and double the answer list without knowing any more vocabulary, which would flatten the challenge. Leaving it out forces the word count to come from actual word knowledge rather than from adding one letter to things you already found.

Are four-letter words worth finding in Spelling Bee?

Yes, and they matter more than their score suggests. Each is worth only 1 point, but they are the densest part of the answer list and they reveal which letter pairs actually work in that day's puzzle. A four-letter word is often the stem you extend into a six or seven-letter word worth far more.

Is using a Spelling Bee helper cheating?

That depends entirely on why you are playing. If the puzzle is a daily test you want to pass unaided, a helper defeats the point. If you play to learn words, checking what you missed after you have run out of ideas is how the unusual entries actually stick. Most players land somewhere in between and use one only when stuck.

Sources: Tulving, E. and Pearlstone, Z. (1966), "Availability versus accessibility of information in memory for words," Journal of Verbal Learning and Verbal Behavior 5(4):381 to 391, for the categorised-list method across 12, 24 and 48-word lists, cued recall exceeding non-cued recall with the gap growing by list length, the second recall test showing cued groups unchanged at means of 21.17 and 21.20 while previously uncued groups improved significantly on all lists but one, and the conclusion that the unrecalled words were available in memory storage but not accessible for retrieval. Peter Norvig, "English Letter Frequency Counts: Mayzner Revisited" (norvig.com), analysing the Google Books Ngrams corpus, for both the single-letter frequencies and the bigram table, including the 2,819,662,855,499 letter pairs counted, the 669 distinct bigrams observed out of 676 possible, and the seven that never occur. Those bigram shares are token frequencies from running text rather than counts of how many distinct words contain each pair, so use the ranking and ignore the decimals. Devanand et al., "Computerized Games versus Crosswords Training in Mild Cognitive Impairment", NEJM Evidence, 27 October 2022, via Columbia University Department of Psychiatry. Corbett et al., PROTECT study, University of Exeter Medical School, International Journal of Geriatric Psychiatry. Troyer, Moscovitch and Winocur, "Clustering and switching as two components of verbal fluency: evidence from younger and older healthy adults", Neuropsychology, 1997, 11(1), 138-146, PMID 9055277 (pubmed.ncbi.nlm.nih.gov). Slamecka, N. J., "An examination of trace storage in free recall", Journal of Experimental Psychology, 1968, 76, 504-513, for the original part-list cuing finding; recall percentages and effect sizes are from the open-access replication at PMC8642343, which is Wallner, L. and Bäuml, K.-H. T., "Self-paced part-list cuing", Psychonomic Bulletin and Review, 2021, 28(6), 2012-2018, and is also the source for the self-paced condition figures of 46.87 percent at d = 0.13 and 52.47 percent at d = 0.03. Those were list-learning experiments rather than puzzle play, so the mechanism transfers but the numbers do not. Novick, L. R. and Sherman, S. J., "On the nature of insight solutions: evidence from skill differences in anagram solution", Quarterly Journal of Experimental Psychology Section A, 2003, 56(2), 351-382, PMID 12613568 (pubmed.ncbi.nlm.nih.gov), for the Experiment 1 pop-out finding and the proposed split between parallel processing of letter-order constraints and serial hypothesis testing. Those were laboratory anagrams with a guaranteed solution, so the mechanism transfers to a perfect pangram far more cleanly than it does to open word hunting. Not affiliated with The New York Times Company. Marion, S. B. and Thorley, C. (2016), "A meta-analytic review of collaborative inhibition and postcollaborative memory: Testing the predictions of the retrieval strategy disruption hypothesis," Psychological Bulletin 142(11):1141 to 1164, PMID 27618544, for the 75 effect sizes from 64 studies, collaborative inhibition as a robust effect, its enhancement in larger groups and under free-flowing, free-order procedures, and the separate 27-effect-size analysis finding that collaborative remembering generally benefits later individual retrieval. Diehl, M. and Stroebe, W. (1987), "Productivity loss in brainstorming groups: Toward the solution of a riddle," Journal of Personality and Social Psychology 53(3):497 to 509, for the four experiments and the finding that production blocking accounted for most of the productivity loss of real groups. Mullen, B., Johnson, C. and Salas, E. (1991), "Productivity Loss in Brainstorming Groups: A Meta-Analytic Integration," Basic and Applied Social Psychology 12(1):3 to 23, for brainstorming groups being less productive than nominal groups and the stronger loss in larger groups and with spoken rather than written contributions. None of these studies tested word puzzles, which the section says.