The most common answers in crosswords are changing

Start with a single number. ERIE is the third most common answer in American crosswords. It is also the 33,389th most common word in English. There are more than thirty thousand words you are more likely to meet in ordinary life than the name of a mid-sized lake, and yet inside a grid it outranks almost all of them.

That gap is the whole subject of this article. It has a name among solvers — crosswordese — but it is usually described rather than measured. It can be measured, and once you measure it two things become visible that a plain frequency list hides: crosswordese lives almost entirely in the shortest slots, and it is being replaced.

What we counted

The data is the xd corpus, an open archive of American crossword clues and answers. After cleaning it gives 8,006,563 answer usages and 368,549 distinct answers, spanning 1913 to 2026 and 46 publications. The New York Times is the largest single source at 2.46 million usages, followed by the Los Angeles Times (834K), Newsday (819K), Universal (714K) and USA Today (684K). So this is not a portrait of one newspaper's taste; it is American crosswords broadly.

For how common a word is in ordinary English we used LexiLab's own 46,832-word English frequency list, which is built from everyday written and spoken language and has nothing to do with puzzles. It matters that these two measurements are independent: one counts grids, the other counts language, and neither is derived from the other. Crosswordese is what you get when you subtract the second from the first.

One consequence of the first measurement is worth flagging, since it changed a tool while this article was being written. The Crossword Helper now ranks its suggestions by how often each word has actually been used as a crossword answer, rather than by how common it is in English — which is why it puts ERIE ahead of BRIE. The English list above is still involved, but only as one half of the ranking, never as the whole of it.

The 25 most common answers

Ranked by raw appearances, with each answer's rank in ordinary English alongside.

Two things stand out immediately. Every single entry is three or four letters long. And the list mixes two completely different kinds of word: ONE, ARE and IDEA are among the most common words in the language, while ERIE, ALOE, ALA and IRE are nowhere near it. Raw frequency cannot tell them apart. That is what the next section is for.

Measuring crosswordese

The intuition is simple: divide how often a word turns up in grids by how often it turns up in English. A word that is common in both, like AREA, scores low. A word that is common in grids and rare in English scores high. That ratio is crosswordese, expressed as a number.

One detail is worth flagging, because it changes the answer. The grid side is counted; the English side has to be estimated from a rank list. What the measure does not do is compare the two ranks directly, which is the obvious shortcut and gives a badly wrong result — ranks compress at the head of the distribution, so ERA and OREO differ by 1.7× in appearances but by 28× in rank. The exact formula is in the method note.

This behaves the way a solver's instinct says it should. ERA, the single most repeated answer of all time, falls to 557th on this measure, because ERA is a perfectly ordinary English word that happens to be three letters of convenient shape. AREA falls to 1,441st. Meanwhile OREO rises into the top five.

The most extreme tier is the one where the right-hand column gives up entirely. 158 of the 2,000 most common answers are real dictionary words that fall outside the frequency list altogether — rare enough in everyday use that a list of the 46,832 most common English words does not reach them. EPEE, ASEA, OLEO, ALEE, APSE, ALIT, STET, ECRU, OLIO. Every one of these has a life outside puzzles — a fencer really does pick up an épée, an editor really does write stet — but for most solvers the crossword is where they meet them.

Crosswordese is a disease of short entries

Sort every answer ever printed by length, and the distribution is startlingly lopsided.

Now walk up the lengths and watch what happens to the answers themselves.

The most constrained slots in a crossword hold the least familiar language. The least constrained slots hold the most familiar.

This is not a coincidence, and it is not a matter of taste. It falls out of the geometry. A three-letter answer in a densely interlocked American grid is crossed on every single letter: all three of its letters must simultaneously work as part of three other answers running the other way. The pool of strings that can satisfy three constraints at once is small, and it skews heavily toward vowel-rich, consonant-light shapes. That is exactly the shape of ALOE, ASEA, OLEO, EPEE. Constructors do not use these words because they like them. They use them because at that length there is very little else that fits.

A fifteen-letter answer sits under the opposite pressure. It is usually a theme entry, chosen first and built around, with the rest of the grid arranged to accommodate it. It is selected to be lively — a phrase a solver will enjoy uncovering. So the entries with the most freedom get the most familiar language, and the entries with the least freedom get the strangest.

The turnover

Everything so far describes crosswords as a single static object. They are not. And once you look at the same answers decade by decade, the most interesting result in the whole dataset appears.

This part needs care. Before 1994 the corpus is 99.7% New York Times, and broadly multi-publication afterwards, so a naive comparison across eras would mostly measure the archive changing shape rather than crosswords changing. To avoid that we restricted this section to the New York Times alone, which happens to be almost perfectly steady: between 308,000 and 333,000 answer usages in every decade from the 1960s to the 2010s. We then compared each answer's share of its decade, not its raw count.

Read the two lists side by side and the pattern is hard to miss. The fading words are Latin, heraldry and classical architecture: ITER is a Roman road, STOA a Greek colonnade, ORLE a heraldic border, ANTA a pilaster, ESNE an Anglo-Saxon labourer, ADIT a mine entrance, ANIL an indigo shrub. The rising words are agencies, currencies, sports bodies, consumer brands and internet-era interjections.

The letter-shape pressure never changed — a constructor in 1965 and a constructor in 2015 both need vowel-heavy three-letter fill. What changed is the body of knowledge a puzzle assumes its solver has. Mid-century American crosswords were built for a reader with a classical education. Contemporary ones are built for a reader fluent in brands and acronyms. Crosswordese did not go away; it swapped registers.

When it happened

Decade buckets imply a slow tide. Plotted year by year, the change is not gradual at all: almost all of it lands inside a single year.

The chart below tracks thirteen classical crosswordese answers as a share of all New York Times answers: ITER, STOA, ESNE, ANTA, ORLE, ANIL, ADIT, SNEE, SMEE, ANOA, ETUI, SPAE and EDILE — the Latin, heraldry and dead-vocabulary tier.

The rate sits between 23 and 39 per 10,000 for every one of the nine years to 1993, then lands at 9.0 in 1994 and never returns. Across the whole Maleska-era window it averages 30.3 per 10,000; across 1994–1999, 9.8. Sixty-eight per cent of a category of vocabulary disappeared, essentially at a stroke.

The timing points somewhere specific. Eugene T. Maleska, who had edited the puzzle since 1977, died in August 1993. Will Shortz's first puzzle ran on 21 November 1993 — so 1993 is almost entirely his predecessor's work and 1994 is the first full year of his. He arrived with an explicit view of what was wrong: as he put it later, "the clues weren't playful, the answers weren't modern."

A change in what the archive contains would produce exactly the same-looking cliff, so that has to be ruled out first. It doesn't survive the check:

That last row is quietly the most telling. Mean answer length rose and the number of distinct answers rose in the same year — a puzzle with longer entries and a wider vocabulary. And the geometric pressure this whole article is about did not go anywhere: the vowel-heavy pillars were used more in 1994, not less. The grid still demanded its ERIEs. What stopped was reaching for ESNE to get them.

The geometry stayed exactly as demanding. What changed was which words counted as an acceptable way to satisfy it.

One caution, and it is a real one. We cannot fully separate "the New York Times changed editors" from "American crosswords changed in the mid-1990s," because no other publication in this corpus has enough pre-1994 material to act as an external control — before 1994 the archive is 99.7% New York Times. Digital grid-filling software also became widely available in this period, and would independently have made obscure fill easier to avoid. What the data does establish is the shape of the change: a step, not a slope, at a moment when one publication changed editors. Gradual technology adoption does not produce a 62% drop between one December and the next January.

Note also what the chart shows after the break. The cliff did not finish the job — it started a long decline. The same thirteen words run at 7.9 per 10,000 through the 2000s, 6.4 through the 2010s, and 1.7 across 2020–2025. Against a Maleska-era baseline of 30.3, the classical tier is now running at about one-eighteenth of its former rate. ITER, STOA, ESNE, ANTA, ORLE and ADIT have not appeared in a New York Times puzzle since 2019.

The pillars do not move

One more result deserves a mention, because it cuts against the story. The very top of the list is remarkably stable. ERIE shifted by a factor of 0.79 between those eras, AREA by 0.88, ALOE by 0.94, ERA by 1.24. Essentially flat. The turnover is happening in the second tier, among the specialists. The handful of answers that solve the hardest geometric problems in the grid are apparently irreplaceable, and sixty years of changing taste has not dislodged them.

Run those same four answers through the periods used in the section above and the contrast is stark. Combined, they appear 25.8 times per 10,000 before 1994, 27.0 in 1994–99, 23.4 in the 2000s, 21.1 in the 2010s and 25.9 in 2020–25 — no trend at all, across the exact window in which the classical tier fell by a factor of eighteen. Whatever the 1994 break was, it was aimed at a specific register of vocabulary, not at short fill in general.

What this is useful for

The 25 above are the head of a longer list. The 200 most common crossword answers are collected on their own page, ranked on the post-2000 window and shown with the clue each one is given most often.

If you solve: the payoff of learning crosswordese is unusually concentrated, but it now has a shelf life. A few dozen short, vowel-heavy answers still cover a startling proportion of the fill you will meet, and the stable tier really is stable — ERIE and ALOE are as safe a bet today as in 1970. The classical tier is a different matter: if you learned ESNE and ANTA from a puzzle book, you learned words that a modern grid has essentially stopped using. And when you have a pattern and no idea, that is a search problem rather than a memory problem — which is what a pattern-based word finder is for.

If you build: the same pressure acts on any grid you make, including a small one built from a vocabulary list. The shorter and more interlocked your grid, the harder it pulls toward strange fill. The practical lever is geometry rather than word choice — a slightly more open grid, or one or two fewer crossings, does more for the quality of your fill than hunting for better three-letter words. That is worth knowing before you build a puzzle of your own, because it is much easier to loosen a grid than to rescue one that is already too tight.

More generally: crosswordese is worth measuring rather than just complaining about, because the measurement says something a word list cannot. It locates the problem in the geometry of the grid rather than in the taste of the people filling it, which is why the same handful of vowel-heavy answers survives every change in fashion. And it shows that the part which is taste — the classical tier, the Latin and the heraldry — turns out to be the part that can move, and has.