Inside the Mind of a Crossword Constructor: Q&A

It's almost always theme first — and specifically, it's a wordplay observation that makes me laugh or go "huh, that's interesting." I'll be reading something, or in a conversation, or just thinking about language, and I'll notice that a phrase works in two unexpected ways, or that if you add a word to the front of several other words you get a funny sequence. That's the seed. From there, I ask myself: is there a puzzle in this? Can I find five or six examples of this same pattern? Are the examples interesting, fresh, varied? If yes, I open the software. If not, I put the idea in a notebook and move on.

Chaotic. There's a notes app on my phone where I capture things on the go, and there's a physical notebook on my desk for when I'm brainstorming intentionally. The phone notes are usually just a phrase and three question marks — like "BUTTERFLY EFFECT... butterfly stroke? butterfly knife? ???" I'll come back to it weeks later and either it clicks or it doesn't. I'd say about one in eight ideas makes it into an actual grid. The rest are either not quite right or, depressingly often, already done. It's heartbreaking to develop an idea for two weeks and then discover someone published that exact theme five years ago.

Most professional constructors use one of two pieces of software: Crossword Compiler or Crossfire. I use Crossfire. It's a grid-building environment that also contains a word list — a database of tens of thousands of words and phrases scored by frequency and crossword-fitness. Once I've placed my theme entries in the grid, I use the auto-fill function to see what the software suggests for the remaining fill. Then I start making decisions. Some auto-fill solutions are great and I keep them. Others are technically valid — they're real words — but they're obscure or stilted or just kind of ugly. Those I replace manually.

The grid itself — from placing the theme entries to having fill I'm satisfied with — is usually 3 to 8 hours, spread across a day or two. The clue writing is another 4 to 6 hours. Then there's test-solving: I send it to two or three trusted friends who solve and give me feedback on difficulty, whether any clues are unclear or unfair, whether the theme works the way I intend. That feedback process adds another day or two before I'm ready to submit. So total, from grid open to submission, it's usually about a week for a themed puzzle I care about. A themeless can take much longer — sometimes a month — because the fill standards are so much higher and there's no theme to give the solver a handhold.

Solvers detect bad fill primarily through two signals: encountering words they've never heard of and can't parse from crossings, and encountering a lot of "crosswordese" — those short, obscure words that are technically valid but feel stale. ESNE, ETUI, ALAE, that kind of thing. Experienced solvers can fill them in on autopilot, but they know they're not learning anything from those answers — they're not surprised or delighted. Good fill does two things simultaneously: it gives the solver clear paths to confident letter-fills through crossing patterns, and it occasionally delivers an answer that makes them grin — a fresh phrase, a vivid word, something they hadn't thought of in years. The best fill entertains. The worst fill just technically completes the grid.

Always. The theme entries determine what letters go where in the grid, and that constrains what fill is even possible in the surrounding areas. If your theme requires the letter sequence X-K-J to appear in a certain position, the software is going to struggle to find good crossing words. This is why constructors often end up compromising: they choose between a really elegant, specific theme entry that causes fill problems, or a slightly less perfect theme entry that allows for much cleaner surrounding fill. I always prefer cleaner fill. The solver spends most of their time in the fill, not the theme. If the fill is painful, the puzzle is a chore even if the theme is brilliant.

It really is a different skill. Grid construction is more like engineering — you're solving a constraint problem with clear rules. Clue writing is creative writing. You're trying to find the most interesting, surprising, or funny way to describe a word while keeping it technically fair. There's a whole craft vocabulary around clues: misdirection, punning, the "?" at the end to signal wordplay, the difference between indirect clues and direct clues. I often write three or four candidate clues for each answer and then choose the one that fits the puzzle's tone — is this a Monday puzzle where I want accessible, direct clues? Or a Thursday where I can be more cryptic and tricky?

Oh, this is dangerous territory because the best clues are never the ones you're proud of — they're the ones the editor keeps, which is sometimes not the same thing. But I had a clue a few years ago for the answer SPINE that read "It supports a lot of ideas." On the surface, that's just a description of the spine (it supports the body). But it also works metaphorically for a book's spine (which supports the binding), and it works as a pun on "backbone" in the sense of having ideas worth standing up for. Three layers in six words. The editor kept it, which means it cleared the bar of being fair at its first reading while rewarding a second look. That's the target.

The process varies by editor. With the NYT, you submit the grid and clues and wait — sometimes weeks, sometimes months. If the editor accepts, they'll often send back a revision request: maybe 10 to 20 clue changes, occasionally a request to revise a section of fill. The clue changes are sometimes improvements I genuinely love, and sometimes substitutions I think are worse than my original. You learn to pick your battles. Fill revision requests are harder — the editor is usually right that a section needs work, but fixing it while preserving the surrounding crosses requires real puzzle-solving on your part.

For a new constructor starting out, the rejection rate at the NYT is extremely high — probably 90% or more. Even experienced constructors like me get rejected regularly. The NYT publishes around 365 puzzles a year and receives many, many more submissions than that. The rejection can be for any number of reasons: the theme idea isn't strong enough, the theme has been done before, the fill isn't up to standard, the topic isn't right for that moment. It's a frustrating process even when you know it's coming. I keep a spreadsheet of submissions: date sent, publication, outcome. Some puzzles sat in submission limbo for over a year before I got a decision.

The aha moment is the reason crosswords exist. It's that specific feeling when something that was opaque becomes suddenly, unmistakably clear — when the theme reveals itself, or when a clue you couldn't parse suddenly makes sense, or when filling in one answer unlocks three others in cascade. It's a small dopamine hit, and the best puzzles engineer it deliberately. For theme puzzles, the aha moment usually comes when the solver encounters the second or third theme entry and suddenly understands what the theme is — that's when they start scanning for more examples and the grid opens up. For themeless puzzles, the aha moments are distributed — they come from individual entries and clues that surprise you. I think about the aha moment constantly when I'm constructing. I ask myself: where will the solver first understand what I'm doing? How can I make that moment as satisfying as possible?

Download Crossfire or Crossword Compiler. There are trial versions of both. Make a small grid — start with a 9x9 or an 11x11, not a 15x15. Pick a simple theme: five words that all contain a color, or five phrases that end with a musical instrument. Get the theme entries in the grid first. Use the auto-fill to see what's possible. Then start making it better. Don't worry about submitting anywhere. Your first puzzle doesn't need to be publishable — it needs to teach you how hard this is, what the constraints feel like, and whether you enjoy the process despite the difficulty. If you solve and finish that first grid and feel satisfied, keep going. If you feel only frustrated, that's valuable information too. The craft rewards persistence more than innate talent.

Three things: First, read Rex Parker's blog and Amy Reynaldo's Crossword Fiend for years of annotated puzzle discussion — you'll learn by osmosis what makes clues work and fail. Second, follow constructor Twitter and the crossword Discord communities — the community is extraordinarily generous with advice and feedback. Third, just solve a lot. Hundreds of puzzles. Thousands. Notice how the themes work, how the fill changes difficulty, how different editors have different aesthetics. Construction is solving with the constraint engine flipped around. The more you understand the solver's experience, the better your puzzles will serve them.