As a hobbyist constructor who isn’t surrounded by puzzle-fiends, I get this question kind of a lot. I would imagine other constructors probably get asked it a lot too. It’s an interesting question because the typical guess I get from people who ask this question actually couldn’t be further from the truth! As such, I thought it might be a fun topic for a blog post [with some updates now that my second New York Times Sunday puzzle is coming up in January 2021]!
Usually, when I get asked this question, the asker then volunteers a guess along the lines of: “So, do you just start with a list of clues and…?” This assumption–that I must start by writing clues–makes sense! When you see a crossword in print, much of the page is devoted to the clues, and it’s what the solver engages with “first,” so it’s reasonable to assume that most of a constructor’s effort must revolve around them. Well, no. The truth may surprise you: When you solve one of my puzzles, the clues were the part of that puzzle I did last and spent the least amount of time on (relatively speaking), and many of the clues won’t even be the ones I wrote! To explain, I need to walk you through how I construct a themed crossword puzzle. Brace yourself—thar be dragons ahead!
Step one is coming up with a theme in the first place.
Let me be clear: Crosswords don’t need to have themes. The books of puzzles you get in the grocery store, for example, may be full of unthemed puzzles–these are puzzles with grids whose answers are just whatever would fit, and there’s no cohesive idea holding any of the entries together. Also, many prominent venues, like the New York Times, run “themeless” crosswords once or twice weekly. Those themelesses are noteworthy for having grids that are “wide open,” in which many long white regions intersect and there are many fewer black squares than normal–which make them much more challenging to both solve and construct!
However, most prominent venues require many of their daily puzzles to have a theme—something that thematically unites several (but by no means all) answers in the puzzle. By convention, these are usually the longest 3-5 answers (but sometimes more or less) in the across direction, but that depends on the particular theme.
Themes can take a lot of forms–it’s nearly impossible to generalize accurately about them. For example, a daily puzzle a while back in the Wall Street Journal by constructor Alex Eaton-Salners (one of my favorites) asked solvers to CROSS BREEDS—in four places in the puzzle, the names of two dog breeds crossed each other (one going across, one going down). A puzzle I co-wrote with Jeff Chen for the Uptown Puzzle Club once asked solvers to put the whole word BOX (as part of answers like CEREAL BOX and BOX GRATER) inside of a single white box 12 times in the puzzle. These twelve BOX squares were themselves arranged in the shape of a 3D box! Many themes involve wordplay of some kind. One of my puzzles in the Uptown Puzzle Club had a theme where I added an O to the ends of common phrases to make silly phrases that I then clued with a literal but silly clue. For example, “Like a sexy ‘hi’?” was the clue for HOTTER THAN HELLO. Puns, celebrations of a pastime like baseball, reading words “incorrectly” on purpose–if you can do it with words, crossword themes can exploit it!
Because themed crosswords revolve so much around their theme, most themed crossword puzzles are rejected or rejected by puzzle editors on the basis of whether or not the theme is clever/fun/consistent/innovative/well-executed. As such, I spend hours (sometimes more than 24!) thinking about my theme before I ever bother moving past that point. Many themes get written in my “theme ideas” notebook and then lost, never to be realized.
Part of why themes occupy so much of my constructing energy is that I don’t come up with good themes very often when I’m trying to come up with good themes—the best ones tend to just come to me somehow, often when I least expect it. I have dreamed up a theme, for example! Often, they are inspired by an interesting phrase I come across, a quirk of the English language I notice, or a slip of the tongue. When I’m not quite so lucky and a theme hasn’t fallen into my lap, I come up with a gimmick (like adding Os to the ends of things) and then just see if I can make something fun of it.
Step two: Get a list of potential theme entries
Once I have a theme, the real fun starts. Let me walk you through how the post-theme process using one of my Los Angeles Times puzzles (screenshot below of the grid), which ran in March of 2018.
This puzzle revolves around a classic wordplay staple: homophones (words that sound the same but are spelled differently and have different meanings), specifically homophones that differ in just their first letter. It was inspired by a friend saying BRITISH ISLES out loud during a conversation. For a split second, my brain “heard” what he had said as BRITISH AISLES, which left me pretty confused! That quirk—that there are homophones that are as similar in spelling as AISLE and ISLE are—felt fun to me, so I started Googling to see how many such homophone pairs there are. Turns out there are several. Here are the ones I was able to find that I thought could have potential to help me build fun theme entries: NEW/KNEW, RITE/WRITE, ROTE/WROTE, ISLE/AISLE, CENT/SCENT, HOLE/WHOLE, OUR/HOUR, NIGHT/KNIGHT, NOT/KNOT, REST/WREST, and RING/WRING. One thing you’ll realize right away: Good theme brainstorming involves a LOT of list-making!
The next step was to find fun and common phrases/expressions that contain one of these two homophones that would still make some kind of sense if the normal homophone was switched for its pair. In other words, BRITISH AISLES still kind of makes sense; you could imagine such things existing, even if only in an alternate reality. We constructors call this “surface sense;” wordplay like this has to result in phrases that still make sense, even if only on the surface.
Step 3–I needed to constrain my theme.
It’s important for themes to have a consistent gimmick so that the solver can catch on to them and then use the gimmick to crack the other theme entries. So, I decided I always wanted to replace the shorter homophone in the phrases with the longer one, just like I would be doing with BRITISH ISLES –> BRITISH AISLES. Some homophone pairs gave me a lot of options for doing this. For example, NIGHT/KNIGHT provided a wealth of fun possibilities because NIGHT shows up in a lot of cool phrases/expressions. So I had options like AMATEUR KNIGHT, ONE KNIGHT STAND, KNIGHT SCHOOL, LADIES KNIGHT, and so forth. Other pairs left me with way fewer good options. For example, all I could come up with for REST/WREST was WREST ONE’S CASE, which only barely makes surface sense to me [“surface sense” is a quality that the products of plays on words may or may not have–something has surface sense when, even though it’s not a real thing in real life, you can still fairly easily imagine how those words or ideas could end up next to each other. For example, you can probably imagine what a “knight school” would be, even though there is no such thing!].
How did I find the phrases I just listed above, you might ask. Well, I do usually spent a little time just trying to come up with them on my own. Once that avenue is exhausted, though, I then move on to using onelook.com. Onelook is one of my all-time favorite constructing tools. What the site does is combine all the online dictionaries it has access to into one giant database that you can search by typing in particular expressions. For example, B??R would return entries found in at least one dictionary or database that start with a B, end in an R, and are four letters long (the ?s are wild cards that can stand in for any letter). So, it would return words like BEER, BOHR, BOOR, BEAR, and so on. You can also use the website to look for phrases. For example * NIGHT will return every phrase across all databases that ends in the word NIGHT (the * is a super wild card that can be any number of letters). So, it would return answers like SILENT NIGHT, WISH SOMEONE GOOD NIGHT, IN THE DEAD OF NIGHT, and so on. Then, mentally, I can sift through all these results, find charming phrases, and see if they have surface sense if I, say, swapped in KNIGHT for NIGHT. SILENT KNIGHT would make surface sense, but IN THE DEAD OF KNIGHT wouldn’t. In practical terms, if a phrase has surface sense, I can imagine a snappy, meaningful way to clue it (SILENT KNIGHT = Quiet member of King Arthur’s court?).
The process of coming up with a list of possible theme answers usually takes hours (or, perhaps, days). Once I have a list of them, the real decisions begin. First, I have to go through the list and decide which possibilities are actually, you know, any good: Is this one fun/funny/interesting? One criterion I use for wordplay themes like this I’ll call “departure value.” Basically, how much does the new meaning of the phrase differ (or depart) from the original meaning of the phrase, now that I’ve worked my workplay magic?? SILENT KNIGHT and SILENT NIGHT are sorta different…a night and a knight are not alike at all, although you could imagine them being “silent” in similar ways. So, this would score a medium on my “departure value” metric–the higher, the better (for me!). Other things I might consider include how colloquial something is, how familiar the original phrase is, how hard the resulting phrase might be to spell for a solver, and so forth. The initial round of filtering usually cuts my original list of possible theme entries in half, but sometimes it rules out the theme idea altogether.
Then, I have to do something that might seem weird to a non-constructor: I have to count how many letters each possible theme answer has.
Why? Keep in mind that most daily crossword puzzles are a 15 by 15 grid, meaning the maximum length a theme entry could be would be 15. WRUNG ON THE LADDER is 16, so it’d have to go. Also, some lengths (9 and 13 come to mind) are harder to “fit” for logistical reasons into a typical grid than other lengths (like 11 and 15)—this has to do with where they force black squares to be and how many such black squares they force. More on that later. But length matters for another key reason: Most crossword puzzles are rotationally symmetric with an order of 2. Oy! This means that if there is a black square or a 6 letter answer in the top-left corner, there is a black square or 6 letter answer in the exact same corresponding place in the bottom-right corner (thought of differently, you could turn a crossword upside-down and it would look the same). This is mostly to make puzzles look pretty, I think, but it also imposes a strict set of constraints on constructors. If I want to use a 12 letter entry like SILENT KNIGHT, I need to create a 12 letter long answer in the grid somewhere for it to go…maybe in the top-left region of the puzzle. Because of symmetry, though, this will force a 12 letter long answer in the bottom-right region too, which will also need to be filled with a theme entry (presumably). This means I need to find not just good theme entries but pairs of good theme entries, with each member of the pair being the same length. That way, I can put them into matching spots in the grid!
Even though 13 letter entries are tricky, I decided BRITISH AISLES was one of my favorite possible theme entries, which meant I had to find another 13 letter entry to pair it with. I had tons of options using the homophone pair NOT/KNOT, so I chose one of these I liked (KNOT FOR PROFIT). So far, so good! Now, I needed to find a second pair of theme answers. Why? As I mentioned early, themed puzzles typically contain between 3 and 5 theme answers (or, more generally speaking, between 33 and 55 individual squares devoted to letters within theme answers). This means that the theme is central to the puzzle and its “vibe,” but it doesn’t occupy the whole puzzle (which would be nigh impossible!). So, two theme answers isn’t enough.
Out of the rest of my options, OUT OF HOUR HANDS made me smile the most. This one is 14 letters, so I need another 14 letter entry to pair with it. HAVE A WRING TO IT also made me laugh, and it was 14 letters too, but I didn’t choose it. Why? Consistency. Themes are at their best (or so it is generally felt) when they are really consistent—solvers want to be able to “get” the theme and understand its rules and then use that understanding to get a leg up on the other theme entries, which means the theme has to have consistent rules to pick up on! Now, my theme obviously has rules—I take a phrase with a homophone in it and then I add exactly one letter always to the front of that homophone to change it to its pair to make a silly phrase. HAVE A WRING TO IT follows that rule, as stated, so how would it still make my theme inconsistent?
Consider where in my theme entries my gimmick plays out:
- BRITISH AISLES (Not at the beginning of the phrase)
- KNOT FOR PROFIT (At the beginning of the phrase)
- OUT OF HOUR HANDS (Not at the beginning of the phrase)
- HAVE A WRING TO IT (Not at the beginning of the phrase)
I have a rule—let’s call it the “one of these things is not like the other” rule. I don’t want a theme entry that feels like it is somehow different than all the others. With this set of entries, KNOT FOR PROFIT would feel like the odd one out because it’s the only one in which the homophone switch occurs at the beginning of the phrase. What’s the big deal, you might reasonably wonder? My concern here is for the solver’s experience: What if they crack KNOT FOR PROFIT first and assume all the rest of the theme answers follow the same gimmick (switching homophones at the beginning of phrases)? They’ll be confused until at least one or two more theme entries are solved, and they might feel like they had gotten tricked in the process!
So, to avoid this potential scenario, I wanted a 14 letter theme entry that would do the switchero at the beginning so that KNOT FOR PROFIT had a “partner.” That way, I’d have two theme entries each that accomplish the gimmick in different places within the phrase. Let’s call that consistent inconsistency :). Following this logic, I decided to go with WRITE OF PASSAGE as the pair for OUT OF HOUR HANDS.
Step 5: Now, I have to figure out how to arrange my theme entries in the puzzle!
[Edit: I plan to do a whole blog post on designing crossword grids–consider this an appetizer in the meantime!] 13 and 14 letter entries can’t go in rows 1 through 3 or 13 through 15. Why? Think about what would happen if I put a 13 letter entry in row 3. In crosswords, entries need to be bounded on either side by either the edge of the grid or by a black square. So, if I put a 13 letter entry all the way to the left in row 3, it’s going to need a black square in column 14 to mark its endpoint. That means the 3rd square in column 14 would now be a black square, which creates something taboo—a 2 letter word at the top of column 14 (the minimum answer length in most crosswords is 3).
So, because all these entries are 13 letters or longer, they all will need to go in rows 4 through 12. Plus, none of them can go in row 8, the center row, because symmetry makes the center row and center column unique. In other words, there’s no “paired” row for row 8. This leaves just 8 rows to cram four theme entries into. That’s tricky because having a lot of theme entries next to each other creates “locked-in letters” problems. Consider that if I put KNOT FOR PROFIT in my puzzle in the across direction, I will have to find answers that go in the down direction that cross every letter in that entry, including that K and P and those two Fs. Fair enough, but add in OUT OF HOUR HANDS just two rows away, and now I suddenly have to find several down answers that could cross both of those entries, with all those “locked-in” letters that I’m not allowed to change from the theme entries. In other words, I can’t make OUT OF OUR HANDS into OUF OF OUR HANDS just because I don’t like where that T happens to be. This all means I might need a word that has a P (from PROFIT) and an H (from HANDS) in it, separated by just one letter. Ugh. It’s not that there aren’t answers that fit the bill (SLEEPY HOLLOW, PSHAW, UPSHOT, and so on), it’s just that there aren’t as many as you’d think! When everything has to work out just right to get 70 some entries to neatly interlock with one another, having more options is definitely better than having fewer!
I mean, theme entries lock in letters that I must fill in other words through and around–there’s nothing inherently bad about that. It’s when theme entries are close together, are longer, and/or have kooky letters in them that you start to run into locked-in letter problems. So, the first and best thing to do is to get the theme entries as far away from one another in the grid as possible, to the extent you can. To do this here, I put the two 14 letter entries in rows 4 and 12 and then the 13s in rows 6 and 10. I put the 13s in the interior of the puzzle and not the 14s because they are shorter, so they will overlap less with each other, so there are fewer down entries that would have to potentially cross both of them. Plus, rows 6 and 10 are quite far apart. That means there is tons of room to place black squares into the grid to separate those two theme entries from each other. In fact, you’ll notice I put a total of 9 black squares in just those 3 intervening rows, which meant only two down entries (11-Down and 26-Down) needed to go through both the entry in row 6 and the one in row 10. Similarly, there are 4 black squares in between rows 4 and 6 (and 10 and 12, by extension), reducing how often the outer two sets of theme entries interact with one another as well. You’ll notice I usually separate kooky letters from each other also. All of this together meant I could avoid the problems caused by all those “locked-in” letters (Ok, except the P?H issue, which I had to solve with the lousy entry PAH at 51-Down).
If you’re keeping track, by this point, I’ve had to 1. Think up a gimmick for a theme; 2. Think up/research possible theme entries; 3. Vet these and whittle them down; 4. Find nice, consistent pairs that match in length; 5. Slot these into the grid; and 6. Put black squares around and in between them to make the puzzle fillable. Besides thinking about how I might clue the theme entries cleverly, I haven’t thought about clues for even a second, but I’m probably hours to days into the construction process!
Step 6: Nope, not cluing–filling!
What follows this point is filling the puzzle (putting entries in all the other spots in the puzzle not occupied by a theme entry…aka gibberish avoidance!). Filling is usually the single longest part of the process; it can take 10-20 hours by itself, although it can sometimes be fast (under an hour). Mostly, this depends on the fussiness of the constructor, the shape of the grid, and the amount of trial and error and backtracking is necessary. Filling could be the subject of several blog posts on its own!
Step 7: And now we clue.
Only once the filling is done do I sit down to write the clues. While I try my best to write fun, accurate, and unique clues (AREA has appeared so many times in crosswords that it has been clued EVERY possible way by now!), I usually don’t try too hard, limiting my cluing to around 2-4 hours of work. Why? Because puzzle editors (the ones who decide whether to accept or reject my puzzle) usually change between 50-90% of my clues anyway. Yep, that’s right–If you solve one of my puzzles, the grid and theme are (usually) all me, but the clues (a big part your experience as solver) might mostly be the editor’s! This fact is, ultimately, why I get such a laugh out of people guessing that I must start by writing the clues…not even close :).
Now, I don’t at all mean to imply I am bitter that my clues get changed! There are a lot of reasons an editor may change a clue. Maybe the one I wrote is too hard or even inaccurate. Maybe that same word was clued the exact same way in the puzzle that ran the day before mine in the paper. Maybe the clue I wrote was good, but it was too long (most newspapers have limited space for the puzzle). Maybe the clue was good, but it was in a challenging region of the puzzle, so an easier clue would help the solver out with that region. Maybe the clue spoke to something that is too regional; it’d make sense to a Mainer but not an Oregonian. Maybe the editor just thought up a clue that was more clever than mine was. Maybe my clue was just a bit vague/imprecise, leaving too many possibilities for the solver to weed through. The list of reasons goes on and on.
Anyway, there you have it, everything you might ever want to know about how a themed crossword puzzle comes to life. Did I cover everything you wanted to know? Did I miss something? Do you have questions? Let me know in the comments!