Interactive | Why AI can’t beat humans at setting puzzles yet

In the 1940s, The New York Times (NYT) introduced its crossword — after dismissing it as a “primitive sort of mental development” two decades earlier — as a distraction in the bleak aftermath of the Pearl Harbour bombing, marking the beginning of puzzles as an entertainment phenomenon. Today, NYT’s major revenue generator is its games subsidiary — NYT Games. A Statista report published in February 2026, noted, “In the second quarter of 2024, the NYT Games app… generated about 11.2 million USD in in-app purchase revenues worldwide. The gaming app is partially subscription-based.”

The credit for the first crossword, though, goes to Arthur Wynne who created a diamond-shaped grid called ‘Word Cross’ for the New York World newspaper in 1913. A subsequent typographic error printed the puzzle as ‘Cross Word’ and it has been called that since.

Currently, the global word games and logic puzzle market is valued at $9.4 billion and is projected to more than double to $20.1 billion over the next decade, according to the ‘Word Games Market Outlook (2025–2034)’ by Dataintelo.

For his readers, Wynne intended the crossword would offer an interactive and stimulating experience. And indeed, our minds today are tested for their dexterity, as a combination of words are thrown into grids making us rack our brains, sometimes in frustration, and at other times in anticipation, in pursuit of that singular ‘aha’ moment.

The Good Puzzle

There are people who are intelligent and then, there are puzzle solvers. Their intelligence comes with a reserve of patience, allowing them to see beyond, in between and underneath the surface. A solver is curious and their never-give-up attitude nearly epitomises human resilience. But, without a good puzzle, there is no good solver.

Will Shortz, editor of The New York Times crossword, equates a good puzzle with a work of art: An attractive crossword grid full of interesting, colourful, lively vocabulary, where every square appears in two answers.

“The whole point of a puzzle is to entertain the solver. You want the solver to win, and you want to push the solver to reach that goal,” says Shortz, 73, who is arguably the only person in the world to have a college degree in Enigmatology, the academic study of puzzles, a major he designed himself at Indiana University.

Shortz’s all-time favourite clue appeared in The New York Times on November 5, 1996. It was Election Day to call the race between Democrat Bill Clinton and Republican Bob Dole. The clue for the 7-letter ACROSS answer read: ‘Lead Story in tomorrow’s paper!’ The DOWN clues from the answer were created such that it could be the name of either candidate (CLINTON or BOB DOLE) and the grid would still work.

Broadly, there are two types of crosswords: Quick, which is American with straightforward clues; and Cryptic, of British origin and followed in most former colonies of the British empire, including India, Australia and New Zealand. A cryptic clue is never what it seems. There are rules and devices — anagram, charades, container — using which you are required to break the clue in parts, indulge in wordplay before the answer reveals itself.

For both, though, a delicate balance is essential — not so hard that it makes the solver give up, but not too easy that the win doesn’t feel earned.

Australian puzzle setter David Astle has the best analogy — an apple tree.

“You don’t want a big bowl of apples sitting for you. You also don’t want a tree that is not going to give you any fruits. There are some low-hanging fruits that give you the courage to keep climbing up the tree. Ultimately, it needs to be this slow surrender of delightful apples,” says Astle, who has had an illustrious career as a TV anchor, radio host and author.

Simultaneously, the 64-year-old has been setting puzzles for publications such as The Sydney Morning Herald, The Age, The Canberra Times among others, although his “gateway drug” into this world was riddles.

Speaking of riddles, do you have the answer yet?

A good puzzle is as much a dopamine hit for the solver as it is for the setter, whose ultimate goal is to create a “satisfying play”, achieved with clever clues, fresh angles, snappy language, smart wordplay, all leading up to juicy answers. This much sought-after satisfaction depends not simply in solving the crossword but also in the fewer hints one requires to reach the goal.

At The Times (UK), for instance, the qualification for a setter is not so much the list on their resume. “We look for people who can write good, fair, concise clues but also have a very good, plausible surface meaning,” says Mick Hodgkin, Puzzles Editor at The Times.

The other non-negotiable quality of a good puzzle is its ability to evoke a triumphant chortle. “I would like to make the solvers laugh at some point,” adds 63-year-old Hodgkin, recalling a funny crossword clue he liked. It was ‘Fly back from China in 747?’ The answer? TANG, referencing the dynasty.

Ben Tausig of the AVCX (American Values Club), an independent puzzles and games outlet, agrees, “Above all else, is the puzzle funny? We want ideas that will put a smile on people’s faces.”

Clues Beyond Borders

Shortz, Astle and Hodgkin began, first solving and then, setting puzzles in a different time. Each of them discovered their passion in their local newspapers as young boys. At the time, it was unlikely for a person in the UK to read an American or Australian newspaper. Besides, the clues were often local references that people outside the community or geography would not necessarily understand. For instance, referring to the Brooklyn-Manhattan transit as BMT.

Today, we live in a different world. Out of the 13.35 million total subscribers of NYT Games as of mid-2026, over three million live outside the US, which means that while the lived knowledge of culture still remains largely elusive, the chances of a non-American accessing a crossword appearing in an American paper have massively increased. This realisation has pushed setters to expand — albeit gradually and often relying on stereotypes — their scope of inspiration to a more inclusive knowledge base, spanning age, culture and even gender.

Clues referencing Indian food items, cities and personalities frequently reappear in NYT and The Times (UK) crosswords. A Naan is: Indian flatbread; Sari: Traditional women’s draped garment; Assam: Northeastern Indian state known for tea and silk production. A memorable double-meaning clue for the answer (Mahatma) ‘Gandhi’ in The Times (UK) was: Indian spinner who played a leading role in getting England out.

Most of AVCX solvers are American, closely followed by British, based in Australia, New Zealand, Germany and other European countries. Asia, including India, also contributes significantly to their subscriber base, which is why their sports clues often vary between baseball and cricket. “There will be quite a lot of cricket references, which probably don’t bother our Indian solvers but the Americans might go, ‘What’s this?’,” Tausig says.

To Each their Own

Like most setters, Tausig, 45, started doing crosswords as a hobby, eventually crossing over to the other side. For him, setting crosswords became a medium of expression when he couldn’t see the voice of his own generation in the puzzles he was solving. “A lot of crosswords felt like they were for people who were old.” That led Tausig to found AVCX in 2006, which either presents traditional topics with a new vocabulary or refer to more recent pop-culture phenomenons like Taylor Swift, which “some of the older solvers may not know about.”

What differentiates AVCX from legacy newsroom crosswords — representing a uniform, largely objective, non-political stance of the organisation — is the unique voice of the setter. This means that conventionally the solution to the crosswords would draw heavily from general knowledge rather than opinion. With AVCX, it is the opposite. “Puzzles always have some kind of a point of view. Our politics are pretty liberal,” Tausig says. One of their visibly opinionated clues was published in 2016 after Donald Trump was elected president (first term): “Fascist-leaning reality TV star who won the electoral college in 2016.”

Patti Varol, editor, Los Angeles Times Crossword, likes to believe that their puzzles are “completely apolitical” but admits letting her own bias guide what is acceptable and what is not. While she hopes to remain “non-controversial”, she refuses to not represent the LGBTQ community in her puzzles. “My goal and my job is to make sure that anybody can pick up the paper and feel included. That means I’m going to acknowledge the existence of different pronouns, choices and colours and the whole messy spectrum of human life,” she says, adding, “If you’re triggered by the existence of trans people, I can’t help you.”

Gender Matters

When it comes to gender, the world of puzzles is not very different. It has traditionally been a male-dominated field, with few women editors, but when their voice does seep through the puzzles, it makes all the difference.

A woman setter is more sensitive. Varol gives an example: It was women who started clueing Ina (Garten) to reflect the renowned chef as a person — (Be My Guest With __ Garten), and not as the modifier phrase — ‘in a’ ( ___ manner of speaking, Garten), often used by male setters.

India’s Sowmya Ramkumar, who has set crosswords for several publications, including The Hindu, The Boston Globe and Gulf Weekly, and is the founder of 1ACross, an independent online platform of cryptic crosswords, likes to indulge in a bit of mischief every now and then.

One of her favourite pranks to play as a setter is to use female pronouns in clues for professions that have both male and female personnel — doctors, lawyers, economists. “Most solvers immediately make a male association with these professions, so a female pronoun throws them off, taking them longer to solve,” says the Bahrain-based setter.

In the Time of AI

It is, thus, the individuality of each setter that makes a puzzle one of a kind. As technology — which has helped create a global community of puzzlers — slowly takes over different industries, the inevitable question remains: Can a machine set and solve puzzles?

Yes, it can. Like humans? The unanimous verdict: No. At least, not yet.

The black-and-white grid became automated long ago, but technology, particularly AI, has a lot of ground to cover as a setter of cryptic crosswords. An improved version of Dr Fill, a computer programme, beat humans at solving quick crosswords in 2021 but, as Indian setter Tony Sebastien says, “AI may now construct a full NYT-style crossword. What it still cannot really do is set good cryptic clues.”

John Henderson, whose extraordinarily challenging puzzles have appeared in The Independent, The Times and The Guardian, recalls The Times Crossword Championship in London in 2024. “The AI application (Crossword Genius) solved five of the six puzzles to qualify into the finals. The one it could not crack was mine.”

Perhaps, because a good puzzle isn’t just about words. It requires an understanding of human behaviour — what they are likely to assume, what will crack them up and ultimately, what will win them over.

One of Sebastien’s most outrageous feats as a setter was a crossword in The Hindu that doubled up as a marriage proposal to his now wife, Rosanna Jacob. While the clues alluded to different aspects of a wedding, the first letters of each clue formed the question: “Rosanna Jacob, will you marry me?”

Riddle answer: Crossword