How NYT Connections’ Algorithm Generates Puzzles

NYT Connections asks you to sort sixteen words into four groups of four, each group sharing a hidden theme. The categories are color-coded by difficulty — yellow easiest, green, blue, purple hardest — and the game has reached the kind of daily-ritual scale that previously only Wordle managed. The first surprise when you look at how NYT Connections works: there is no algorithm. The puzzles are written by humans.

The format

A Connections puzzle presents sixteen words in a 4×4 grid. You select four words you believe share a category, submit, and the game tells you whether your group is correct. If correct, the group is locked in with its color and theme revealed. If incorrect, you lose one of your four allowed mistakes.

The objective is to identify all four groups within those four mistakes. If you exceed the mistake limit, the game ends and reveals the remaining categories. Either way, the puzzle is once-per-day — same for every player on the same calendar date.

Why it’s not algorithmic

The thing Connections does that no algorithm could do well is design the misdirection. A good Connections puzzle has multiple words that plausibly fit multiple categories, and the construction of those overlaps is what makes the puzzle interesting.

Consider a hypothetical puzzle where the categories are “Apple products” (Phone, Watch, TV, Music), “Hot ___” (Dog, Cake, Spot, Air), and so on. The word “Watch” could plausibly fit both “Apple products” and “Things on your wrist” — and good Connections puzzles exploit exactly this kind of multi-category candidacy to make the player work.

Generating misdirection algorithmically is theoretically possible — you could enumerate word associations and find words with many cross-cluster ties — but the result wouldn’t be a fair puzzle. Fair misdirection requires editorial judgment about what kinds of overlaps are clever versus annoying. That’s a human-skill problem, not an algorithmic one.

The editorial process

Connections puzzles are written by a team of constructors at The New York Times, edited by Wyna Liu (the game’s puzzle editor as of the public reporting on the format). The construction process involves choosing four categories, generating candidate word lists for each, and then iterating until the overall puzzle has the right balance of obvious entries, misdirection, and an aha-moment payoff for the trickier categories.

The NYT has discussed parts of the process in interviews — Wyna Liu has described how the team aims for a specific difficulty curve where one or two categories should feel approachable, the middle two should require some thought, and the purple category should reward lateral thinking.

For the broader public-facing background on the game, the Wikipedia entry on Connections covers the development and launch context.

The difficulty colors

The four colors signal expected difficulty:

• Yellow (easiest) — usually a direct semantic category. Example: “Citrus fruits” with LEMON, LIME, ORANGE, GRAPEFRUIT.
• Green — slightly less obvious, often a category with one or two less-common members.
• Blue — typically requires more thought; categories may involve compound phrases or thematic groupings.
• Purple (hardest) — often wordplay-based. Categories like “Words that can precede STREET” or “Homophones of body parts” appear here. The purple category is where misdirection lives.

The colors are revealed only after you correctly identify a group. Until then, you don’t know which category any group corresponds to — only that the four words you’re guessing share something.

The “one away” feedback

The single piece of mid-puzzle guidance Connections offers is the “One away!” message that appears when your guess includes exactly three words from a real category and one outlier. This is genuinely useful information — it confirms your category exists and lets you reconsider the outlier — but it’s also the source of frustration when you can’t identify which word is wrong.

“One away” doesn’t tell you which word is the outlier or which category your three correct words belong to. You have to deduce the outlier yourself, often by considering whether the outlier might belong to a different (still-unresolved) category.

The mechanic is well-balanced: it provides enough information to keep good players engaged, but not enough to trivialize the puzzle. You can lose a Connections puzzle even after receiving multiple “one away” hints, which is part of what gives the game its long-term replayability.

Strategies that work

The dominant strategic advice from regular Connections solvers boils down to a few principles:

• Solve purple first if you spot it. The purple category is the trickiest, and identifying it early removes a major source of misdirection from the remaining words. If you spot the wordplay theme, lock it in immediately.
• Don’t rush yellow. The “obvious” yellow category often has decoy candidates — a word that fits the yellow category but actually belongs to blue or purple. The two-pass approach (find purple, then resolve middle categories, then yellow) avoids this trap.
• Use “one away” carefully. If you get “one away” twice with the same three core words, the outlier is one of the two remaining candidates. Bookkeeping helps.
• Step away. When stuck, looking at the puzzle for the third or fourth time often reveals connections that weren’t visible during the first attempt. Lateral thinking benefits from breaks.

Why an algorithm couldn’t replace it

Pretend you wanted to generate a Connections puzzle automatically. You’d need:

1. A graph of words and the categories they could plausibly belong to.
2. A scoring function that rewards puzzles with the right misdirection density and difficulty curve.
3. A constraint solver that picks four categories and sixteen words satisfying overlap constraints.

This is technically possible. But the resulting puzzle would feel mechanical — the overlaps would be statistically optimal rather than thematically clever, and the “aha” moments that make a good purple category satisfying would be replaced by clean but soulless wordplay.

The NYT editorial team’s value is in producing puzzles that feel hand-crafted because they are. The day-to-day variance — one puzzle leans on celebrity names, the next on chemistry terms, the next on song lyrics — is itself a feature, not a bug.

The shared-daily-puzzle structure

Like Wordle, Connections is the same puzzle for everyone on a given day. This is the foundation of the social layer — when someone shares their colored-square Connections result on social media, every reader played the same puzzle and can compare.

The share format uses the four difficulty colors arranged in rows showing the order you solved groups. Mistake rows show as four mixed-color squares. The format is intentional: it conveys the shape of your solve without revealing words.

Comparison with Wordle

Wordle and Connections share several design principles — daily synchronous play, shareable colored-square results, no-account-required access — but solve different cognitive challenges. Wordle is a deductive guessing game with a structured feedback loop. Connections is a categorization puzzle with looser feedback and a heavier reliance on lateral thinking.

The NYT Games suite includes both, plus Spelling Bee, the Mini Crossword, Tiles, and several others. Connections has reached audience numbers comparable to Wordle within its first two years — a meaningful achievement given how much harder it is to construct.

The accessibility design

Connections is unusually accessible for a puzzle game. There’s no time limit, no premium tier required (the puzzle is free on the NYT site), and the four-mistake budget means the game tolerates several wrong guesses before ending. Players can also access historical puzzles through NYT Games’ archive (subscribers only for older entries).

Mobile play works in the browser without an app, and the puzzle’s small footprint makes it suitable for short sessions — a typical puzzle takes between three and ten minutes.

For a different kind of puzzle

If you want a puzzle that’s pure reflex rather than pure deduction, the Chrome Dino game at the top of this site is the polar opposite of Connections — there’s nothing to categorize, no editorial intent to reverse-engineer, just a button and a cactus.

Frequently asked questions

Are NYT Connections puzzles generated by an algorithm?

No. Each daily Connections puzzle is hand-authored by a team of constructors at the New York Times and edited by the game’s puzzle editor. The four categories, the sixteen words, and the misdirection between them are all editorially curated.

How many mistakes can you make in Connections?

Four. Each incorrect group costs one mistake. If you make four wrong guesses, the puzzle ends and the remaining categories are revealed.

What do the colors mean in Connections?

Yellow is the easiest category, green is slightly harder, blue is harder still, and purple is typically the trickiest — often involving wordplay or compound phrases. The colors are only revealed after you correctly identify a group.

What does “one away” mean in Connections?

It means three of the four words you guessed belong to the same real category, and one word is the outlier. The game doesn’t tell you which word is wrong or which category your three correct words belong to — you have to deduce the rest.

Who creates the Connections puzzles?

A team of puzzle constructors at The New York Times. Wyna Liu has been publicly identified as the game’s puzzle editor. The editorial process involves selecting four categories, generating word candidates, and iterating until the puzzle has the right balance of clarity and misdirection.

The bottom line

NYT Connections works because the puzzles are made by people who think carefully about how a puzzle should feel. Algorithms could generate something that looked like Connections, but the result wouldn’t satisfy the “aha” moment that makes the real game worth coming back to every day. The lack of algorithmic magic is, paradoxically, the magic.

Related: How Sudoku Puzzles Are Generated (Algorithm Notes) · Wordle vs Connections: Comparing Two NYT Hits · NYT Connections Review: Best Daily Puzzle Hit?