Mild spoiler warning: this post uses real groups from our puzzle library as examples.
Sixteen words. Four hidden groups. Four mistakes allowed. That’s the whole idea behind Connected Words, and it takes about ten seconds to explain. Writing a grid that’s actually fun to solve takes much longer. Every puzzle in our library of 250 went through the same process, and here’s what it looks like from our side of the board.
Step 1: Start with the purple group
You’d think we’d start with the easy group and work up. We don’t. We start with the hardest one, the purple group, because it gives the puzzle its personality.
Purple groups are about the words themselves, not what they mean. Some of our favourite types:
- Hidden words: OFTEN, CANINE, WEIGHT and PHONE each hide a number (ten, nine, eight, one).
- Fill in the blank: SOUND, LOVE, SNAKE and MOSQUITO all go before “bite”.
- Sounds like: MUSSEL, KNOWS, HARE and AYE sound like body parts.
- Anagrams: TEARS, RATES, ASTER and TASER use exactly the same letters.
A good purple group feels impossible until it suddenly feels obvious. If solvers groan when they see the answer, we’ve done our job.
Step 2: Build three groups around it
Next come the other three groups, ordered by difficulty:
- Yellow (straightforward): a clear category most people will spot quickly, like fruits, planets or kitchen utensils.
- Green (moderate): still a plain category, but one that needs a moment’s thought.
- Blue (tricky): more specific knowledge, such as Bond villains, Cluedo rooms or Roman emperors.
Yellow isn’t there to be boring. It gives solvers a foothold, and once four words are cleared the remaining twelve start to make sense. A puzzle with no easy way in just isn’t fun.
Step 3: Plant the red herrings
This is where a list of four categories becomes a puzzle. A red herring is a word that looks like it belongs in one group but really belongs in another.
One of our grids has a group of snakes (COBRA, VIPER, MAMBA, PYTHON) next to a group of programming languages (RUBY, JAVA, SWIFT, RUST). PYTHON is both a snake and a programming language, so a solver who spots the languages first is very likely to grab it.
Another grid pairs famous painters (MONET, PICASSO, DALI, REMBRANDT) with the Ninja Turtles, who are all named after Renaissance artists. Solvers who think “painters” for LEONARDO walk straight into the trap.
The aim is to make your first instinct slightly wrong, without being unfair.
Step 4: The fairness check (the step most people never see)
Red herrings come with a risk. If a word truly fits two groups, the puzzle might have two correct answers, and since our game only accepts one, you’d be marked wrong for a perfectly good guess. That’s the worst thing a puzzle can do.
So every grid has to pass one rule: there must be exactly one way to split the sixteen words into four groups. A word can tempt you towards two groups, but only one arrangement can work for all sixteen words.
Take the snakes again. PYTHON could be a language, but the language group already has four other members. If PYTHON goes there, the snake group is left with three words. The puzzle forces PYTHON back where it belongs, and that’s what makes the trap fair.
We check this for every red herring in every grid, and it catches real mistakes. Two examples from writing our library:
- A grid had “ocean life” (CORAL, KELP, PLANKTON, REEF) and “shades of pink” with SALMON in it. SALMON is ocean life too, and CORAL is a shade of pink, so the two words could swap and both answers would work. We replaced SALMON with BLUSH.
- A grid had “yellow fruits” including LEMON alongside “things you peel” including BANANA. Both words fit both groups. That whole purple group went, and a cleaner one (BANANA as in “banana boat”) took its place.
Neither problem shows on a quick read. You only find them by asking, word by word, “where else could this go?”
Step 5: The final polish
Before a grid goes live, it gets a last round of checks:
- Exactly 16 unique words. No duplicates, even ones that differ only in capitalisation.
- Short, readable tiles. Long phrases get squeezed onto mobile screens, so we keep them short where we can.
- No obscure trivia in yellow or green. Specialist knowledge belongs in blue, where solvers expect a challenge.
- A shuffled preview. We look at the grid the way you will, in random order, to make sure it doesn’t give the game away at a glance.
Why the game feels the way it does
A few choices in the game itself come from the same thinking:
- “One away…” appears when three of your four words belong together, so a near-miss still teaches you something.
- Repeat guesses are free. Submit a combination you’ve already tried and we’ll tell you, without costing you a mistake.
- Random puzzles, no repeats. Every time you press New puzzle, you get a grid you haven’t played yet, until you’ve worked through the whole library.
Your turn
Now that you know how the traps are set, see whether it helps. Play a random Connected Words puzzle and look out for the word that fits a little too well. It’s probably purple.
Want more word games? Browse all the games on Wordletz.