Start with a broad question
Good prompts are easy to understand in a second and have more than one good answer. “Name something you would take on a picnic” works because everyone has a picture of it, nobody agrees on the list, and the obvious answers (sandwiches, a blanket) and the rare ones (a citronella candle) are both clearly real. Very narrow prompts, such as ones that ask for exactly five letters or a single obscure breed, are fun for specialists but frustrating for everyone else, so we keep those out of the daily mix.
Build the list of valid answers
Every prompt needs a list of answers the game will accept. Some lists come from open data. For categories such as countries, rivers or chemical elements we pull candidates from Wikidata, which is released into the public domain. Other prompts, especially everyday ones like “something that makes a loud noise”, have no database behind them, so we write the list ourselves, with AI assistance, and keep adding to it.
Open data is never clean. The same query that returns the right rivers also returns a games festival, and the list of famous painters can include people who painted a few pictures as a hobby. So every list is reviewed and corrected, with AI assistance, and revised whenever players report a mistake. Wrong entries are removed, aliases and spellings are added (a player may type “Van Gogh” or “Vincent van Gogh”), and anything inappropriate is blocked by a shared word list.
Rate how obvious each answer is
Each answer gets a number for how likely a person is to say it first. We set this in advance, from our own estimates and from how common the word is in everyday writing, rather than from other players. The most obvious answers get the lowest scores; a small number of well-known but second-choice answers sit in the middle; the long tail of genuinely rare answers scores highest. Each prompt also has one chosen summit: a real answer almost nobody would say, worth exactly 100. See rarity without a crowd for the full method.
Test before it goes out
Before a prompt is allowed into the schedule it has to pass automatic checks. Every common answer a player would plausibly type must be accepted, along with plurals and a handful of typical typos. At least sixty answers must exist, every score band from Tarmac to Vacuum must be populated, and only the summit can score 100. The checks run again on every change, so a mistake in one list cannot quietly break another.
Schedule the week
Each day uses seven prompts from different families, with at least two classic lists and a good share of everyday and pop-culture prompts. No prompt repeats within 14 days, and a day never has more than two prompts from the same domain, so you do not get five food questions in a row. About one day in six is a theme, such as Food Day or Pop Night. The new daily appears at midnight US Eastern time for everyone at once.
If an answer you know is real is rejected, please tell us. Missing answers are the thing we most want to fix.