Turn ChatGPT's questions into a Jeopardy game

Write a Jeopardy game in exactly this format, as plain text with no table, no bold and nothing before or after it. If you've already written questions in this chat, reuse those.

Category: <category name>
100 | <clue> | <answer>
200 | <clue> | <answer>
300 | <clue> | <answer>
400 | <clue> | <answer>
500 | <clue> | <answer>

Repeat that for 6 categories, with the clues getting harder as the values go up and every answer different. Then finish with:

Final Jeopardy: <category name>
<clue> | <answer>

Topic: your topic

Asking for clues that get harder as the values go up is what makes the board play like Jeopardy. Left to itself, ChatGPT tends to write five clues of about the same difficulty, and then the values mean nothing.

There's no need to ask again in a new format. Paste what it gave you. These layouts are all read correctly:

• A heading or a "Category:" line for each category, then a line per clue, with or without the dollar value.
• The answer on the same line, after "Answer:", "A:", an arrow or in brackets, or on the line underneath.
• "Q:" and "A:" pairs, or the show's own naming, where the clue is labelled Answer and the response Question.
• A table with a row per clue, or the board drawn as a table with a column per category.
• A Final Jeopardy section at the end, with its own category.

The one layout that doesn't work is an answer key at the very end, with all thirty answers listed apart from their clues. Ask ChatGPT to "put each answer next to its clue" and paste again.

A board ChatGPT wrote has the same failure modes as one a person wrote in a hurry, plus one of its own. Read it once with these in mind:

• Wrong facts stated with confidence. Dates, record holders, "the first" and "the largest" are where it slips most. If you couldn't defend the answer to the room, check it or cut it.
• Two right answers. "This planet has rings" has four. A clue needs a second fact that rules out the rest.
• The answer is in the clue. ChatGPT often names the thing it's asking about, especially in the easy rows.
• The same answer twice. Across thirty clues on one topic, it repeats itself more than you'd expect.
• Pitched at the wrong room. A board for a class of ten-year-olds and one for a pub quiz differ in every clue, and the prompt is the place to say which you want.

It's easier to fix the board in the chat before pasting than tile by tile afterwards. A few follow-ups that work well:

• "Make the 400 and 500 clues harder. Most of the room should miss them."
• "Replace the third category with one about the 1990s and keep the format."
• "Rewrite each clue as a statement, so the answer is a short name or phrase."
• "Check every answer and tell me which ones you're unsure of." It won't catch everything, but it flags some.

On Pro, the editor can also write a board from a list of categories, and rewrite a single clue in place, which saves the round trip through the chat.

ChatGPT will host a game of Jeopardy in the chat window, and for one person on their own that works. For a room it falls apart: everyone is reading the same screen, there's no way to tell who answered first, and the score is whatever the person typing remembers.

The board is what turns thirty questions into a game. Everyone can see what's left, picking a tile is a choice with a value attached, and the buzz order is settled by the server rather than by whoever shouts. On this site, a 6 × 5 board with phone buzzers, live scores and Final Jeopardy is free, and pasting the questions in is part of that.