Write your questions with your AI.

Five steps

1. Download the templateA spreadsheet with the right columns and one example row to delete. Download goodcall-question-template.csv Want something to play tonight while your AI writes? Download the starter CSV: 48 ready questions in four categories, 12 each.
2. Copy the promptUse the Copy button below. Paste it into ChatGPT, Claude, Gemini, whichever you use, and attach the template.
3. Tell it what you wantThe last line of the prompt: topic, how many, the category name, who's playing.
4. Upload the CSV it gives youOn your host page, open Questions and press Import. Each category becomes a game once it holds 12 questions; until then it shows "Needs 12". Bad rows are listed by line; the rest go in.
5. Play itOn the host page, under Tonight's game, open the shelf named after your room, pick the category and press Start a game. Your questions stay in their own game: they only turn up in Anything Goes and the other broad games if you switch on Mix in for that category.

You need a room to import. Host a Game Night; it's free.

The prompt

Copy everything in the box, paste it to your AI with the template, and fill in the last line.

You are writing multiple-choice questions for Good Call Trivia, a live game where players answer on their own screens against a clock. I will tell you the topic, how many, and who is playing. Give me back one CSV that imports cleanly. Follow every rule: the importer rejects rows that break the hard limits. 

1. THE FILE. One CSV, UTF-8, header row first, exactly these columns in this order:
category,difficulty,question,answer_a,answer_b,answer_c,answer_d,correct_answer,fun_fact_blurb,clue_1,clue_2
The first eight are required; fun_fact_blurb (the fun fact) is strongly wanted; clue_1 and clue_2 are optional. Quote any field containing a comma, a quote or a line break; double any quote inside a quoted field. Use a CSV library. At most 2,000 rows. Write at least 12 questions per category (30 to 60 is better). A category with fewer than 12 won't play. Save as .csv, not .xlsx. 

2. HARD LIMITS (count characters, don't estimate): category 1-20 characters, spelled identically on every row · difficulty a whole number 1-5 · question <= 150 · each answer <= 40, all four different · correct_answer exactly one of A, B, C, D · fun_fact_blurb <= 180 · clue_1 and clue_2 <= 51 each. 

3. WHERE THE CORRECT ANSWER GOES — the rule that gets broken. The game shows answers in the order you write them; nothing shuffles them. So: (a) write every question with its correct answer and three wrong answers, without letters; (b) when all are written, make a list of N/4 As, Bs, Cs and Ds (+-1), shuffle it, and deal one letter to each question — right answer in that slot, wrong ones in the others in random order; (c) re-deal if any letter appears more than twice in a row; (d) count the correct_answer column: each letter between 20% and 30% of rows; report the counts. The correct answer must not usually be the longest option (a third of rows at most). If you can run code, do (b)-(d) in code. 

4. THE FUN FACT (fun_fact_blurb). After the answer is revealed, every screen shows it for about seven seconds. Write one for EVERY question. Not a restatement — say WHY that answer is right: the fact, the rule, the reason, in a way someone could remember. Trivia: Q: Which planet has the most known moons? -> "Saturn overtook Jupiter in 2023 when dozens of tiny moons were confirmed, most only a mile across." Study: Q: Which listing lets the seller sell the home themselves with no commission? -> "Exclusive agency: one broker is hired but the owner keeps the right to find their own buyer commission-free. Exclusive right-to-sell pays the broker regardless." Max 180 characters; name the correct answer; one idea; never hint at another answer. 

5. DIFFICULTY. 1 = nearly everyone knows it; 2 = most do; 3 = about half; 4 = one or two people in a group; 5 = a real stumper that is still knowable. Mostly 2s and 3s, a few 1s to get people on the board, a few 4s, 5 sparingly. For a study set, difficulty tracks how often the topic is tested and how easily it's confused. 

6. GOOD QUESTIONS. One defensible answer. Direct questions, not riddles, readable in one breath on a TV. Wrong answers are tempting and the same kind of thing as the right one (four cities, four years, four legal terms); for study sets they are the classic confusions. No joke options, no "all/none of the above", nothing that goes stale, no duplicates (including the same fact asked two ways). Unsure? Leave it out. 

7. CLUES (optional; blank beats bad). Clued In style only. clue_1 rules out one wrong answer by wordplay without naming it; clue_2 is a pure wordplay giveaway on the correct answer ("Springfield" -> "A spring in his step."). Neither contains the answer text. One sentence each, ending in a period, <= 51 characters. 

8. BEFORE YOU HAND IT OVER, verify and report: header exact; every limit in (2); four different answers per row; letter counts each 20-30%, no run of three; correct answer longest on <= a third of rows; a fun fact on every row; no duplicates. Then give me the CSV, the verification summary, and one line on anything you weren't sure of.

What I want: [TOPIC or STUDY GOAL] · [NUMBER] questions · category: [NAME] · players: [WHO]

What makes a good question

• One defensible answer. Direct questions, not riddles, readable in one breath on a TV.
• Four different answers of the same kind, one clearly right: four cities, four years, four legal terms. The wrong ones should be tempting. No joke options, no "all of the above", nothing that goes stale, no duplicates.
• The fun fact. The line after every answer saying why it's right, up to 180 characters. It's on screen for about seven seconds, and you can hold on it when people want longer. For a study set this is the whole point: the prompt tells the AI to explain the rule and name the classic confusion, so every wrong answer teaches.
• Unsure? Leave it out. A short file you trust beats a long one you don't.

Categories and difficulty

• Categories are up to 20 characters, spelled the same on every row. One category is one game on your picker. Ask for 30 to 60 questions per category: a game of your own category is 12 questions, and it needs 12 before it plays, so a bigger category lasts several nights without repeats.
• Difficulty 1 to 5, honestly rated: 1 is nearly everyone in the room, 3 is about half, 5 is a stumper that is still knowable. Mostly 2s and 3s.

The optional extras

Clues. Two optional columns for the Clued In style. The first clue rules out one wrong answer by wordplay without naming it; the second is a pure wordplay giveaway on the right answer. Each is one sentence, up to 51 characters. Blank beats bad: if you don't want clues, leave both empty.

Upload rules

• One CSV, UTF-8, header row first, at most 2,000 rows. Save as .csv, not .xlsx.
• Columns, in order: category, difficulty, question, answer_a, answer_b, answer_c, answer_d, correct_answer, fun_fact_blurb, clue_1, clue_2. The first eight are required; the fun fact is strongly wanted.
• Limits: question up to 150 characters, each answer up to 40 (all four different), correct_answer exactly one of A, B, C, D, fun fact up to 180.
• The right answer has to move around. The game shows answers in the order they're written, and nothing shuffles them. The prompt makes the AI deal the correct letter across A, B, C and D evenly (each between 20% and 30% of rows, no run of three) and report its count. This is the mistake every first file makes, so check the counts it reports.
• No duplicates, including the same fact asked two ways.

If the import complains

It lists the bad rows by line number and keeps the good ones. The usual causes: an answer over 40 characters, a fun fact over 180, a correct_answer that isn't A to D, two identical answers, or a file saved as .xlsx. Fix those rows and import just them again.

Studying with it

Say what you're studying for in the last line of the prompt, for example: Texas real estate licensing exam, 60 questions, category RE: Contracts. Use one category per topic area. Play a category on Standard until you're clearing it, then Pro. Scores are kept per category, so weak topics show.

An AI can be wrong, especially on exam material. Read the file before you rely on it, and check anything you're going to be tested on against your course book.