How AI Writes a Fresh Trivia Round Before Every Game Night

The oldest problem in bar trivia is not the crowd, the sound system or the prize budget. It is the questions. Somebody has to write them, week after week, and that somebody eventually runs dry, recycles a round nobody notices is a repeat, or quits. Freedom Bar Games solves it by writing every round fresh, minutes before it runs. Here is what actually happens under the hood.

Nothing is pulled off a shelf

There is no static question bank shipped with the app that every bar in the country draws from. When a host starts a game, the server calls a large language model with the category, difficulty and round length, and asks for a brand-new set of questions written for that specific game. Two bars running a 'Nineties Music' round on the same Tuesday get two different rounds.

That matters more than it sounds. Shared question banks are why regulars at a chain trivia night can start recognising answers after a couple of months — and why the sharpest teams start Googling the provider instead of the question.

The generation step is the easy part

Asking a model for ten trivia questions is trivial. Getting ten questions that are actually usable in a loud room, with a single defensible answer, at a consistent difficulty, is the engineering problem. The pipeline the studio that builds the app put around that call does several things beyond the raw generation call:

• Constrains the answer format so options are short enough to read on a phone at a dark table.
• Rejects questions whose answer depends on a date after the model's knowledge cutoff.
• Screens out anything with two arguably-correct answers — the single fastest way to lose a room.
• De-duplicates against what that bar has already played, so regulars never see a repeat.
• Grades difficulty so a round ramps instead of opening with the hardest question.

Answers never touch the client

Generated questions are split the moment they exist. The prompt and options go to players' phones; the correct answers are written to a server-only collection that no client can read, and scoring happens on the server. It is the same reason the game clock is server-authoritative — anything a phone can see, a determined player can inspect.

If the answer key ever reaches the device, you do not have a trivia night. You have an honour system.

Why a question bank still exists

Generating everything live has one real cost: latency and API spend. So good questions get kept. Once a round has run cleanly, its questions are banked per-category and can be re-served to a bar that has never seen them, while generation runs ahead in the background to keep the bank topped up. Players get instant starts; the catalogue keeps growing; nobody gets a repeat.

The part that has nothing to do with AI

The model writes the questions. It does not decide that Tuesday is your slow night, that your regulars love music rounds, or that a 3x multiplier will pull thirty people through the door. That is still the owner's call, and the tools that make it work are ordinary product design. AI removed the bottleneck; it did not remove the job.

This is roughly the shape of every project Impartial AI Tech ships for local businesses: a narrow, genuinely annoying task handed to a model, wrapped in enough validation that the result is safe to put in front of paying customers. If you want the longer argument, see what actually works with AI for local businesses.