What kinds of questions can it generate?

It produces either multiple-choice questions with four options and a marked correct answer, or short-answer questions with a model answer. Pick the type that matches how you plan to use the quiz.

Are the answers reliable?

The model grounds questions in the text you paste, so answers reflect that content. Always skim them before publishing, since LLMs can occasionally introduce a subtly wrong option or answer.

Is my API key stored anywhere?

No. The key lives only in your browser tab and is sent directly to OpenAI for the single generation request. Nothing is logged or sent to Gera.

Can I use this to build evaluation datasets?

Yes. Short-answer mode is handy for creating question-answer pairs you can use to test other models or RAG pipelines against the same source material.

How long can the input be?

It works best with a few paragraphs to a few pages. Very long inputs cost more tokens and may exceed the model context window, so chunk large documents and quiz each section.

Quiz Generator from LLM Output (BYO-key)

Quiz generator from LLM output

Whether you are building e-learning content, onboarding material, or an evaluation dataset, turning a passage into good comprehension questions by hand is slow. This tool sends your LLM-generated content to your own OpenAI key and returns ready-to-use multiple-choice or short-answer questions, each grounded in the text you provide.

How it works

On run, the tool makes a single request to https://api.openai.com/v1/chat/completions. It instructs the model to read your content and produce exactly the number and type of questions you asked for, returning them in a clean, copyable format. For multiple-choice it includes four options and marks the correct one; for short-answer it includes a model answer. Everything happens directly between your browser and OpenAI — there is no Gera backend in the loop, and your key is never stored.

Tips and notes

Quiz one section at a time. Tight, focused passages produce sharper questions than a long mixed document.
Review before publishing. Skim the generated answers; occasionally a distractor option is too close to correct, and you may want to tweak it.
Use short-answer for eval sets. The question-answer pairs make a quick starting dataset for testing other models or RAG retrieval against the same source.
Increase count for spaced practice. More questions per passage gives you a bank you can rotate through across multiple review sessions.