Free · No loginRuns in your browser

AI Batch Question Generator

Set stage, subject, textbook edition, grade and difficulty to build an AI question-generation prompt, paste it into any AI chat, then paste the result back to validate and export a standard K12 question-bank JSON with answers. Free, no login, nothing uploaded.

← Back to question bank downloads

How it works — 5 steps

Set the question criteria: Choose stage, subject, grade, textbook edition, question type, difficulty, volume, unit and lesson, plus how many questions you need.
Copy the AI prompt: The tool assembles a prompt with the JSON schema baked in — copy it in one click.
Paste it into any AI chat: Open any listed official web chat (DeepSeek, Qwen, Doubao, Kimi, Yuanbao, GLM, ERNIE, Spark, MiniMax, MiMo), paste and send to generate questions in bulk.
Paste the result back and export JSON: Paste the AI's JSON back to validate every question item by item, drop invalid ones, then export a standard question-bank JSON file to your computer.
Cross-review with a second AI: Generate a review prompt for a second AI, paste its verdict back and adopt the fixes, then export the corrected JSON.

Supported scope: primary / junior / senior, 12 subjects, six objective question types, up to 1000 questions per batch. Exported JSON follows the AIVI question schema and can be imported by the Windows client.

① Configure question criteria (build prompt)

Stage

Subject

Grade

Textbook edition

Question type

Difficulty

Question count (max 1000)

Base points per question

Volume

Unit (-1 = any, 0 = none)

Lesson (0 = any)

Knowledge points (optional)

② Copy the prompt → paste into an external AI (Qwen / DeepSeek, etc.)

你是一名资深 K12 命题教师。请严格按以下要求出题，并以 JSON 数组格式输出结果。

一、出题范围
- 学段：小学（stage 固定为 "primary"）
- 学科：语文（subject 固定为 "chinese"）
- 年级：1 年级（grade 固定为 1）
- 题型：除填空题混合（在 single/multiple/judge/match/order 中大致均匀分布，禁止输出 fill 填空题）
- 难度：不限（low/mid/high 大致均衡）
- 册别：不限（按该年级教材常见分册自行分配，volume 必填且只能取 "upper"/"lower"/"whole"）
- 单元：不限（按教材实际内容分配，unit 必填且取值 0-99；0 = 教材开头未编号单元内容，如一年级准备课；1-99 = 实际单元编号）
- 课时：不限（按单元内实际课时编号 1-99 分配，lesson 必填）
- 知识点范围：不限（knowledge_point 必填，颗粒到具体考点）
- 每题基础积分：points 固定为 2
- 出题数量：100 道

二、JSON 字段规范（输出数组的每个元素为一条题目对象，字段与顺序如下）
{
  "id": "全局唯一不重复的随机字符串 id",   // 建议 12~20 位随机字母数字（可加 q_ 前缀），禁止使用含学段/学科/年级信息的顺序编号格式
  "stage": "primary" | "junior" | "senior",
  "subject": "chinese" | "math" | "english" | "science" | "physics" | "chemistry" | "biology" | "history" | "geography" | "politics" | "ict" | "habit",
  "grade": 1,
  "difficulty": "low" | "mid" | "high",
  "type": "single" | "multiple" | "judge" | "fill" | "match" | "order",
  "volume": "upper" | "lower" | "whole",
  "textbook": "教材版本" 或省略（省略 = 通用/不限版本题）,
  "unit": 0-99（0=教材未编号单元内容，如一年级准备课）,
  "lesson": 1,
  "knowledge_point": "知识点",
  "tags": ["检索标签"],
  "stem": "题干（表述清晰完整，不含答案提示；match 连线题必须含「左列依次是「项1、项2…」」句式）",
  "options": ["选项A内容", "选项B内容", "选项C内容", "选项D内容"],  // match 连线题：此为右列各项（顺序打乱）
  "answer": "见下方题型规则",
  "answer_accept": ["其他可接受答案"],
  "explanation": "解题思路简述",
  "points": 2
}
字段说明：textbook（教材版本）可省略（= 通用/不限版本）；存在时不得含空格、不超过 32 字符。options 仅 single/multiple/match/order 必填（judge/fill 省略该字段）；answer_accept 仅 fill 题可填，其余题型省略；explanation 可选。

三、各题型 answer 规则
- single（单选）：字符串选项字母，如 "A"
- multiple（多选）：字母数组，如 ["A","C"]
- judge（判断）：字符串 "true"（正确）或 "false"（错误），禁止输出布尔值
- fill（填空）：字符串标准答案，如 "0.75"；同义可接受答案写入 answer_accept 数组
- match（连线）：题干必须含「左列依次是「项1、项2、…」」句式列出左列各项（顿号或逗号分隔，项数 2~8）；options 为右列各项内容（顺序打乱，禁止与左列一一对应，数量与左列项数一致）；answer 为按左列行序逐行配对右列的字母数组，如 ["B","D","A","C"]（左列第 1 项配 options 第 2 项则首元素为 "B"）
- order（排序）：options 为待排序元素列表，answer 为正确顺序的字母数组，如 ["C","A","B"]

四、硬性要求
1. 只收录有唯一标准答案的客观题；禁止主观题、开放题、无标准答案的题。
2. 禁止出任何需要图形/图片才能解答的题（如几何图、统计图、函数图象、示意图、实物照片等）；仅当图形可以用纯文本字符完整表示时才允许（如 ○△□●★ 等字符符号、字符表格、"3 × 3" 点阵、纯文字描述位置关系），否则必须把题意改写为纯文字表述或换题。题干、选项、解析中禁止出现「如下图所示」「见图」等依赖图形的表述，也禁止输出图片链接、base64 或 LaTeX 绘图。
3. 题目内容科学准确、无歧义，难度与表述符合该年级学生认知水平。
4. 枚举字段必须严格取上述允许值，不得使用中文标签。
5. 只输出一个 JSON 数组；禁止输出 markdown 代码围栏、注释、解释、标题等任何其它文字。

③ Open a Chinese AI chat (top 10 — pick any one)Clicking a card opens that vendor's official web chat in a new tab and auto-copies the prompt above — paste and send to generate questions in bulk. Listed in no particular order.

1DeepSeek↗

DeepSeekOpen-source leader — strong reasoning, math and coding at very low cost

2Qwen↗

Alibaba CloudAlibaba flagship — balanced Chinese understanding and long-context, open weights

3Doubao↗

ByteDanceLargest consumer base in China — mature multimodal and everyday Q&A

4Kimi↗

Moonshot AIVery long context and long-horizon tasks — good for a whole textbook in one pass

5Tencent Yuanbao↗

Tencent HunyuanWeb-grounded answers, multimodal generation, fast/slow dual thinking

6GLM (ChatGLM)↗

Zhipu AIGLM series — strong coding and long-horizon tasks, open weights for self-hosting

7ERNIE↗

BaiduNatively omni-modal — strong Chinese writing and search grounding

8iFlytek Spark↗

iFlytekStrong Chinese and voice interaction — deep education-scenario expertise

9MiniMax↗

MiniMax (Hailuo)1M context + native multimodal + agent capabilities in one model

10Xiaomi MiMo↗

XiaomiStrong reasoning and math — free on the web

All links point to each vendor's official public web chat (no particular ranking). This site does not call any model API on your behalf and is not responsible for their content or availability.

④ Paste the AI result → parse & preview → export JSONPaste the AI's question JSON below, click Parse & Preview to validate every item, then Export JSON to save it locally. Nothing is stored on the server — keep the file or hand it to an admin.

Export validated questions only

⑤ Cross-review: let another AI check and fix the resultCopy the review prompt above (send it to another AI — or the same one), paste the returned verdict below, then Replace with fixes to overwrite step ④ and export again.

Review prompt (send to a second AI)

你是一名资深 K12 审题教研员。请对以下 AI 生成的题目逐题独立复核，并以 JSON 数组格式输出复核结论。

一、待检题目（JSON 数组）
请把上一步 AI 生成的题目 JSON 数组原样粘贴到下方占位处（不要改动内容）：

__QUESTIONS__

二、判定基准
- 学段：小学（stage = "primary"）
- 学科：语文（subject = "chinese"）
- 年级：1 年级
- 教材版本：不限
- 题库只收有标准答案的客观题；题型白名单 single（单选）/ multiple（多选）/ judge（判断）/ fill（填空）/ match（连线）/ order（排序）。

三、逐题检测维度（逐项核对，不得跳题）
1. 答案正确性：独立重算一遍，禁止沿用原题答案；答案错误直接判 fail。
2. 科学性：题干、选项、解析是否存在知识性错误、表述歧义或常识错误。
3. 图形依赖：题目是否依赖图形/图片才能解答（几何图、统计图、函数图象、示意图等）——仅当图形可用纯文本字符完整表示（字符符号、字符表格等）时才允许；含「如下图/见图」表述、图片链接或 base64 的直接判 fail，可纯文字改写的给 fixed。
4. 年级适配：难度与知识点是否符合该年级认知水平，是否超纲或过于简单。
5. 格式契约：枚举字段（stage/subject/difficulty/type/volume）取值合法；answer 形态符合题型规则
   （single 为选项字母字符串；multiple / match / order 为字母数组；judge 为 "true"/"false"；fill 为答案字符串）；
   match 题干必须含「左列依次是「…」」句式且 options 为打乱的右列项、数量与左列项数一致。
6. 选项质量：single 有且仅有一个正确项、multiple 至少两个正确项；选项互不重复、不含「以上都对」类兜底项。
7. 重复度：与其它题是否存在题干雷同、同知识点同难度重复。
8. 元数据合理性：knowledge_point、textbook（教材版本归属唯一且与内容相符）、unit、lesson 是否与内容相符。

四、输出格式（只输出一个 JSON 数组，每条对应一道被检题目，顺序与输入一致）
{
  "id": "被检题目的 id（必须原样回写，禁止改写或省略）",
  "verdict": "pass" | "warn" | "fail",
  "issues": ["问题说明（按严重度排序；pass 时可为空数组）"],
  "fixed": null 或 { 完整题目对象 }
}
- verdict：pass = 无问题；warn = 有小问题但修正后可用；fail = 答案错误或无可救药，建议弃用。
- fixed：pass 且无需改动时为 null；warn 必须给出修正后的**完整**题目对象（未改动字段原样保留，id 保持原值）；
  fail 若无法修复则填 null，并在 issues 首条写明「建议弃用」。
- fixed 的完整字段：stage / subject / grade / difficulty / type / volume / textbook / unit / lesson /
  knowledge_point / tags / stem / options / answer / answer_accept（仅 fill）/ explanation / points。

五、硬性要求
1. 逐题独立复核，输出条目数必须**等于**输入题目数，禁止合并、省略或新增条目。
2. 修正以「最小改动」为原则：能改答案/选项就改，不要整体重写题干。
3. 只输出一个 JSON 数组；禁止输出 markdown 代码围栏、注释、解释、标题等任何其它文字。

Nothing parsed yet: parse the AI result in step ④ first — copying will then embed the question JSON automatically. This is currently a paste-in template (__QUESTIONS__ placeholder).

Paste the review result (verdict from the second AI)

Frequently asked questions

Is this tool free? Do I need an account?

It is free and open to everyone — no login, no registration. Open the page and start generating questions right away.

Does it upload my data or call any AI model for me?

No. The whole workflow runs in your browser: the tool only builds a prompt and parses what you paste back. It never calls a model API on your behalf and never writes to the AIVI question bank.

Which question types are supported?

Six objective types: single choice, multiple choice, true/false, fill-in-the-blank, matching and ordering. Every question must carry a standard answer, because the bank only accepts objectively gradable items.

How do I get the questions onto my child's computer?

Export the JSON in step 4, then import it in the Windows client under Question Bank → Import. Administrators can also submit questions straight into the public bank from the console.

How many questions can I generate at once?

Up to 1000 per batch. Splitting by unit or lesson into several batches keeps each batch easier to review.

Why run a cross-review?

A second AI re-checks stems, options and answers, and flags wrong or inconsistent items. Adopting its fixes before export noticeably cuts the error rate.