Market research · July 2026
市場研究 · 2026年7月
Between bubble machines and essay AI sits the weekly quiz — Taiwan's most-taken test. No teacher's tool serves it — not in the app stores, not in the teacher communities, not inside real classrooms.
畫卡機和作文 AI 之間，是每週都考的小考——台灣最常考的試卷。卻沒有任何工具為老師而做：App Store 沒有，教師社群沒有，真實教室裡也沒有。
01 — The gap
01 — 市場缺口
Grading tools cluster at the two solvable extremes — bubbles you can machine-read, and essays publishers monetize. The everyday quiz between them is where teachers actually live.
批改工具全擠在兩個做得起來的極端——機器讀得懂的畫卡，和出版社能收費的作文。夾在中間的日常小考，才是老師每週真正的生活。
Hanlin's answer-card scanner (閱卷大師), OMR card readers (讀卡機), ZipGrade. Fixed answer cards, single-choice only. A booked, shared machine for term exams — and a camera app frozen since 2020.
閱卷大師、讀卡機、ZipGrade。固定答案卡、只認單選。段考要預約共用機器——照相 App 從 2020 年就沒更新過。
Fill-in-the-blank, short answers, shown-work math — the shape of every elementary quiz. On paper or on a screen, no teacher-facing tool in Taiwan grades a handwritten answer. Publisher banks stop at multiple choice; HiTeach's new GPT grading reads typed text only; the global apps that touch handwriting have no Traditional Chinese and no gradebook.
填充、簡答、計算——每一張國小考卷的樣子。不論紙本或螢幕，台灣沒有任何「給老師用」的工具能批改手寫作答。出版社題庫停在選擇題；HiTeach 新的 GPT 批改只看打字；碰得到手寫的國外工具，沒有繁中、也沒有成績簿。
Hanlin's AI essay grader (翰林 AI寫作批改) and its peers. AI does read handwriting here — but only essays, sold inside textbook bundles.
翰林 AI寫作批改與同類工具。這裡的 AI 確實讀得懂手寫——但只改作文，而且綁在教科書套裝裡賣。
02 — Why it's empty
02 — 為什麼是空的
The gap isn't an oversight. It's protected on one side by a hard technical problem, and on the other by doors that stay shut on the incumbents who already solved it.
這個缺口不是沒人發現，而是被兩道牆保護著：一邊是真的很難的技術問題，另一邊是已經解出來的巨頭，進不了台灣的門。
There is no published benchmark for children's handwritten Traditional Chinese. On the newest OCR evaluations, handwritten text is still what frontier AI models fail at most, and even on clean, typed math, roughly 1 in 10 AI grades is wrong. Better prompts can't fix what the model can't read.
兒童繁體手寫，全世界沒有公開的辨識基準。最新評測裡，手寫內容仍是 AI 模型最弱的一環；連乾淨的打字數學題，AI 評分都還有約一成出錯。模型讀不懂的字，再好的提示詞也救不回來。
iFLYTEK scaled machine grading on standardized answer cards and professional scanners; Xiaoyuan AI (小猿AI) photo-grades 200 million problems a day. All of it Simplified Chinese, mainland curriculum — and no path into Taiwan. The demand is proven. The door is shut from their side.
科大訊飛靠標準答題卡＋專業掃描器規模化；小猿AI一天拍照改兩億題——全是簡體、大陸課綱，進不了台灣。需求被證明了，門卻是從他們那邊關上的。
A DotDotDot quiz is answered on-screen, so we never photograph handwriting and try to decode it. Multiple choice is a tap; fill-in-the-blank and show-your-working math are handwritten — and we grade the pen strokes as they're drawn. That's the whole trick: in every published benchmark, reading live strokes beats reading a photo of them, because the stroke data a photo throws away is exactly what makes grading accurate. Built first for practice at home, the same engine also runs on the tablets schools already have (the national one-tablet-per-student program, 生生有平板), where every quiz counts toward the tablet-usage numbers schools must report.
點點點的考卷在螢幕上作答，不必拍下手寫再猜它寫什麼。選擇題用點的，填充和計算用寫的——筆跡一邊寫，我們一邊改。關鍵在這：公開評測裡，直接讀筆劃都比讀照片準，因為照片丟掉的筆劃資料正是批改準確的關鍵。它先用於在家的個人化練習，同一套引擎也能在學校現有平板上跑（生生有平板國家計畫），每次小考都算進學校要交代的平板使用率。
And the paper that still happens — term exams, publisher tests, the quiz you photocopied? We read the one number that's always there: the teacher's red-pen total. (Prefer to print a DotDotDot quiz? That works too — photograph the stack.)
至於還是紙本的部分——段考、出版社考卷、影印的小考？我們讀那個永遠都在的數字：老師的紅筆總分。（想把點點點的考卷印出來考也行——整疊拍照就好。）
03 — The field
03 — 盤點戰場
Here's every grading tool a Taiwanese teacher could reach — and why none of them serves the weekly quiz.
這是台灣老師找得到的每一個批改工具——以及他們為什麼都碰不到每週的小考。
| Tool | 工具 | What it does | 它做什麼 | The catch | 但是 |
|---|---|---|---|---|---|
| Hanlin's answer-card scanner (閱卷大師) | 翰林閱卷大師 | Camera-scores answer cards | 拍照讀答案卡 | Hanlin's own answer cards only, single-choice only — frozen since Nov 2020, four ratings on the TW App Store | 只認翰林自家答案卡、只認單選——2020年11月起沒再更新，台灣 App Store 只有四則評分 |
| OMR card readers (讀卡機) | 讀卡機 OMR | Machine-scores term exams | 機器閱段考卷 | A shared machine you book by the hour; unheard of in elementary schools | 要登記借用的共用機器；國小幾乎沒人聽過 |
| ZipGrade · Plickers | ZipGrade · Plickers | Camera MC scoring | 拍照改選擇題 | Fixed bubble sheets / one projected question at a time; no Traditional Chinese | 固定畫卡／一次投影一題；沒有繁中 |
| HiTeach (Habook 網奕) | HiTeach 網奕 | Smart-classroom suite; GPT now auto-grades open-ended answers | 智慧教室套件；GPT 已能自動改問答題 | Typed-on-tablet only — no handwriting input; photo answers still graded by hand; enterprise procurement | 只吃平板打字——沒有手寫輸入；拍照作答仍靠老師手改；企業採購制 |
| Hanlin AI essay grading (翰林 AI寫作批改) & peers | 翰林 AI寫作批改 等 | AI essay grading from photos | 拍照 AI 改作文 | Essays only, bundled with textbook licenses | 只改作文，綁教科書授權販售 |
| Publisher question banks (TEAMS · OneClass · 出題高手) | 出版社題庫平台（TEAMS · OneClass · 出題高手） | Question banks + auto-scored tests | 題庫＋自動改分的線上測驗 | Locked to their textbook's content; MC/typed only — no handwriting; their AI push is quiz generation, not grading | 綁自家教科書內容；只認選擇／打字——沒有手寫；他們的 AI 主打「出題」，不是「批改」 |
| Kahoot | Kahoot | Live quiz games, auto-scored | 即時搶答遊戲，自動計分 | MC-family and typed exact-match only; engagement tool — and only ~4–6% of accounts ever paid | 只有選擇類與打字精確比對；是炒氣氛的工具——而且只有約 4–6% 的帳號付過錢 |
| Wayground (formerly Quizizz) | Wayground（前 Quizizz） | Quiz platform; AI now evaluates photos of written "show your work" | 測驗平台；AI 已能評「拍照上傳的手寫過程」 | The closest global tool — but English-first, no Traditional Chinese curriculum, session reports rather than a per-student gradebook | 全球最接近的工具——但英文優先、沒有繁中課綱，只有場次報告、沒有每個學生的成績簿 |
| 中小學題庫 exam-bank app (2025) | 中小學題庫（2025） | Taiwan exam-paper bank; photo-and-ask-AI correction of handwritten papers | 台灣考古題庫；拍照問 AI 批改手寫考卷 | Student/parent-facing — no roster, no class workflow, no gradebook; proof of demand, not a teacher's tool | 給學生家長用的——沒有班級名單、沒有班級流程、沒有成績簿；證明了需求，但不是老師的工具 |
| ClassDojo | ClassDojo | Points & parent comms | 點數與親師溝通 | No grades, no quizzes, no analytics | 沒有成績、沒有測驗、沒有分析 |
| School grade systems (欣河 / 全誼) | 校務系統（欣河／全誼） | Official grade records | 正式成績系統 | Term exams only — Excel paste exists, but nothing flows from daily quizzes; teachers still key or paste by hand | 只收段考——Excel 貼上是有，但日常小考的分數到不了那裡；老師還是手動 key 或貼 |
| FOLIO | FOLIO | Scan-and-email English essay reports | 掃描寄件的英文作文批改報告 | English essays only; days of turnaround; a service, not a tool — so small it's invisible to search | 只改英文手寫題；要等好幾天；是服務、不是工具——小到搜尋引擎都找不到 |
| DIY ChatGPT | DIY ChatGPT | Free; teachers already do it | 免費；老師已經在用了 | Drops rows, invents numbers, no batches, no history — and MOE's 2025 directive bars student data in personal AI accounts | 會漏列、會亂給數字、不能批次、沒有歷程——而且教育部 2025 年函釋已禁止把學生個資丟進個人 AI 帳號 |
Shipped ≠ adopted. The real incumbent isn't any of these thirteen. It's the red pen — and, since this year, a personal ChatGPT account. Demand so real that teachers improvise the tool themselves; nothing reliable enough that they keep it.
做出來 ≠ 有人用。真正的對手不是這十三個工具，是那支紅筆——還有今年起，老師自己的 ChatGPT 帳號。需求真實到老師自己動手土法煉鋼；但沒有一個東西可靠到讓他們留下來。
04 — Why us
04 — 為什麼是我們
Six advantages — each one either technically hard, structurally blocked for the incumbents, or compounding over time.
六個優勢——每一個不是技術上難，就是對手在結構上進不來，或者會隨時間複利。
Digital quizzes where students write — and the AI grades the ink live on the canvas, at home or in class. Publisher banks stop at multiple choice; the global apps that touch handwriting have no Traditional Chinese and no gradebook. Your own questions, too — not a bank locked to one textbook.
學生用「寫」的數位小考——AI 在畫布上即時批改筆跡，在家在校都行。出版社題庫停在選擇題；碰得到手寫的國外工具沒有繁中、沒有成績簿。而且考的是老師自己的題目，不是綁定某一版教科書的題庫。
Excel round-trip with the official school grade system (校務系統). The paste box already exists on the school side — we're the first thing that fills it automatically. Double-entry is how teacher tools die.
跟校務系統 Excel 雙向互通。學校那格「貼上」早就在了——我們是第一個自動把它填好的工具。要老師輸入兩次，就是教師工具的死法。
Every quiz adds to a multi-year, per-student history — the "who in the middle is moving" view teachers now compute by hand. Rivals can copy features. They can't copy your last three semesters.
每一次小考都累積成跨學年、每個學生的歷程——老師現在用手算的「中段誰在動」，一眼看見。對手抄得走功能，抄不走你過去三個學期。
Built with a practicing elementary-school teacher on the team, used in her classroom daily — with stamps, points, and lucky draws in the design language, not enterprise grey.
團隊裡就有現任國小老師，每天在真實教室裡用——設計語言是印章、集點、抽籤，不是企業灰。
We ship daily; incumbents ship by semester. The window below is closing — speed is how we hold it.
我們每天出貨，對手照學期出貨。下面那扇正在關的窗是真的——速度就是我們撐住它的方式。
Consent that's compliant with Taiwan's Personal Data Protection Act (個資法), built in. MOE's 2025 directive already bars student personal data in generative AI — we're the school-sanctioned answer to work pasted into personal ChatGPT accounts.
個資法合規、同意書內建。教育部 2025 年函釋已明定學生個資不得輸入生成式 AI——我們就是「貼進私人 ChatGPT」這個灰色地帶的體制內解答。
05 — The honest ledger
05 — 誠實帳
Teachers adopt free tools and rarely pay for them. ClassDojo's official policy is that teachers never pay — parents do, and subscriptions are most of its revenue. At Kahoot's last public reports, roughly 4–6% of active accounts had ever converted to paid. DIY ChatGPT anchors "AI grading" at zero. Usage and revenue are different products — and this page would be dishonest if it pretended otherwise.
老師會用免費工具，但很少為工具付錢。ClassDojo 官方明言老師永遠免費——付錢的是家長，訂閱佔了營收大宗；Kahoot 最後一次公開財報裡，付費轉換只有約 4–6%；免費的 ChatGPT 把「AI 批改」的行情錨定在零。使用量和營收是兩個不同的產品——這一頁如果假裝不是，就不誠實了。
Our answer, in order: publisher bundling (publishers pay for the stickiness their textbooks lack) → schools, via the tablet-usage KPI budget line (funded through 2029) → later, a cram-school diagnostic product built on this data. Teachers are the distribution. They are not the revenue. "Who pays" is an open question with a deadline — not an assumption.
我們的答案，按順序：書商合作（出版社為教科書缺少的黏著度付費）→ 學校（平板使用率 KPI 的預算線，這條線已編列到 2029）→ 之後，建立在這些數據上的補習班診斷產品。老師是通路，不是營收。「誰付錢」是一個有期限的開放問題，不是一個假設。
HiTeach already GPT-grades typed answers. Wayground already AI-reads photos of shown work. GoodNotes Classroom already grades tablet ink — in English. Any one of them localizing Traditional Chinese quiz grading shrinks the middle.
HiTeach 已能 GPT 批改打字作答；Wayground 已能 AI 讀「拍照上傳的過程」；GoodNotes Classroom 已能改平板筆跡——用英文。任何一家做了繁中小考批改，中間就縮小一圈。
Counter: daily shipping + the head start — localization, curriculum and a gradebook are a product, not a feature.
對策：每天出貨＋先行者優勢——繁中在地化、課綱、成績簿是一整個產品，不是一個功能。
Births fell 48% from 2016 to 2025 (208k → 108k); elementary enrollment drops below one million within four years.
出生數從 2016 到 2025 掉了 48%（20.8 萬 → 10.8 萬）；四年內國小學生數會跌破一百萬。
Counter: the engine travels — HK is proven ground — our engine already runs in 400+ schools there; Indonesia is next, with a teammate on the ground — once Taiwan proves retention.
對策：引擎帶得走——香港已被驗證——同款引擎已進 400+ 校；印尼是下一站，團隊有人在當地——前提是台灣先證明留存。
Teachers one-star flaky scanners, then never come back (Hanlin's answer-card scanner 閱卷大師 died this way — abandoned since 2020).
不穩的掃描工具，老師給一顆星就再也不回來（閱卷大師就是這樣死的——2020 年起沒人維護）。
Counter: an accuracy benchmark gates launch; low-confidence reads become one-tap confirms, never silent errors.
對策：準確率基準沒過就不上線；沒把握的判讀變成一鍵確認，永遠不會默默出錯。
Sources: the app stores, the trademark registry, MOE and Ministry of the Interior statistics, teacher communities (Flip Education 翻轉教育 · Dcard · PTT · Facebook groups), vendor product pages, and working sessions with practicing teachers. Every claim on this page traces to a linked primary source.
資料來源：App 商店、商標資料庫、教育部與內政部統計、老師社群（翻轉教育、Dcard、PTT、臉書社團）、廠商官網，以及與現任老師的工作會議。本頁每個主張都能對回原始出處連結。