I Already Have ChatGPT. Why Would I Need a TOEIC App?

Great engine. Now build the car.

Let's be clear about where we stand: we are not anti-AI. Question generation, explanations, audio — AI has done a great deal of the heavy lifting in this app. So this essay won't tell you "AI can't do it." It can, and very well. In fact, the developer of Handy 990 has spent many years building AI applications and has written a book about AI — hardly the profile of someone who thinks AI doesn't work. The issue is that there is a long road between "a general-purpose AI that can do anything" and "a tutor who knows you're sitting the TOEIC next month."

Someone has to walk that road. Either the app developer does, or you do.

1. Getting AI to remember you is a job — and that job isn't learning English

Today you tell ChatGPT, "I'm weak on Part 5 prepositions; give me more of those." It obliges beautifully. Tomorrow you open a new conversation and, most likely, it has forgotten. Of course AI can remember: you can switch on memory, set up a project, write a "my exam profile" document and paste it in every time — target score, weak areas, where you left off, which questions you've already done.

It's doable. But notice what you're doing: you have become your own learning-management system. Maintaining that profile, reformatting your answer history into something the AI can read, reminding it not to repeat questions — every minute of that is subtracted from actually practicing English. An evening spent teaching an AI who you are and an evening spent doing a hundred questions: only one of them shows up on your score report. This is work an app should simply do for you, and do more precisely than any hand-maintained document: it remembers every question you've answered, knows which word is due for review today and which can wait a month, never quizzes you on the same word three times in a row, and never permanently forgets the one you actually needed.

2. AI-written questions contain errors — and you can't see them

This is the point almost nobody raises, and the most important one.

A certain share of AI-generated multiple-choice questions are simply broken: the key is wrong, two options are both correct, or the right answer happens to be the longest one. How do we deal with it? We keep a human in the loop: every question is cross-reviewed by people and AI checking each other (the stories of AI doing something dumb could fill an evening), then run through a layer of rule checks — option length, answer distribution, similarity to existing items. Even then, the occasional one slips through, and we rely on user reports to catch it.

Now picture a learner practicing in a chat window. The AI makes a mistake; the learner has no way of knowing. All they see is "the correct answer is C," and they dutifully learn the wrong thing. Learning something wrong is worse than learning nothing — and a chat window has no quality-control department.

3. AI writes "English questions," not "TOEIC questions"

TOEIC items have a very specific shape: Part 1 is built around photographs, Part 2 is listen-only with nothing on the page, Part 3 has three-speaker conversations with charts, Part 7 has double and triple passages to cross-reference — and the whole thing is a 200-question, 120-minute endurance event.

Ask an AI for a mock test and you'll get something that resembles "an English exam" — but not the TOEIC. It won't play you the listening section in American, British, Australian, and Canadian accents, at the real speed, once, with no pausing. It also has no idea whether the questions it wrote are easy or hard for real test-takers, because it has no answer data from thousands of people who've actually taken them. Difficulty is measured, not guessed. And half the value of a mock test is discovering, before exam day, what your brain does at minute ninety. A chat window can't give you that fatigue.

4. Weaknesses aren't found by asking. They're found by counting.

AI can give a dazzling commentary on the question in front of it. But "you got this one wrong" and "your error rate on inference questions is double your own average, and it hasn't improved in three weeks" are two entirely different things.

The second requires three ingredients: structured tags on every question (which grammar point, which listening skill), a record of several hundred of your answers, and the statistics that connect the two. A chat has none of them. AI is like an excellent doctor who never has your medical history — it can talk about today's symptoms, but it can't see a trend.

5. Exam prep is a willpower game, not a tooling game

Honestly, the biggest enemy in exam prep has never been "the questions weren't good enough." It's "I didn't practice again today."

Open the app: today's questions are already waiting. Open ChatGPT: first decide what to ask for, then paste a prompt, then steer it back on topic when it drifts. That friction sounds trivial, but on some exhausted evening in week three it will decide whether you continue. Streaks, daily goals, reminders at the right hour — these "boring" features are precisely what carries people to exam day. A chat window has no concept of progress.

6. AI isn't free — and AI good enough for this is definitely not

"AI is free" is only true while you're on the free tier. And the free tier is exactly the one most likely to write broken questions, least likely to remember you, and most likely to start phoning it in by question three. The AI that can genuinely teach you the TOEIC is the paid plan — and the longer the conversations, the more questions you want, the better the model you'll need.

Do the math: a month of an AI subscription may well cost more than a month of a dedicated TOEIC app. And then you take that more expensive tool and serve as its quality-control department, its learning-management system, and its listening-audio studio. That arithmetic is strange whichever way you run it.

So, you'd be walking the road we already walked

To tune a general-purpose AI into something that genuinely prepares you for the TOEIC, you'd roughly need to: generate a question bank, de-duplicate it, review it, calibrate difficulty, record multi-accent audio, build a weakness model, implement spaced repetition, and add whatever it takes to make you come back every day.

We did every one of those steps as "AI plus human review," over many months, and we have fallen into all the available holes. You could certainly redo it yourself in a chat window after work — just more roughly, and by the time you finish, the exam will probably be over.

In fairness: three things AI really does do better than an app

Without this section, this essay is just an advertisement.

1. Open-ended follow-up questions. "Why can't I use which here?" "What's the difference between these two words?" Five layers deep, AI beats any app.
2. Custom example sentences. You work in semiconductors, so it builds sentences in a semiconductor setting; an app can't.
3. Instant feedback on writing and speaking. The TOEIC Listening & Reading test doesn't assess them, but if you need English for real, AI is the best sparring partner there is.

One more thing worth saying plainly: we believe learning to use AI tools matters for everyone. It's just that turning an AI into a TOEIC tutor is far too specialized a project for most people — that's our job, and it shouldn't have to become yours.

So the right answer isn't either/or; it's a division of labor: the app practices, measures, and accumulates; AI explains. That's exactly how we designed it — finish a question, tap AI 詳解 on anything you didn't understand, and both jobs happen in the same place.

Closing

Of course you can prepare for the TOEIC with ChatGPT. Just as you can buy an engine and build your own car — it's only that most people want the one they can drive to the test center on Sunday morning.

We use the engine too. The car, we've already built for you.