english.passed.jp is a free English problem set for junior high school English, run by Amazing engine, with no sign-up required. It is built for students in the first through third years of junior high school, and for Japanese adults who want to relearn the English they studied in junior high school from the ground up. Over eight years of stacking questions one at a time, the count has passed 1,000 (1,079 at the time this article was first published).
The person who set it in motion was not the operator but an internship student. At the time, that student was with Amazing engine and, outside of work, was volunteering as an English tutor for junior high students. In that classroom, they noticed that there was no free, low-friction way for a junior high student to practice English. I, the operator, heard that and said, "Then let's build it."
From there, questions were added one at a time. Then the project stalled for a while, and now it is moving again. What restarted it was AI making SEO, AEO and GEO (optimizing for search engines, answer engines and generative AI) realistic to carry out. This article lays out, from the operator's side, who built it, why, why it stopped, and why it restarted.
What is english.passed.jp?
english.passed.jp is a free problem set covering junior high school English grammar and vocabulary, organized by grade and by topic. There is no account to create and no app to install. Open it in a browser and start solving.
| Item | Details |
|---|---|
| Run by | Amazing engine |
| Price | Free (no paid course, no app edition, no subscription) |
| Sign-up | None |
| For | Junior high school students (years 1–3) and Japanese adults relearning junior-high English |
| Questions | 1,000+ (1,079 when this article was first published) |
| Coverage | Junior-high grammar from the be-verb through relative adverbs, plus vocabulary and idioms |
| Formats | Grade-level drills (sets of 5, 10 or 15), topic pages by grammar point, and the special "Impossible English" section |
| URL | english.passed.jp |
In the grade-level drills, you get a score and an accuracy rate when you finish. For every question you miss, the explanation goes beyond the right answer into why you were likely to pick that particular wrong one. The topic pages are a faster format: tap an answer and it is color-coded right or wrong on the spot.
The content centers on junior-high grammar and vocabulary. It does not reproduce the format of the TOEIC or Eiken exams. That scope comes directly from the starting point described next: it was built for junior high students.
The cycling method (every question on the first pass, only the ones you missed from the second pass on) is covered in How to Cycle Through a Problem Set.
Why build a free junior-high English problem set?
Because an internship student who was teaching English to junior high students noticed that, at the time, there was nothing free that a junior high student could casually use to practice English.
Back then, Amazing engine had an internship student. Outside of work, they were volunteering as an English tutor for junior high students. In that classroom, they could not find a free, easy practice resource to point students to. That is what they told us.
This was not demand found through market research, and it was not the operator's idea. It was a gap found in the field by someone who was actually teaching. I heard it and said, "Then let's build it." That is all I did.
There is a reason for putting "who found it" first. Someone has to be actually stuck, the trouble has to be concrete, and it has to be untended. Only when those three line up does "then let's build it" mean anything.
The person who lined them up was the one in the classroom; the operator took the side that turns it into a thing. Get that order wrong, and you get the next thing you build wrong too.
The internship student's name and affiliation are not part of this article. What stays is the sequence of events.
How did 1,000+ questions get built over eight years?
One at a time, by hand. As far as I know, none of them were bulk-generated.
Eight years after the first question, the count has passed 1,000. At the time this article was first published, it was 1,079. The number is not a banner. It is simply where the running total happens to stand.
By topic, vocabulary and idioms are the largest groups, followed by grammar units such as infinitives and gerunds, comparison, and the past tense. The tilt toward junior-high grammar and vocabulary is a direct result of the starting point: a free resource for junior high students. The content was built for them from day one, and that has not changed.
One more note, about timing. These questions were stacked before AI could work the way it does now. So the tidying I am doing today is work done standing on top of that foundation. Laying the foundation and straightening the shelves afterward are different jobs.
For the AI-side view of the same problem set, there is the entry where Claude Code itself wrote that all it did was tidy up.
Why did it stop, and why did it restart?
After the questions were stacked, the project stalled for a while. Two things restarted it. First, AI made SEO, AEO and GEO work realistic to carry out. Second, a plan I had wanted from the start — using Twitter (now X) to bring companies and well-known people into the picture — came within reach.
I am not going to compose a reason for the pause here. The questions did not disappear; they stayed where they were. It simply was not moving. That is all.
The restart was a matter of volume, not motivation. What I wanted was to reshape the stacked questions so that both people and machines could tell what was there.
Concretely, that means giving every question a fixed URL, building entrance pages, and naming the content in headings and descriptions. None of that is hard on its own, but it applies to every single question. By hand alone, the end was out of sight the moment you started.
Once AI became usable for real work, that wall of volume dropped. The plan that had been sitting on a shelf turned into a task list. That is when things started moving. The order was not "motivation showed up, so it moved" but "it came within reach, so it moved."
I have written about the same structure in the context of study plateaus (restarting with systems rather than willpower).
What we are doing now for SEO, AEO and GEO
There is one policy: leave the content (question text and answers) untouched, and fix only the container and the way it reaches people.
A quick glossary:
- SEO (Search Engine Optimization): getting search engines such as Google to understand the content correctly
- AEO (Answer Engine Optimization): getting systems that answer questions directly (such as AI overviews in search results) to pick the content up as an answer
- GEO (Generative Engine Optimization): getting AI that generates text in reply, such as ChatGPT or Perplexity, to cite the content as a source
The three are not separate jobs; the core is the same. It comes down to four tasks, all of them unglamorous:
- Name the content in each page's title and description
- Straighten the heading hierarchy
- Give the pages a structure that lets a machine tell "this is a learning question"
- Give every question a fixed URL
Here is what is already in place. On english.passed.jp, every question has a fixed URL, and all of them are listed in the sitemap. On passed.jp, we publish an llms.txt, and robots.txt explicitly allows GPTBot, ClaudeBot, PerplexityBot and Google-Extended.
Articles are written in both Japanese and English and linked with hreflang, so that both readers who search in English and AI that answers in English can reach them.
What changed is the recipient. It used to be enough to assume a person would open search results one by one. Now there are more paths where a machine reads on the person's behalf, summarizes, and returns an answer.
On those paths, being good is not enough; the content has to be in a shape a machine can tell apart. That is likely one reason it felt "good, but buried."
The results are not in yet. This is not the stage where I can report "we tidied up, and here is what happened." It is the stage where the work finally became reachable.
Why do famous people and companies show up in the questions?
To tie grammar to a sentence that sticks in memory. At the same time, that sentence becomes an opening to reach out on Twitter (now X) to the person or company it features.
The "Impossible English" section of english.passed.jp has questions with headlines like "Gaga wore a dress made of meat!?" Human memory catches on information that stirs emotion. If a name you know or an event you have seen shows up inside a sentence, that sentence tends to stay, and grammar can be tied to it.
The thinking behind the design is in the article on building the "Impact" department.
From the start, this section has had one line drawn through it: being fun is no excuse for distorting facts. So the material is limited to things that actually happened, with sources attached. That line does not move when the subject is a real person or a real company.
Rather than producing separate promotional material, the learning content itself becomes the reason to reach out. That is what the plan I wanted from the start looks like.
But it has only just begun. There are still few such questions, and whether any company or public figure will respond is unknown. No results are promised here.
In short: who did what
Here is the order, one more time.
- Noticed the gap: an internship student teaching English to junior high students in the field
- Said "Then let's build it": the operator (me)
- Stacked the questions one at a time: 1,000+ over eight years (1,079 when this article was first published)
- Stalled: for a while
- Restarted: because AI put SEO, AEO and GEO work, and the Twitter (now X) plan, within reach
- Doing now: leaving the content alone and fixing only the container and delivery
Within that order, the operator's part is only the middle. Neither the first observation nor the substance of the questions is something I made alone. Keeping straight who did what is, I think, the cheapest insurance against building the wrong next thing.
The questions are lined up at english.passed.jp. The container is still being straightened.
FAQ
Who made english.passed.jp?
It is run by Amazing engine. The trigger was an internship student, teaching English to junior high students, who noticed that there was no free, easy way to practice. The questions were then built by hand, one at a time, over eight years.
Is it free? Do I need an account?
It is free, and there is no account. There is no paid course, no app edition and no subscription. Open it in a browser and start solving.
Can I use it to prepare for TOEIC or Eiken?
As an entry point or for rebuilding the basics, yes. The content centers on junior-high grammar and vocabulary, so for the actual exam formats (long passages, listening, time management) you will need the official practice books or past papers for each test.
Were the questions written by AI?
No. The questions were stacked by hand, before AI could work the way it does now. AI has been involved only in the container: fixed URLs, headings and descriptions. The question text and answers have not been changed.
Why do famous people and companies appear in the questions?
To tie grammar to a sentence that sticks. It also creates an opening to reach out on Twitter (now X) to the person or company featured. The material is limited to things that actually happened, with sources attached. This is a new effort, and there are no results to report yet.
