I'm Not the One Who Found the Gap

I want to write down, properly and once, the story of english.passed.jp, the site we run.

Let me put the numbers down first. Eight years since we started stacking it, and 1,079 questions today. We added them one at a time, and that's the number we ended up with.

What this article is about is the order of things: how it came to exist, where it stalled, and why it's moving again now. I'd rather it be read as a record of the order than as a success story.

Let me start with the most important part. I'm not the one who found the gap this problem set grew out of.

We didn't find it through market research. It wasn't my idea either.

The one who found it was an internship student. That person was actually teaching English to junior high school students. Out of that work, they brought back the point that something was missing.

I'm not going to summarize what that gap was in one line here, because I'm not the one who found it. When someone who arrived later summarizes what a person standing in the field actually saw, the summary usually turns into something slightly different. What I can say for certain is only this: it wasn't a guess made at a desk. It was a real gap, found in the field by someone who was actually teaching.

What I did was hear about it and say, "Then let's build it." That's all.

"Then let's build it" is something anyone can say afterward

This is the part I want to be honest about.

"Then let's build it" is something anyone can say — afterward. That is, once the gap has already been found. Say "let's build something" while no gap has been found, and what usually comes out is a thing nobody was struggling with. Building isn't a motive by itself. There has to be a real person who is 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 one who lined up those three wasn't me. It was the person doing the teaching. So telling the story of this problem set as "I had the idea and built it" gets the order wrong. I was the person standing next to whoever found the gap, taking the side that turns it into a thing.

I don't write this as modesty. It's a matter of stating facts. Get this backwards and you also lose track of what to build next — because the next gap probably comes from the field too. Whether you can properly listen to the people who have a field is, I think, what changes the next thing you build.

Stacked one question at a time

From there, the questions grew one at a time.

The current count is that sum. As far as I know, it wasn't bulk-generated — it was added by hand, one question at a time. The number isn't a banner; I take it as nothing more than where we currently stand.

One more thing, not about numbers. The substance was stacked back when AI couldn't work the way it does now. So the tidying work I'm doing today is work done standing on top of that. Stacking the foundation and straightening the shelves afterward are different jobs. On that asymmetry, there's an entry where Claude Code itself wrote that all it did was "tidying up".

And then it stalled

After that, it was stopped for a while.

I'm not going to compose a reason here. What was stopped was stopped. That's all. The questions themselves didn't disappear — they stayed right there. It just didn't move any further from that point. A good thing, sitting quietly on a shelf.

What was missing wasn't motivation.

It started moving again — but not because motivation showed up

This is the structure I most wanted to write down in this article.

The reason it restarted isn't "I got fired up again." It's that a plan I had wanted to carry out all along, but couldn't reach, became reachable for the first time with AI.

Here's what I had wanted to do. Take the stacked substance and re-shape it so that both people and machines can tell "here is what's inside." Build entrance pages. Make it readable in English as well as Japanese. Give it structure so machines don't mistake what the content is. Build departments and hubs so that someone who solves one question reaches for the one next to it.

The first pass of that was already done by the time of the entry I linked above. What I'm doing now is the continuation. There are still places that haven't been straightened out.

Individually, none of that is hard work. What's hard is the volume. It's every question's worth. Try it by hand alone and the end is invisible the moment you start. It's the kind of work where, even if you want to do it, your spirit breaks on day one. So it stayed wanted, and untouched.

Once AI became usable for real work, that wall of volume got lower. The moment it did, a plan that had been sitting on the shelf abruptly turned into an actual task list. That's when it started moving.

In terms of order: not "motivation showed up → it moved," but "it became reachable → it moved." That surprised me. What was needed to move a stalled thing wasn't willpower, but a change on the side of how much work it took. I've written the same thing about studying (the piece on restarting a stalled study routine with systems rather than willpower) — what didn't stop for lack of motivation, I think, doesn't come back by adding motivation.

SEO, AEO, and GEO

Let me get slightly concrete about the work happening now.

In the old vocabulary, this is SEO: getting search engines to correctly understand the content. But that's not all we're actually doing. It also covers the directions people call AEO (Answer Engine Optimization — optimizing for systems that return answers) and GEO (Generative Engine Optimization — optimizing for systems that generate them). Work aimed at getting the substance picked up correctly by systems that answer a question directly, and by systems that generate text in reply.

Honestly, these three aren't cleanly separated jobs. The core of the work is the same. Name the substance in each page's title and description. Straighten the heading hierarchy. Give it a structure where a machine can tell "this is a learning question." Give each question a fixed URL. All of it is unglamorous.

What changed is the recipient. It used to be safe to build on the assumption that 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 probably isn't enough. If the content isn't in a shape a machine can tell apart, I'd expect the chance of being found to drop accordingly. That's probably one of the reasons it felt like "good, but buried."

I've drawn one line for myself here. Don't change the substance. In this work I don't touch the question text or the answers that were stacked up. What I touch is only the container and the way it reaches people. Thin out the substance in order to improve delivery, and there's nothing left worth delivering. Adding new questions is a separate job, kept apart from this one.

And, to be honest about this too: the answer to whether this work is paying off isn't in yet. This isn't the stage where I can report "we finished tidying, and here's what happened." Right now, it's just that it finally became reachable.

Still only an idea

One more thing, which hasn't taken shape yet.

We're considering using well-known people and companies as material for questions. Human memory tends to catch on information that moved you emotionally — that's been the premise since we built the department. If a name you know, or a product you've seen, shows up inside a sentence, maybe that sentence tends to stay. Then you tie the grammar to the sentence that stayed. The thinking itself is an extension of the design behind our "Impact" department.

But it's still an idea. Before the question of whether we can ship it, there's an unresolved problem. This department has had a line drawn through it from the start: fun is no excuse to distort the facts. At the same time, the trick of this department's questions is deliberately building a sentence that couldn't happen. Use real people and real companies as material, and those two collide head-on. Is it acceptable to build a sentence that isn't true about a real person? If so, where is the boundary between acceptable and not? Until I have my own answer to that, we don't ship it.

So I'm not going to promise results here. I can't even say "we've started."

Don't get the order wrong

To close, let me lay out the order again — the thing this article was for.

The one who found the gap was an internship student teaching junior high school students in the field. The one who heard about it and said "then let's build it" was me. The result of stacking questions one at a time is today's 1,079. After that, it was stopped for a while. And then, not through motivation but because AI put the plan within reach, it started moving again.

Within that order, the part I handled is only the middle. Neither the first gap nor the substance itself was made by me. So I wrote this as a record of the order, not as a triumph. Keeping straight who did what is, I think, the cheapest cost there is for not getting the next thing you build wrong.

The questions themselves are lined up at english.passed.jp. The container around them is still being straightened.