Website / International
A century old film studio, rebuilt for the producers who read it first
Toei Kyoto Studios’ English site was a 2010s brochure with ninety words of English on it. We rebuilt it as toeistudios.co: one page, both lots, all eighteen sound stages drawn to scale, and readable by the AI assistants producers now ask before they ask anyone else.

- 12.8 s → 0.5 sPage load on a phonemeasured
- 66 → 16Requests on first viewmeasured
- 90 → ~1,000English words on the homepagemeasured
- 0 → 2 + llms.txtStructured data blocksJSON-LD
Where it started
The old English site had stopped somewhere in the early 2010s. The dated look was the smallest problem. The real one was that it told an international reader, a search engine and an AI assistant almost nothing.
- The English page declared itself Japanese (lang="ja"). The only h1 on it was 東映京都撮影所, in Japanese.
- No viewport tag. A phone rendered the 980 px desktop layout and you pinched.
- The navigation was images. About us, Stages and Art Services were JPG and GIF buttons. Text inside an image cannot be translated, searched or read aloud.
- Ninety words of English on the homepage. The news column was Japanese only, even on the English page.
- jQuery 1.12 from 2016 plus a slideshow plugin. 8.4 seconds to load on desktop, 12.8 on a phone, 66 requests.
- A Universal Analytics tag (UA-9558041-7). Google switched Universal Analytics off in July 2023, so nothing had been measured for over three years.
- No structured data, no hreflang, no Open Graph, no sitemap reference. To a machine the page was close to blank.


Who the site is for
We narrowed the reader to one person. A line producer in Los Angeles or London deciding whether to bring the next shoot to Japan.
That person’s questions are known. How big are the stages. What can the LED volume do. Who has shot here. Is there an incentive. Who do I email. Answering those five, in that order, is the structure of the page.
Both lots, Kyoto since 1926 and Tokyo since 1935, went into one site, because a production picks Japan first and the lot second. English only, on purpose. The Japanese audience already has its own sites.
What we built
One page. A 19 second, 4 MB hero film cut from eight shots of Toei’s own footage, then the two studios, the eighteen stages, the craft on the lot, the filmography, a century strip, the practical side of filming in Japan, and a contact form.
The centre of it is the stages section. Every published length, width and catwalk height became a three.js wireframe with a 1.7 m figure standing beside it. Metres and feet side by side, and where Kyoto does not publish a height it says so and draws 9 m as an assumption. A producer looks at dimensions before photographs.
The craft section is nine departments as short loops, from sword forging to facial capture, to make one point: it is all on the lot. The filmography puts a poster under your cursor as you move down the list.




Readable by people and by machines
The next producer’s first question does not go into a search box. It goes to an assistant: which studios in Japan have a permanent LED volume. This page answers that in text an assistant can quote, with a URL it can cite. Four decisions made that true, all taken at design time.
- All eighteen stage rows are static HTML in the page, not drawn by JavaScript, so a crawler or an agent with scripts off gets the whole table.
- JSON-LD declaring the Organization, with Kyoto and Tokyo as departments, and the WebSite.
- A /llms.txt with every stage dimension, the credits, the incentives and the contact in plain text, linked from the head with rel="alternate".
- A robots.txt that welcomes crawlers, a sitemap.xml, a canonical, Open Graph. The boring things, all of them.
Light on purpose
There is about 12 MB of media on the page and a first visit loads 1.8 MB of it on a phone, 4.6 MB on desktop including the hero film. Video and three.js load on idle, posters load lazily, and every canvas stops the moment it leaves the viewport.
Assets are immutable for a year, HTML is never cached. It is served from Cloudflare Pages, and an unknown URL gets a 404 rather than the homepage. Under prefers-reduced-motion the film, the morphs and the loops all stop.
It works from the keyboard. Stage tabs are buttons with aria-pressed and arrow keys, the menu traps focus and closes on Escape, the form validates name and email separately.
Measured from day one
This replaces a site that had measured nothing for three years. Google Analytics 4 and Search Console, verified as a Domain property, are in from the start, with the sitemap submitted. Next month the question "did international traffic go up" has a number for an answer.
How it was made
Two days from first commit to live. Design and build ran alongside Claude Code. We are an Anthropic recognised Claude Partner and this is the tool we use on our own work every day. The speed was not only the tool. One reader and five questions leave very little to argue about.
Before and after
| Before | After | |
|---|---|---|
| Declared language | Japanese, on the English page | English |
| On a phone | Desktop layout, scaled down | Responsive |
| Navigation | Image buttons | Text, keyboard operable |
| English words on the homepage | 90 | About 1,000 |
| Page load on a phone | 12.8 s | 0.5 s |
| Requests | 66 | 16 |
| Analytics | Universal Analytics, dead since 2023 | GA4 and Search Console |
| Structured data | None | Organization and WebSite |
| Summary for AI assistants | None | /llms.txt with every stage dimension |
| Hosting | Apache, one server | Cloudflare Pages, at the edge |
Measured 2026-09-09 in headless Chrome from Tokyo, one run per viewport. Phone is 390 × 844, desktop is 1440 × 900. These are single readings from that day, not averages.
What a company in Japan can take from this
- The content already existed. The container was the problem. Every stage dimension and every credit was somewhere on the old site, just not in a form anyone could read.
- International ready is not a translated menu. It is answering the reader’s first five questions, in the reader’s order.
- Being readable by AI is a build decision, not a campaign. It costs a day and it is where discovery is moving.