What’s on the site
| Section | What it does |
|---|---|
| The archive | 29 cars from Toyota, Nissan, Honda, Mazda, Mitsubishi and Subaru, from the AE86 to the R34 GT-R, browsable by maker |
| Car entries | Engine, power, drivetrain, transmission and weight in a fixed format, then the story of why the car matters and what to watch for today |
| Guides | The bigger picture: RB26 vs 2JZ, the real story of Japan's 280 PS gentlemen's agreement, and importing a JDM car to the USA under the 25-year rule |
The idea
JDM car information is everywhere and inconsistent: forum posts, wikis and listings that disagree on basic specs. I wanted a reference that answers the same questions for every car, and that you can trust because every entry is checked.
How it’s built
This is my test bed for an AI content workflow. AI gathers sources and drafts each entry into a fixed structure; I check the specs against those sources, edit the story and publish. The structure is published as structured data, so search engines and AI assistants can quote a spec with confidence.
Stack Next.js · structured content model · LLM API · Schema.org
Skills it shows
- AI transformation: an editorial workflow where AI does the research and first draft, and a person owns accuracy. The same shape works for product content in a Shopify store.
- AI development: prompts, a content model and checks that keep 29 entries consistent.
- Shopify-relevant thinking: a car entry is a product page by another name. Clean, structured attributes are what AI shopping assistants read.
What I learned
- When AI can write anything, "curated and cross-checked" is the point of difference.
- A strict content model matters more than the writing. It's what makes a page quotable.
Behind the scenes
Why I built it
I was born in the 80s and grew up watching Japanese performance cars hit their peak. I wanted one place that explains why each of those cars matters, written with care and checked before it goes live.
What exists today is split in two. Classifieds sites show price but little context, and open wikis have depth but inconsistent structure and quality. Forums hold great knowledge, but it is buried across years of threads.
The primary audience is enthusiasts researching a specific chassis, often before buying or importing one. The secondary audience is newcomers who want to understand JDM culture, plus people dreaming about their next build.
What it deliberately does not do:
- It is not a classifieds site and does not sell cars.
- It is not an open wiki, so nobody can edit entries freely.
- It does not chase every model ever sold. It covers the cars that defined an era, one finished entry at a time.
The methodology
The site runs on a short set of rules, set out on the About page.
- One finished entry per chassis. An entry goes live complete, not as a stub to fill in later.
- Specifications are checked. Figures are checked against factory data and period source material before publishing.
- Disputed numbers are shown as disputed. Where sources disagree, the entry gives the range and says why rather than picking a convenient figure.
- Same structure every time. Every entry uses the same sections, so readers can compare cars quickly.
- Honest about AI. Some drafting and formatting is assisted by AI, and the site says so. Final editorial decisions and accuracy checks stay with me.
- Visible review dates. Each entry shows when it was last reviewed, so readers know how fresh it is.
- Open to corrections. Every entry links to a correction form, and the operator details are published on the Contact page.
- Readable everywhere. Pages are fast and mobile friendly, with specs labelled in both English and Japanese.
Where the content comes from
This is an editorial site rather than a data feed, so most sources are research inputs. These are the ones each entry draws on.
| Source | Publisher | What it powers | Refresh |
|---|---|---|---|
| Factory data and period source material | Manufacturers and period press | Specification of record and history sections | Checked at each entry review |
| Wikimedia Commons | Wikimedia Foundation | Documentation photographs, as images move to openly licensed sources | Per entry |
| YouTube | Embedded reference footage on each entry | Per entry | |
| Google Search Console | Which searches bring readers in, and which titles and entries to rework next | Exported and compared period over period |
Live market listings and auction price data aren't covered yet.
The hard problems
A few problems took real work to solve, and some are still in progress.
Writing that sounded like AI. Early entries had clear AI tells: bolded topic sentences, stacked intensifiers, unverifiable superlatives and inspirational closings. I named six patterns, built a banned words list and hard formatting rules into the drafting process, and now edit every draft by hand. Older entries are being rewritten to the same standard.
Ranking but not getting clicked. One Supra entry reached page one of Google with zero clicks. The content was fine; the title and meta description weren't earning the click. I now treat titles and descriptions as conversion copy, the same way I'd treat a product page.
Stuck on page two. Several high impression entries, including the EK9 Civic Type R, Impreza 22B and NSX, sat around positions 14 to 16. I rewrote six of them to match what searchers were actually asking.
Templated copy everywhere. All six brand hubs shared the same intro text, so search engines had little reason to tell them apart. Each hub now has its own intro, title and meta description.
Numbers that don't agree. JDM specs are messy because of Japan's 276 hp gentlemen's agreement and different market versions. Rather than hide the conflict, entries show the official figure, the measured figure and the reason they differ.
Images I didn't control. Some photos were loaded from other sites, which hurt speed and reliability and raised licensing questions. I'm moving galleries to openly licensed images, resized so pages stay fast.
Technical SEO and organic traffic
Search is the main way readers find the site, so I treat it as a product in its own right. In the first three months, search impressions grew roughly sevenfold.
Matching pages to real searches. I read search query data before writing anything new. Entries are named the way people search, by model and chassis code, and guides target questions people actually type, like how to import a JDM car to the USA.
Content clusters and internal linking. Brand hubs group every entry for a maker. Each entry links to a sibling, a rival and a related guide, and guides link back to the entries they mention. Readers and crawlers can reach any page in a few clicks.
Technical foundations. Every page has a unique title, meta description and canonical URL, plus share images and breadcrumbs. Pages are fast and mobile friendly, and FAQ, article and breadcrumb structured data are being rolled out across entries.
Fast indexing. New and updated pages are submitted through Google Search Console and Bing Webmaster Tools, with IndexNow pinging supporting search engines on publish. Clear structure and plain answers also help the site surface in AI search tools that cite their sources.
Trust signals. Review dates, a published operator, a correction form and an honest AI disclosure all show readers and search engines that a real person stands behind the content.
Things worth linking to. Buyer's guides, import walkthroughs and head to head comparisons are the pages other enthusiasts share. That is where links come from, not outreach spam.
What's next
- More entries. I have a prioritised list of 40 cars in three tiers, starting with the top ten.
- Import guides for more countries. Guides for the UK, Australia and New Zealand, alongside the existing USA guide.
- More head to head pages. Direct comparisons between rival cars, built from the same specification data.
- Current market listings. Live pricing on each entry so readers can see what a car costs today.