The setup
RPMVault is a motorcycle garage and catalog. The web side is a Next.js app (App Router) on Vercel. Search is where most people start choosing a bike, so the catalog was built to be found from day one. Today the public sitemaps list:
- about 19,600 model pages, one per model and year;
- 221 brand pages that act as hubs;
- a set of hand-picked comparison pages;
- the editorial guides, and a few dozen static pages.
Here is what mattered most, in the order I would do it again.
1. URLs that never change
A model page lives at /motorcycle/{make}-{model}-{year}-{id}. The readable part tells people and search engines what the page is; the id at the end makes every URL unique, even when two variants share a name, and lets the page be found again after a model is renamed in the data.
The rule I follow: a URL is a promise. Once a page is indexed, its address should survive data clean-ups, renames and redesigns. If it ever has to change, it gets a permanent redirect.
2. Titles and descriptions from templates, with care
Every model page title follows one pattern: 2025 Make Model Specs & Details | RPMVault. Year first, because that is how people search for a specific bike, and "specs" because that is what they want from the page.
Templates are fast, but they write exactly what the data gives them. If a field is empty, the template will happily print a placeholder word into a description that thousands of people see in their results. Two habits help:
- Write a fallback sentence, not a fallback word. If the category is unknown, leave it out of the sentence entirely.
- Spot-check a random sample of generated titles and descriptions after every data import, not only the pages you know well.
3. Structured data that matches the page
Each model page carries JSON-LD: a Vehicle with its Brand, an EngineSpecification, measurements as QuantitativeValue and the remaining specs as PropertyValue entries, plus a BreadcrumbList that runs from home through the catalog, brand and category to the model.
The rule is simple: structured data describes what is visibly on the page, nothing more. It helps search engines understand a spec page; it is not a place to add claims the page does not show.
4. Split sitemaps
A single sitemap file may hold up to 50,000 URLs, but one giant file is hard to debug. RPMVault uses a sitemap index that points to smaller files by type: static pages, guides, brand pages, comparisons, and the models split into files of at most 5,000 URLs each.
The benefit shows up in Search Console. When each file is a type of page, you can see at a glance which kind of page is being indexed and which is not, instead of staring at one number for 20,000 URLs.
5. Keeping crawlers out of infinite pages
This is the lesson I would put first if I could only share one. A catalog with filters and a comparison tool can generate far more URLs than it has pages:
- Every combination of page number, sort order and search term on the catalog is a new URL.
- Any two models can be compared. With about 20,000 models, that is roughly 200 million pairs.
None of that should be crawled. So robots.txt blocks the parameter versions: catalog URLs with page=, sort= or search=, model URLs with a year= parameter, and comparisons built from query parameters. Private areas like the garage, profile and account pages are blocked too.
Comparisons people actually search for get their own clean URLs instead, like /compare/honda-cb750-hornet-2024-vs-yamaha-mt-07-2025, and only those go into the sitemap. The tool still works for any pair; search engines just see the ones worth reading.
6. Hub pages and internal links
Twenty thousand pages that nothing links to are twenty thousand orphans. Brand pages are the hubs: each one lists its models and links into them, and every model page links back up through the breadcrumb. The goal is that a crawler landing on any page can reach the rest of the catalog in a few steps.
This is also where I still see room to grow: guides that answer a buying or maintenance question are a natural place to link to the exact models they talk about.
7. Caching instead of building everything up front
Building 20,000 pages on every deploy would make each deploy slow. Instead, pages are rendered on the server, cached at the edge and refreshed in the background when they go stale. Visitors and crawlers get a cached page instantly, and a data change shows up without a full rebuild.
What I learned
- Indexing is earned per page. Search engines do not have to index every URL in your sitemap. A page that only repeats a spec table found everywhere else has little reason to rank. The pages that add something, like a clear comparison or an ownership guide, are the ones that pull their weight.
- Data quality is SEO. Duplicates, missing fields and inconsistent units do not just look bad; they turn into duplicate pages and odd snippets at scale.
- Decide what not to index before launch. Blocking a parameter pattern is easy on day one and slow to clean up once millions of URLs have been discovered.
If you want the story behind the product itself, I wrote about why I built RPMVault.