How I launched a 1,140 page directory with free public data
I launched a new directory recently, and within 24 hours Google had already indexed it (90 pages) and it was ranking for 4 keywords.
Five days later, it's ranking for 20 keywords.
It's headlightbulbsizes.com. Pick your year, make and model, and it tells you which headlight bulbs your car takes (low beam, high beam and fog lights).
20 keywords. 5 Days later. Found in the "Performance" tab in Google Search Console.
Top-level view of the keyword volume.
Way too early to call it a win. But it's exactly the signal I look for after every launch, so here's how I built it (the simple version, not the hour long one).
How to check if Google picked up your site
Two free ways. First, go to Google and type site: "headlightbulbsizes.com" (swap in your own domain). Every page that shows up is indexed. If nothing shows up, Google hasn't gotten to you yet.
Second, open Google Search Console and go to the Performance tab. That shows the actual searches you're appearing for, with impressions and clicks.
Early on I don't care about clicks. I care whether those are the keywords I was aiming for.
I found the idea right next to a directory that already works
I'd been going through mpgbuddy.com for a video and wondered what other boring car databases might just work.
So I searched a few car part keywords plus the word "directory" and found this.
DR 12, about 4,500 visitors a month, with affiliate links to every wiper brand underneath.
(Google's AI Overview answers the question right at the top, and the site still gets clicked. Don't let that scare you off a niche.)
I didn't want to copy it though. Wiper blades are $20 to $40 and people barely replace them. So I went looking for the car part next door.
Headlight bulbs. About 684,000 searches a month across the keyword family, tons of it at KD 0. And when I googled "headlight bulb directory," there wasn't a real one.
That's the whole filter. A proven model next door, big search volume, and no dedicated directory.
The data was free, it just took 5 days
Before I build anything I check two things. Is there public data, and how many pages is this going to turn into?
The vehicle list came from four free government databases, merged by Claude Code into one list of every year, make and model.
The bulb sizes were the hard part. So first I asked Claude Code to find the bulbs for a 2011 Subaru Outback from the official owner's manual. It got it right, and that one test was enough.
Pro-tip: When dealing with a potentially massive directory, you can export Ahrefs keyword data and hand it to Claude code. This helps prioritize which data and pages to build first.
Then I let it run. Wednesday to Sunday.
By the end it had about 26,000 bulb rows across 8,462 vehicles, mostly from manufacturer manuals. That covered 97.2% of the search volume in my export (and I sent maybe 5 to 7 prompts the whole time).
Why I ran it on a VPS instead of my laptop
A scrape like this takes days, and I'm still not technically "finished" with my data scraping.
A VPS is basically a computer in the cloud that's always on. I kicked off the scrape, went to the gym, went to sleep, and it just kept going.
My favorite part was the usage limits. Every time I hit my Claude Code limit, the server picked right back up when it reset. No babysitting.
And when the 2028 models come out, it can scrape those and add them automatically (same idea for events, job boards, anything that needs constant updating).
- Paid Partnership -
I ran this whole build on Hostinger's Claude Code VPS. You pick a plan, choose Claude Code during setup, and it comes installed and ready to go (I went with KVM 2, but KVM 1 is plenty if you're running one lighter directory).
It stays online around the clock even when your computer is off, so long scrapes and data updates keep running in the background.
One page per year, make and model would've meant thousands of thin, nearly identical pages. Tough sell for a brand new domain.
So I went with pillar pages. One page per make and model (say, the Honda Accord) with every year on it, grouped wherever the bulbs are the same.
Each year group gets its own heading, so the page can still rank for "2015 Honda Accord headlight bulb." And picking your car on the homepage jumps you straight to your section.
Then bulb size hubs. "H11 headlight bulb" is a shopping search, so each size gets exactly one page (that doesn't repeat the pillars, so they never compete).
Launch day was mostly housekeeping
GitHub repo, Vercel project, domain. Claude Code even ran the Vercel command line (CLI) for me to connect the domain and verify Search Console.
In non-coder terms: I prompted Claude code to "use the Vercel CLI" and it did all the work for me once I was ready to launch the site!
Then the boring essentials: sitemap in Search Console, Google Analytics, a contact form, and a click tracker so I know which bulbs and stores people click before I pick affiliate programs.
The data is still scraping in the background and gets added as it comes in. I didn't wait for 100% to launch.
The game plan from here
Let it sit for a couple of weeks. It's already showing up for the right kind of keywords, so directionally it's aimed correctly.
Then backlinks. Relevant ones, from pages already ranking for searches like "Honda Accord headlight bulb size," pointing to my matching page.
Then I watch for traffic. Affiliate links come after, once the click data says which programs are worth it. I installed a click tracker to see which brands get the most clicks so I can apply to the affiliate programs according to data.
Honestly the only two real costs on this build are Claude Code and (soon) backlinks. The data was free. I used the Astro framework too, so I'm not paying Supabase for data storage!
Will it work? No idea yet. 20 keywords in a few days is a nice early sign, not proof. I'll share the numbers in a future update.
If you want the full build, every prompt and mistake is in the video. And if you've built something off public data recently, hit reply. I'd love to see it.
