Using data to build with AI
birthday-cards.ai started as a fun way to make cards for friends and family. I built it end to end with AI, and its analytics connect to the same AI system in Claude Code that builds it. The numbers decide what gets built next. That is the 2026 model I work in. AI does as much as it can, and a person sets the direction.
The numbers as of today
Antigua and Barbuda, Argentina, Aruba, Australia, Azerbaijan, Bahamas, Belarus, Belgium, Brazil, Bulgaria, Canada, Croatia, Czechia, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, India, Iraq, Ireland, Israel, Italy, Japan, Kazakhstan, Liberia, Malaysia, Malta, Mexico, Myanmar (Burma), Namibia, Netherlands, New Zealand, Nigeria, Norway, Pakistan, Papua New Guinea, Poland, Portugal, Qatar, Romania, Rwanda, Saudi Arabia, Seychelles, Singapore, South Africa, Spain, Sri Lanka, St. Lucia, Sweden, Switzerland, Trinidad and Tobago, Turkey, Ukraine, United Kingdom, United States, Uzbekistan, Zimbabwe
Refreshed October 20, 2026 from Vercel Web Analytics on birthday-cards.ai. Test traffic excluded.
- Vercel Web Analytics
- Google Search Console
- Bing Webmaster
- Airtable
- Vercel runtime logs
- Claude Code
How it started
My friend gave me a personalized gift made with AI. Personalized gift giving is already very popular, and I knew that with AI it will only grow.
I also love giving cards and writing cards. It's one of the few times you have a good excuse to be vulnerable with someone and share how you feel.
So I built birthday-cards.ai. I was the only user for a few months. Then a few friends and family tried it and I got good feedback. I was printing the cards at home on cardstock, and the cards I gave to people in real life were well received.
Getting set up
I write SQL. I'm not a full stack developer. So there was a learning curve. I had to get set up with GitHub, get comfortable on the command line, and learn Claude Code. Vercel hosts the site. OpenAI draws the cards. Airtable holds the emails.
The piece that guides all of it is a system of .md notes. A briefing, a state file that gets rewritten every session, a session log, a style guide. Claude reads them before it writes any code.
- Build
- Claude CodeGitHubcommand line
- Host
- Vercel
- Draw
- OpenAI
- Measure
- Vercel Web AnalyticsGoogle Search ConsoleBing WebmasterAirtable
- Notes
- .md notes
The tools behind birthday-cards.ai.
A few sessions from random places
Anyways, I saw a few sessions from random places. So I set up analytics further and kept working on this as a passion project.
Visitors from search by month.
In the last 30 days 297 of 818 visitors came from a search engine.
Three pieces of work moved those bars. In July, one indexing fix. Every search signal had been pointing at the wrong host. In August the analytics went live and the first weekly read ran. In September I built thirteen pages, each around one thing people search for.
One page, from a guess in June to the best converting page on the site. One minute, no sound, and it loops.
I can also see how people make their cards.
Search visitors, September 20 to October 19. 291 cards.
More than half of the cards start with someone's photo.
A glimpse into the future
Once I got all the systems connected, it became really cool to see a glimpse into the future (or the now). Claude Code reads Vercel Web Analytics, Google Search Console and Bing Webmaster and provides the analysis. Then we work together to plan the next product development steps.
The loop. Five sources feed Claude Code, a person decides, and the next build ships to the site.
AI is not doing all of it. It does the heavy lifting on the analysis, and it ties things together so nicely. We set rules on what works and what doesn't, so as we develop it learns about what worked and what didn't. We have brand guidelines and tone guidelines, and all of these things mean the project compounds knowledge.
I'm sure there are some risks associated with this workflow. But I move slow and deliberately, or at least I try to.
Where it stands
It's all free now. It's my lead generator for GLF Analytics, to show what I can do. If it gets enough traffic, I might consider paid upgrades and ways to monetize.
It's all new. So this is my test project, my passion project, my playground, and a utility that has saved me a lot on cards I would otherwise be buying from the store.
-Gabe
Questions
Where do the numbers on this page come from?
From Vercel Web Analytics on birthday-cards.ai, read when this site builds. Test traffic is filtered out.
Who makes the decisions?
A person does. Claude thinks through each trade-off with me, builds what we pick, and verifies it. The release goes out after my review.
Working on something like this?