I built an AI web monitor in 6 weeks because every existing tool broke the moment a site redesigned
I was refreshing the same five pages every day. A flat listing in London. A restock page for a piece of gear that's perpetually sold out. A council planning page that affects the road I live on. A competitor's pricing page. A small forum thread I cared about.
Every "web monitor" tool I tried — Visualping, Distill, a couple of Chrome extensions — asked me to click a CSS selector or draw a box around the thing I wanted to track. Which works for about a week, until the site redesigns or pushes a new build, and then your alert silently dies. I'd find out three weeks later that the thing I was tracking had changed and nobody told me.
I wanted to just describe what I cared about in a sentence. "Tell me when this flat drops below £2,000/month." "Alert me if this page mentions our street name." "Ping me when this product is back in stock — I don't care if the button says 'Buy now' or 'Add to basket'." I wanted an AI to read the page the way a human would.
So I built it. PageAlert. Six weeks, mostly evenings and weekends.
Stack, for the curious
- Next.js for the app
- Convex for the backend / realtime state
- Playwright in Docker for the actual scraping (handles JS-rendered sites, which 90% of "monitor" tools fail on)
- Hono for a small scraper API
- Claude API for the reasoning step — given the rendered page, does the user's condition hold?
The interesting bit isn't any single piece of that stack. It's that I shipped the whole thing as a solo founder while holding down a senior engineering role, by treating Claude Code, Cursor, and CodeRabbit as my engineering team. I was the architect and the reviewer. They wrote 80% of the code. I think this is going to be the default way solo founders ship in 2026 and people are still under-appreciating how big a shift it is.
What surprised me
I assumed PageAlert would mostly be used for price tracking. That's the obvious use case. It's maybe 15% of what people are actually doing with it.
The use cases that have emerged:
- Parents watching school admissions and council nursery pages
- Journalists monitoring government publication pages for new filings
- Small-business owners tracking competitor pricing pages and job listings
- Indie founders watching forums and subreddits for mentions of their own product
- A guy in Australia monitoring a single product page on a Japanese-only site he can't read, with the condition translated through the LLM
The pattern: people have one specific page they care about, the page changes irregularly, and there is no API for it. That's the entire market. It is much, much bigger than I thought.
What I'd love feedback on
1. Free vs paid from day one. I'm currently free during beta. Half the IH playbook says "charge from day one or you'll never charge." The other half says "free until product-market fit." Which side did you actually fall on, and would you do it the same way again?
2. Niche subreddit outreach — has anyone here found that posting in genuinely small subreddits (the 2k–20k range) actually converts to signups, or is it mostly vanity traffic that bounces?
3. The smallest thing I could do next that would actually move the needle. I keep getting tempted to build features. I suspect the answer is "post in 5 more communities this week instead."
If you want to try it, it's at https://pagealert.io — no signup needed for the first monitor.
Happy to answer anything in the comments.
James
