What I learned building an AI chatbot for websites and docs

I’ve been building an AI chatbot for websites and docs for about 6 months

For the last few months, on and off, I’ve been building a side project called ChattyBox.

It’s a simple AI chatbot for websites, docs, CMS sites, and help centers. The idea is that you can create a project, scrape your site, test the chatbot, and then add it to your website.

The chatbot answers from your own content and shows source citations, so users can see where the answer came from.

The honest reason I started building it

Part of the reason was the product idea itself. I kept seeing AI chatbot tools for websites and docs, but a lot of them felt very enterprise-focused.

Some required booking a demo before you could try the product. Some had unclear pricing. Others felt more like custom consultancy work than a simple self-serve product.

But honestly, a big part of it was that I wanted an excuse to build something real.

I had been hearing about tools like Convex, Clerk, and Autumn, and I wanted to see how fast I could build by leaning on modern infrastructure and product tooling instead of doing everything the hard way myself.

In the past, I’d usually reach for Docker containers, write my own backend, manage SQL queries directly, wire up auth, handle deployments, and build a lot of the plumbing from scratch.

This time, I wanted to try a different approach:

• use Convex for the backend/database layer
• use Clerk for authentication
• use Autumn for billing/subscription logic
• keep the app mostly serverless
• avoid managing infrastructure too early
• focus more on the product and less on backend plumbing

I wanted to see how far I could get by using tools that scale up and down without me needing to think too much about servers at the start.

What I was trying to build

The flow I wanted was simple:

1. Sign up
2. Create a project
3. Add your site
4. Let the system scrape the content
5. Test the chatbot
6. Add it to your site

No sales call. No complicated setup. No unclear pricing.

The goal is to make AI support easier to launch for SaaS sites, documentation sites, CMS sites, and support-heavy websites.

What I’ve learned so far

The biggest thing I’ve learned is that first impressions matter a lot.

I had one early user sign up and email me about a bug. I fixed it, but by the time I had solved it, they had already moved on.

That was annoying, but also useful. It made me realise that for this kind of product, onboarding has to work almost immediately. If someone tries to add their site and hits friction, they probably won’t come back later.

So now I’m focusing more on the setup flow, error handling, and making it obvious what to do next.

The hard part is trust

The technical side is important, but I think trust is the real challenge.

If someone is going to put an AI chatbot on their website, they need to know:

• what content it is using
• where the answers come from
• whether it will make things up
• how easy it is to remove or change
• whether it will be expensive if traffic grows

That’s why I’m trying to make citations and simple pricing part of the core product, not an afterthought.

The modern tooling helped a lot

Using tools like Convex, Clerk, and Autumn changed how I approached the project.

Instead of spending weeks building auth, billing, backend structure, and deployment plumbing, I could get a lot of that handled by tools that already exist.

That does not mean everything is effortless. You still have to understand the product, data model, edge cases, onboarding, errors, and user experience.

But it did let me spend more time on the actual product rather than constantly getting pulled into infrastructure work.

For this kind of early-stage side project, that has been a big advantage.

Where it is now

The MVP is live, and I’m starting to look for early users and feedback.

I’m especially interested in feedback from people who run:

• SaaS websites
• documentation sites
• CMS sites
• help centers
• support-heavy websites

I’m not treating this as a huge polished launch yet. It’s more of a soft launch while I work out what people actually need and where the product is unclear.

The site is here if anyone wants to take a look: https://chattybox.ai

I’d be interested to hear what would make you trust an AI chatbot enough to put it on your own site.