Opportunity Factory: Auto AI Websites for Trusted Businesses

In the age of AI, why do so many businesses STILL have no website? This project automatically generates custom websites and sells them for a reasonable price to reputable businesses in your ZIP code!

Comment

The gap

Some strong local businesses are almost invisible online. Their customers have already done the hard part: high ratings, years of review history, fresh positive feedback, and photos of real work. Yet a prospective customer still has to piece together basic details across maps and social profiles because there is no independent website.

Given all the benefits of a strong online presence, customer acquisition, and brand management, and given how easy AI has made it to make a website, it's surprising to see that many, many local businesses (especially reputable ones with strong public reviews) STILL don't have their own landing page on the internet aside from social media. This is a huge gap that AI can quickly and cheaply close.

This project see this gap as a major opportunity where businesses and AI-enabled developers can enter a win-win partnership to bring them into the 21st century.

What Opportunity Factory does

To keep costs low, we're using Codex and GPT-5.6 to build and power an engine that not only generates custom sites but also finds the right businesses, researches their industry and existing social media, writes a pitch letter selling the website with no obligation, and tracks leads/sales.

The Opportunity Factory starts with a ZIP code as its input and creates whole campaign by following these steps:

  1. Discover businesses through authorized sources and deduplicate them by durable provider IDs.
  2. Apply fixed gates: public review average rating above 4.0, at least 10 reviews, a precisely dated review in the past six months, and no apparent independent website.
  3. Preserve the evidence behind every decision instead of turning search results into unsupported copy.
  4. Build a private, conversion-focused concept site with an industry-specific structure, social metadata, a favicon, a visible concept notice, and noindex protection.
  5. Create a personalized one-page letter with a short customer URL and QR code.
  6. Merge selected letters into one print packet.
  7. Record what actually happened by phone, email, letter, or in person.
  8. When a business says yes, prepare a Stripe Checkout plan for the $500 setup fee and $25 monthly maintenance.

The 74112 campaign proves this loop with 51 researched businesses and 20 strict qualifiers. All 20 now have a private site and print-ready letter, and I plan to personally reach out to them this week (wish me luck!).

Where GPT-5.6 fits in

I leverage this powerful new model to do all the research and prep work for each website. This is what keeps Opportunity Factory from making boring clones instead of truly custom, evidence-backed websites that businesses will be happy to buy.

GPT-5.6 is the evidence critic at the consequential part of the workflow. The dashboard sends a compact sales brief plus the allowed evidence ledger through the Responses API. The model must return strict Structured Output and cite only evidence IDs supplied in the request. Any unknown citation fails validation. The adapter uses low reasoning, store false, no tools, a request quota, a declared cost ceiling, a kill switch, and a dry-run plan that can be inspected without calling the API. This is intentionally narrower than asking a model to invent a website from a business name. GPT-5.6 spots unsupported or weak claims and recommends an evidence-grounded angle.

How Codex helped

Codex was the development environment for the Build Week release. It read the product and compliance rules, implemented the TypeScript packages, coordinated parallel research and adversarial review, migrated SQLite safely, created the 74112 campaign, redesigned the dashboard, and drove Playwright against the real product at desktop and mobile widths. It caught and corrected failures caused by stale local servers before accepting the final test result. I probably looked at the actual code a total of 5 times during this whole process, which is a testament to how far this tool has come just in the last few months.

The result is an operator tool with audit trails and explicit state transitions. The current suite has 186 passing unit and integration tests plus 15 passing Playwright checks, all of which I wouldn't have been able to build independently. Browser tests cover bulk PDF order, contact persistence, responsive layout, distinct inert previews, and the Stripe/GPT dry-run boundary.

Some challenges

The hardest design problem was creating a system that automates almost everything AND still intelligently makes room for accurately tracking manual steps. For example, my theory is that I will have better customer acquisition if I sell these websites in person to local businesses, so I'm printing a pitch letter to hand to business owners in my neighborhood. But printing a letter itself does not mean I've successfully reached out to the business yet. A Stripe success redirect is not payment proof. Those distinctions now live in the state machine, database, tests, and interface, but it took some work to get there. I predict that, as I use this tool, I'll find more opportunities for automation and get a better sense of what exactly needs to be tracked manually.

The hardest production problem was variety. A paint-and-body shop and a vintage store should not receive the same page with different nouns. It took a few iterations to get to an acceptable place, but now the Opportunity Factory uses several conversion architectures and keeps research, visual direction, claims, and evidence separate so a batch can scale without becoming a clone factory. There is still opportunity to grow here and make each website less cookie-cutter. The /impeccable skill has been indispensable when it comes to purposefully staying away from UI that screams AI slop, but I'm still looking for strategies to take this engine's graphic design intuition to the next level.

What is next

The immediate goal is one verified sale and an owner-approved launch. After that, the system can add Stripe webhook verification, owner fact and image approval, client-site indexing, and repeatable neighboring-ZIP campaigns. Acquisition coverage can expand category by category while the same cost, compliance, and evidence guards stay in place. My wife and I are planning to start a family this year, and we hope operating the Opportunity Factory will give us a reliable income stream as a side hustle while benefiting businesses in our city.

Built With

Share this project:

Updates