Building the AI Readiness Audit Tool

It started small, as most projects do. I wanted to know if my website would get mentioned in AI chatbots as an authoritative source for the services I offer. I tried a few online “score my site” tools and while they weren’t necessarily bad (those that worked), they didn’t address the question in the way I would have liked.

I wanted a tool that would explain what was needed and why, in language that anyone could understand. Since AI is pretty good at summarizing things when given the proper instructions, I decided to build something that other website owners would find useful. And full disclosure, if they liked it, they might want to hire me to do it for them.

I Am Not a Programmer

I’ve been building websites for years and can find my way around most code. However, I don’t consider myself a programmer who can build an application from the ground up. That’s okay though, because in the brave new world of AI, the robots can write the code – as long as they have a detailed specification. And that I can do.

My platform of choice is Claude by Anthropic. The first step was to open up a conversation and start describing what I wanted to build. I found that the interactive process of working through the problem with Claude helped to refine my thinking considerably. While it’s most certainly not human, it asked good clarifying questions and made suggestions that would not have occurred to me.

One of the more important decisions that came out of those conversations was that the tool needed two distinct outputs:

1. A prospective client gets a plain-language report organized around four questions every homepage has to answer.
2. I get a working brief with specific technical findings, organized by type and prioritized for a redesign engagement. Same input, two very different results.

Once I was satisfied that we had worked through all of the decisions, I asked Claude to prepare a detailed specification for the tool.

The “Programming” Process

After some trial and error, we settled on using a combination of Claude.ai and Claude Code to build the plugin. Claude.ai is a web-based chat interface – good for planning, drafting specifications, and working through decisions. Claude Code is a separate tool that runs on a local computer and works directly with project files. It can read, write, and edit code in your actual development environment, rather than just displaying code in a chat window for you to copy and paste. Think of Claude.ai as the architect’s desk and Claude Code as the construction site.

The specification included phases, just as it would if it were being used by human programmers. That way, Claude could start by building a minimum viable product which could then be tested. Once that stage was working, we’d move on to the next. The actual handover prompt to Claude Code was:

“Build a WordPress plugin following the attached specification. Start with Phase 1 only: plugin scaffold with activation and deactivation hooks, database table creation, settings page with API key entry and connection test, URL fetching and HTML parsing, Claude API integration, client report rendering (functional before styled), basic submission logging, and a SETUP.md file. Do not begin Phase 2 until Phase 1 is working and reviewed.”

When in Doubt, Iterate

Then followed a couple of weeks of trying things out, going back to Claude, explaining what needed to change, and having it generate new code. As Reid Hoffman, co-founder of LinkedIn, put it:

“If you’re not embarrassed by the first version of your product, you’ve launched too late.”

The audit tool is well past its embarrassing early version, but that version was a necessary step.

What We Launched

This free, self-service AI Readiness Audit Tool is live now on the Webdancers site. It runs a structured analysis of any site’s homepage and produces a report showing what’s working and what needs attention.

You enter a URL and answer a few questions about your business type and location. The tool fetches your homepage, analyzes the content, and produces a report organized around four questions that every homepage needs to answer:

• What is this?
• What can I do here?
• Why should I be here instead of somewhere else?
• What do I do next?

For each question, the report explains what the site communicates, what’s missing, and what that means for AI search visibility. It closes with a summary of what’s working and what’s needed.

On the back end, each time an audit is run, it creates a structured brief for me. The brief surfaces specific issues and organizes them by workstream – content, schema, technical structure, navigation – so that I have an overview of the work that might need to be done.

Lessons Learned

• AI programming (a.k.a. vibe coding) only works well when it has a clear objective and a detailed specification.
• The rounds of iteration that followed the initial build weren’t signs that something went wrong, they were how the tool learned edge cases that couldn’t have been anticipated. Some things only surface when the tool meets real-world conditions.
• The tool is only as good as the criteria behind it. Claude can evaluate a site against a framework, but it can’t invent the framework. The four-question structure came from years of client conversations, not from the AI. Domain knowledge is still the scarce ingredient.
• Knowing what to hand off and what to keep is its own skill. The decisions that required judgment stayed with me. The execution that was well-defined enough to delegate went to Claude Code. Getting that boundary right is what makes the result unique.

Postscript: How This Post Was Made

This post is itself an example of the process. Rather than asking Claude to write from scratch, I wrote a rough draft with my own words and left bracketed instructions in the places where I wanted Claude to fill in specific sections – explaining Claude Code, finding the right quote, describing what the audit report shows.

One of the more useful things Claude did here goes beyond filling in the blanks. This project has a dedicated workspace in Claude where the full plugin specification, workflow documents, and related files are stored. When it came time to describe what the audit report actually shows the user, Claude went back into that project knowledge, pulled out the relevant details, and wrote that section accurately – without me having to look it up or rewrite it. For a post like this, that’s a minor convenience. For a longer or more technical document, the ability to draw on a structured body of project knowledge without being prompted to do so is genuinely powerful.

Claude generated the bracketed sections in my voice, working from a style guide developed from my past writing, and left my own text largely intact.

What AI did: generated the bracketed sections, retrieved accurate technical details from project files, maintained consistency of tone throughout, and suggested small edits for flow.

What I did: wrote the original draft, made every structural decision, chose which sections to write myself and which to delegate, and edited the final result before publishing.

It’s a reasonable division of labor. The parts that required my experience and judgment stayed with me. The parts that were well-defined enough to hand off, I handed off.

AI disclosure: This post was developed with AI assistance. I use Claude (Anthropic) to fill in designated sections of my own draft, working from a style guide. All content reflects my own views and judgment.