Bob Michaels/ai

The AI-readable web

Sites AI can read

Websites are becoming software that AI reads, recommends and transacts with. I build the surfaces an AI uses to read a site, and I measure what the major models say about it. This site runs every one of them.

Open to

  • Web and digital director roles
  • Forward deployed engineer and AI solutions architect roles
  • Fractional CTO or interim technical leadership
  • Remote first, Austin. On site when the work needs it.

The surfaces I build

A machine identity

An agent card at /.well-known/agent-card.json and a callable endpoint at /api/a2a, so another AI system can find this site and ask it questions.

Content as code

llms.txt, structured data and plain, quotable copy. Content is the substrate AI reads to decide what to recommend, so I treat it as code.

Agents on the page

A chat agent that answers only from a customer's own content, a WordPress plugin and a script-tag widget that put it on an existing site, and executive agent sites.

A front end for people and machines

AI web front-end design: one set of content served to a visitor and to a machine reader, from a headless CMS that I implement with AI.

Measured, not guessed

The Research Engine reads how ChatGPT, Claude, Perplexity, Gemini and Grok describe, cite and recommend a company, on locked prompts with official-domain verification.

Results and their limits

Client case studies with the numbers and the limits stated next to them, including a head-to-head series won four out of four.

Where to see it

A head-to-head read is a point-in-time measurement of what the models did on those questions, on those dates. Rerunning it is part of the work.

Want your site read by AI?

Tell me what you sell and who should find it. I will measure how the models describe you today and tell you what to fix.