shipped

Healthcare AIO

Why AI search won’t cite you, and what to fix.

Visit site

Patients search differently now. It’s not “pediatrician in Scranton” anymore. It’s “what rash is this — do I need to go in?” And they don’t click blue links on Google. They click through AI chat.

SEO is becoming AIO. We built Healthcare AIO to measure how a provider actually shows up in ChatGPT and Perplexity, then show them exactly what to fix. Every claim about a physician is checked against the federal registry, so the audit is grounded in real records, not model guesses.

At a glance

Problem

Patients increasingly ask AI search engines for provider recommendations. A hospital that isn’t cited is invisible at the moment of intent — and there’s no “page two” to inspect.

Solution

I built a vertical platform that crawls a provider’s site, measures real citation share across the AI engines, grounds an LLM diagnostic in federal physician records, and outputs a scored, persona-tailored audit of exactly what to fix.

Why it was hard

A wrong claim about a physician is a liability, so the diagnostic had to be grounded. I verify physician facts against the federal NPPES registry with confidence labels, keep the scoring deterministic, and detect bot walls that hard-cap the headline score so it can’t lie.

Impact

It’s a working end-to-end engine: one URL in, a multi-pillar, grounded, persona-tailored audit out, with provider failover. The public demo reports double as real sales collateral.

AI components

  • LLMs

    Claude Sonnet (primary), Gemini, and GPT behind one interface with schema-validated failover — a structurally wrong answer is treated as a failure, not shipped.

  • Agents

    None — a deterministic Inngest workflow with scoped per-page LLM judgment, a deliberate choice over autonomous agents.

  • RAG

    Structured grounding — crawled page text, NPPES-verified physicians, and live citation results assembled into each prompt; no vector store.

  • Evaluations

    Structural guarantees (Zod gates, deterministic floors) in place; automated quality eval identified as the next investment.

  • Workflows

    A durable crawl → enrich → measure → diagnose → score → narrate pipeline with per-step retries and bounded concurrency.

  • Automation

    An 8-query AI-search citation battery; bot-wall detection by fetching the homepage as five real crawler user-agents.

Technology

Next.js • Claude • Gemini • GPT • Inngest