AI SEO & automation case study
An AI SEO agent that gets a local business named in ChatGPT and Perplexity answers.
BlueSail built an agent that checks what ChatGPT, Perplexity, Claude and Google AI Overviews say every week, reads Google Search Console to see how the site performs in Google itself, works out why a business is missing, and fixes what it can on the website by itself. Built for BlueSail first.
What it does
The problem
When someone asks ChatGPT or Perplexity for a software studio near St. Augustine, the engine doesn't evaluate businesses. It reads a handful of sources it trusts (directories like Clutch, third-party "top companies" lists, a few well-targeted pages) and names whoever is in them.
When BlueSail started measuring in August 2026, it wasn't in those answers, and a traditional SEO update a few weeks earlier hadn't changed that. Writing more blog posts wasn't going to help either: the gap was which sources the engines read, not how much content the site had.
What BlueSail built
An agent that runs the whole loop every week without anyone starting it. It asks each engine the searches a real buyer would type, records whether BlueSail is named and which sources the engine cited, and keeps every result so progress is measured as a trend, not a snapshot.
For every search BlueSail loses, it works out why, with deterministic rules rather than a model's guess, so every diagnosis links to the pages that prove it: not listed on the directory the engine reads, listed but outranked on reviews, missing from the third-party lists, or no page that answers the question in the format that wins.
Then it acts. Title, description and structured-data fixes found in Google Search Console go straight to the site as pull requests and merge themselves once the production build passes, with an automatic revert if a deploy fails. New pages and copy are drafted from research into the pages currently winning, then wait for a human to approve them. Anything only a person can do, like asking a client for a review, lands in the weekly email with a draft ready.
Google Search Console
AI answers are half the picture. The other half is Google itself, and Search Console is where Google reports it. Every week the agent pulls which searches the site appears for, with impressions, clicks, click-through rate and average position by search and by page, inspects every page in the sitemap for indexing problems, and checks the sitemap Google is reading.
It acts on what it finds. A page that ranks near the top of Google but gets no clicks means the title and description aren't earning the click, so the agent rewrites them and ships the fix to the site. It flags when the business ranks poorly for its own name, shows which searches the site is gaining and losing impressions on, and alongside Search Console tracks Google rankings for every target search and PageSpeed and Core Web Vitals for every page.
The engineering challenge
The hard part wasn't asking the engines. It was being honest about the answers. AI answers vary from run to run, the consumer apps can answer differently from the APIs, and a tracked list of searches changes as the strategy does. The agent compares each search with its own previous result, stores raw responses, and separates what it knows from what it's guessing.
Letting software change a live website safely was the other half. The agent tiers its own permissions: small, reversible metadata fixes merge automatically after the build passes; anything that adds a page or makes a claim waits for review. It keeps a history of every change so it never re-edits a page before the last edit could show up in the data.
How progress is measured
AI answers and Google rankings shift from week to week, so a single score means little. The agent measures every week and compares each search with its own previous result, so the report shows exactly which searches were won, which were lost, and why.
Every week's email covers what the engines said, which sources they relied on, what Search Console showed, what the agent changed on the site, and the next steps that need a person.
Technical stack
Python workers and CLI, SQLite for every check and citation, Claude for generating searches and drafting pages, the Perplexity, OpenAI and Anthropic APIs for the engine checks, SerpApi for Google AI Overviews and rank tracking, the Google Search Console and PageSpeed Insights APIs, and GitHub plus Vercel for site changes gated on a passing build. It runs weekly on Windows Task Scheduler and emails its report through the Gmail API.
Compared with the alternatives
- Comparable done-for-you AI visibility services charge around $3,500 a month
- Dashboards that report a score but don't say why you're missing or fix anything
- Blog-post schedules that add content without closing the gap engines actually use
Want to know what AI says about your business?
We'll run the searches your buyers type through ChatGPT, Perplexity and Google, and show you who gets named instead of you, and why.
Get an AI visibility check