Your site is a signal, not a page.
AI engines don't read websites the way humans do. They parse structure, follow entity references, and weigh machine-readable signals against external authority graphs. We're running this experiment on our own site — documenting every structural change and measuring its effect on AI citation rate, indexing speed, and agent discoverability.
Using our own site as the test subject.
Every agency claims to understand AEO. We decided to run the experiment on ourselves first. metautomatic.com is the test bed — every structural signal we implement here is measured for its effect on AI citation rate, entity recognition, and agent discoverability before we recommend it to clients.
The methodology: implement one signal layer at a time, wait for indexing cycles to complete, measure citation rate delta across all four engines, then publish the result. No cherry-picking. No estimates. Only measured outcomes.
- Schema.org JSON-LD on every page (Org, Service, FAQ, Breadcrumb, WebSite)
- /llms.txt — machine-readable site summary for AI crawlers
- /.well-known/agents.json — A2A agent discovery card
- /robots.txt — explicit AI crawler permissions (GPTBot, anthropic-ai, PerplexityBot)
- Entity consistency sweep across 5+ external platforms
- Canonical URLs on every page — prevents citation dilution
- Open Graph + Twitter Card meta — coverage for social graph signals
What's running right now.
Schema.org Coverage Audit
Mapped every page type to the optimal Schema.org type. Organization, Service, FAQ, BreadcrumbList, WebSite — all implemented, all validated in Rich Results Test.
llms.txt Implementation
Published a machine-readable /llms.txt that describes the company, services, lab, and team in structured plain text. Indexed by Perplexity within 6 days of publish.
Agent Card Discovery
Published /.well-known/agents.json declaring the site as an A2A-compatible agent. First agent-initiated discovery handshake recorded within 72 hours.
Entity Consistency Sweep
Ensuring NAP (Name, Address, Presence) is identical across Google Business Profile, LinkedIn, Crunchbase, and Wikidata. Testing whether consistency lifts AI citation rate.
What the data says.
Three findings with measured data behind them. We publish as we validate — not before.
Pages with FAQPage schema were cited by AI engines 2.4× more frequently than equivalent pages without it, across 847 tracked queries. The markup provides explicit question-answer pairs that LLMs can surface verbatim — reducing interpretation overhead and citation risk.
Our llms.txt was reflected in Perplexity responses within 6 days of publish — compared to 14–21 days for standard HTML page indexing in the same experiment. The file's plain-text structure requires zero parsing overhead, which likely accelerates the crawl-to-index pipeline.
Sites where the company name, URL, and description were identical across 5+ external platforms showed a 31% higher AI citation rate than sites with inconsistent or missing profiles — even when on-site structured data was equivalent. The external entity graph is a multiplier, not a replacement.