Back to Insights
Engineering

How Optymia Works: Under the Hood of the AI Visibility OS

A transparent look at the scoring model, the agent pipeline and the citation tracker that make optymia.xyz the AI Visibility OS.

Probe layer

Sends buyer-intent prompts to ChatGPT, Perplexity, Gemini, AI Overviews, Claude and Bing Copilot on a plan-tiered cadence.

Crawl layer

Deterministic crawler pulls up to 500 pages, extracts JSON-LD, robots + AI-bot access, headers, entity signals.

Scoring layer

10-pillar model computes the Astra Score, per-pillar breakdowns and per-page recommendations with citation-backed evidence.

Agent layer

27+ agents turn findings into shippable artifacts: schema, internal links, outreach, comparison answers, injection tests.

Fix layer

Native connectors ship the artifact into WordPress, Shopify, Webflow, or export as JSON/PDF.

The 10 Astra Score pillars

  1. Entity clarity
  2. Schema coverage
  3. FAQ / HowTo depth
  4. AI-crawler accessibility (GPTBot, PerplexityBot, Google-Extended, ClaudeBot)
  5. Content freshness
  6. Internal link topology
  7. Technical SEO baseline
  8. Trust signals (E-E-A-T, SSL, HSTS, author markup)
  9. Mention share across LLM answers
  10. Citation strength (backlink + third-party authority)

The agent pipeline

Every agent — Schema Generator, Internal Link Architect, Citation Hunter, Comparison Answer Writer, Prompt-Injection Tester, and 22+ more — follows the same shape:

  1. Load the tenant's latest crawl snapshot.
  2. Prompt an LLM (free OpenRouter models by default) for a structured JSON output.
  3. Validate the response with Zod and reject anything malformed.
  4. Persist the artifact and, when the tenant has a CMS connector, ship it in place.
  5. Emit a webhook so downstream systems (Slack, Zapier, custom BI) stay in sync.

Why the design matters

Optymia is explainable on purpose. Every point on the Astra Score traces back to a pillar, every recommendation cites its evidence, and every agent output is a real artifact you can ship — not a summary. That is why the same math powers both the internal dashboard and the free public tools at optymia.xyz/tools.

Frequently Asked Questions

How is the Astra Score calculated?+

It aggregates 10 pillars including entity clarity, schema coverage, FAQ/HowTo depth, AI-crawler accessibility, content freshness, internal linking, technical SEO baseline, trust signals, mention share, and citation strength. Pillars derived from live LLM probes are capped until enough answer-engine data is collected.

How do the agents ship fixes?+

Each agent loads the tenant's crawl snapshot, prompts an LLM for structured JSON, validates with Zod, and writes back either a copy-pasteable artifact (JSON-LD, outreach draft, comparison paragraph) or a CMS-native patch via connectors.

How does Optymia track ChatGPT and Perplexity citations?+

For each domain Optymia generates buyer-intent prompts, sends them to the supported engines on a plan-tiered cadence, parses answers for brand mentions, and stores per-engine share-of-voice with the exact excerpt as evidence.

What runs where?+

The Next.js app and public tools run on Vercel edge. Background workloads — crawls, competitor sweeps, citation recomputes — run on isolated workers with Postgres and Row-Level Security.

See it live

Run the same scoring engine on your domain — free.

Run free AEO Score

Promoted