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
- Entity clarity
- Schema coverage
- FAQ / HowTo depth
- AI-crawler accessibility (GPTBot, PerplexityBot, Google-Extended, ClaudeBot)
- Content freshness
- Internal link topology
- Technical SEO baseline
- Trust signals (E-E-A-T, SSL, HSTS, author markup)
- Mention share across LLM answers
- 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:
- Load the tenant's latest crawl snapshot.
- Prompt an LLM (free OpenRouter models by default) for a structured JSON output.
- Validate the response with Zod and reject anything malformed.
- Persist the artifact and, when the tenant has a CMS connector, ship it in place.
- 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.