Technical Architecture
From URL to citation score in one pipeline.
Every CITAQ analysis runs the same deterministic pipeline. No black box, no sampling. Here's exactly what happens.
01 · Crawl
5-tier fetch ladder
CITAQ tries up to 5 methods to retrieve your product page, in order. Each tier escalates only if the previous fails or returns insufficient content.
The escalation stops the moment a tier returns enough structured content. Most DTC pages resolve at tier 1–2.
02 · Extract
4-pass DOM cleaning + schema extraction
The crawled HTML is cleaned in 4 passes: noise removal, attribute normalization, schema.org parsing (JSON-LD + Microdata), and field deduplication. Output: a NormalizedField[] array, the engine's sole input.
<div class="pdp"> widgets noise JSON-LD meta tags </div>
noise removalkey normalizationschema.org parsefield deduplicationNormalizedField[]key: "material" value: "304 steel" source: "schema" confidence: 0.95
03 · Score
7 independent scoring modules
Each module scores a different dimension of AI-readiness. No module sees another's output. They run in parallel against the same NormalizedField[] array.
Total: 100 points. Hard caps apply when critical data is absent: a missing price field can cap the overall score at 82 regardless of other module performance.
04 · Recommend
Archetype-aware recommendations
The engine knows what fields a product in your category should have (155 archetypes). Recommendations are generated only for missing or weak fields that are canonical for your archetype, not generic suggestions. Each recommendation has a fixability tier.
This skincare archetype expects ingredients, a field AI systems check first when evaluating formula claims. You can add it.
Your JSON-LD has a price value but no priceCurrency property. Currency-ambiguous prices are unreliable for AI citation.
This listing is on Amazon. Inventory data is suppressed by platform policy (non-fixable). Move to DTC for full availability-signal control.
05 · Verify
Real AI queries. Real results.
Verification goes beyond static analysis. CITAQ queries real AI platforms with buyer-intent questions in your product's category. It reads the AI's answer, extracts 5 identity claims (brand, model, price, availability, key claim), and compares each against your catalog truth. Contradictions are flagged, escalated, and shown in the identity grid.
This is infrastructure, not analytics. You're not watching a dashboard. You're running a test.