PROVENANCE LAB

AI Citation Provenance & Referral Router

Citation Ingestion & Routing AI Search Active

96
Attribution Score
2
Tracking Tokens
High
Primary Source Trust
Parameter Value Category
utm_sourcechatgpt.comProvenance Origin
ref_typeai_search_citationCitation Classification

Cross-Lingual Audit & Manifest Verified Cross-Lingual Linkage

1
Source Language Detection: zh-CN (Simplified Chinese)
2
Publisher Identity: The Paper (澎湃新闻 - thepaper.cn)
3
Forwarding Linkage Integrity: Direct News Detail Forwarder (ID: 33723920)

    

What the citation router actually infers

Read the explanation

The router parses the address you enter with JavaScript URL. It calls origin plus pathname the canonical target. Here a product address with a variant query and a fragment becomes only the product path. Every query is removed, including parameters that might identify different content. There is no network lookup, publisher canonical tag, redirect resolution, or page verification. This is local string decomposition, not proof that the resulting address names the correct article. Query entries are assigned into a plain object keyed by parameter name. Repeated names overwrite earlier values. For x equals one followed by x equals two, the retained object contains x equals two, and the displayed token count is one. That count measures unique names, not all query occurrences. The table categories use key spelling: U T M prefix, ref substring, then session or token. These categories are assigned heuristics, not authenticated telemetry. The score is a decision tree over user-controlled text. A source containing chat g p t, or the special AI search citation ref type, chooses ninety six first. Otherwise perplexity gets ninety four, copilot or bing ninety one, and everything else eighty. A perplexity source with that special ref type still takes the first branch. Anyone can type those strings. The score and verified labels are fixed assignments, not calibrated probabilities or evidence that a search engine actually cited the page. The remaining audit labels also follow address patterns. A dot C N hostname or newsdetail in the path assigns Chinese; other addresses assign English. A hostname containing thepaper dot C N gets a publisher label even if those characters are merely a substring of another host. Forward in the path assigns a resolved label without resolving anything. An invalid entry replaces the canonical field and manifest with an error, but leaves earlier engine and score fields visible. Export preserves the current local manifest. Check these assignments and stale fields before using the file as evidence; this tool has not inspected article language or publisher identity.

Super generates helpful tools and automates fact-checking across the internet proactively. If you enjoyed this tool, build your own with Super and share it with a friend.