ASCA01
VERIFIED RE-ANALYSIS

AI SEARCH CITATION ATLAS · OBJECT 01

AI search source overlap changes after hostname normalization.

Removing only a leading www. more than doubled the measured AI Mode–Gemini domain overlap: 4.66% → 10.95%.

OBJECT IDasca:source-overlap@2026-05.v1
OBSERVED
STATUSREPRODUCIBLE

A hostname prefix can masquerade as platform disagreement.

The source dataset records hostnames exactly as returned. Gemini frequently emits a leading www. where another platform does not. Treating those as different domains understates shared sources for every Gemini comparison.

NORMALIZATION SENSITIVITY

Selected pairwise Jaccard overlap

RAW → WITHOUT LEADING WWW.
Platform pairRaw hostnamesNormalizedChange
AI Mode × Gemini4.66%10.95%
Gemini × Grok3.23%9.10%
Gemini × Perplexity3.10%8.68%
ChatGPT × Gemini1.16%3.70%

Jaccard = intersection ÷ union of unique domain sets. The complete 15-pair matrix is in the downloadable CSV.

153,425citation records parsed
6AI search platforms
5,000source queries
15platform pairs

Small transformation. Visible consequence.

  1. Parse the public CSV with a standards-compliant CSV reader.
  2. Build one unique hostname set for each platform.
  3. Calculate pairwise intersection, union, and Jaccard overlap.
  4. Remove only a leading www. and calculate again.
domain = domain.lower().removeprefix("www.")

What this object does not claim

  • The source dataset is independently produced and not independently audited here.
  • Domain overlap does not measure retrieval quality, clicks, or repeat traffic.
  • The snapshot reflects a May 2026 collection and can change by model or interface.
  • Only a leading www. is normalized; no registrable-domain folding is applied.

DOWNLOAD · VERIFY · CITE

Carry the object, not just the number.

CSV SHA-25651f401fd338823440dcb6be4dc622073134b7c23f608917603748fe227302b66
“After removing only a leading www prefix, AI Mode–Gemini domain overlap increased from 4.66% to 10.95%.”asca:source-overlap@2026-05.v1