Methodology
Calculation version 1 is methodology metadata, not a product version or score.
A. Purpose
The AI Search and GEO Agencies Citation Index measures which web sources were explicitly cited during the AI research used to produce AI Marketing Consensus Index consensus studies. It is a descriptive record of observed citation behavior, not an evaluation of product quality or a source-authority score.
B. Dataset Scope
This snapshot includes 14 published AI Search and GEO Agencies studies, 7 AI platforms, and the valid research responses recorded in the published study snapshots. The studies cover distinct AI Search and GEO Agencies buyer needs in the United States.
The measured AI platforms are:
- OpenAI
- Anthropic
- DeepSeek
- Grok
- Perplexity
- Kimi
- Google
C. Unit of Analysis
The primary unit is a normalized URL cited in a valid AI research response. Domain-level results are aggregations of those URL records.
D. Citation Occurrence Counting
A normalized URL may contribute no more than one Citation Occurrence within a single valid response. It may contribute another occurrence when cited in another response, study, or platform. Multiple different URLs from the same domain may each contribute an occurrence within the same response.
E. Citation Share
Citation Share is a domain’s or URL’s Citation Occurrences divided by all Citation Occurrences in the snapshot. It is not Response Coverage. Rank does not use Citation Share; the percentage is a descriptive share of observed citations.
F. Citing Response Counting
A domain or URL is counted no more than once per valid response when calculating Citing Responses and Response Coverage.
G. URL Normalization
Hostnames are lowercased, www. is removed when it does not represent a separate property, fragments and known tracking parameters are stripped, default ports are removed, and trailing-slash behavior is standardized. Original cited URLs are preserved in the underlying record.
H. Redirect Wrappers
Google grounding redirect URLs and other platform-generated redirect wrappers are resolved to the final publisher URL whenever possible. Unresolved wrappers are retained in the audit file but excluded from publisher-domain rankings because the wrapper platform is not the underlying source.
I. Domain Normalization
Domains are grouped using the registrable domain. Subdomains are generally merged into the registrable domain unless a subdomain represents a materially separate publishing entity that the methodology intentionally preserves.
J. URL Validation
Every observed citation is retained as evidence of what the AI platform returned, even when the destination is later broken or unavailable. Destination availability is not validated in this snapshot (ASG-2026-09-V1). A later snapshot may report Live, Redirected, Broken, or Unresolved status after an automated check. A valid HTTP response would not establish that a page’s claims are accurate.
K. Response Inclusion
A response is valid only when it is complete, the required schema can be parsed, the study and platform are identifiable, the response can be associated with its correct research record, and the citation data is not corrupted.
L. Response Exclusion
Failed platform calls, empty responses, truncated responses, malformed structured data, duplicate response records, and responses that cannot be associated with a valid study record are excluded from the denominator. A failed response is never replaced with an assumed zero-citation response.
M. Source Relationship
Company-owned is assigned only when the cited registrable domain matches the official domain of a qualified company evaluated in the research (including that company’s known parent or subsidiary identity in the research entity table). Independent, in this column, means the source is not that evaluated-company domain. A commercial publisher is not Company-owned merely because it is a business. Mixed is used when a domain contains both classifications in different study contexts. Unclear is used when ownership could not be established with sufficient confidence. Source Relationship is not used in ranking or Citation Rating.
N. Publisher class and evidence mix
Publisher class is a separate layer from Source Relationship. Company-owned uses the same evaluated-company rule. Remaining domains are classified as Independent publisher, Government / nonprofit, Community / UGC, or Other using a curated domain table plus heuristics such as .gov/.mil hosts. Unclassified remainder is Other, not Independent publisher. The evidence-environment mix is occurrence-weighted Citation Share across those five classes. It is not a count of domains and does not measure influence or quality.
O. Ranking Rules
Domain and URL rankings are ordered first by Citation Occurrences, then by Citing Responses, study coverage, platform coverage, and alphabetical name as the final tie-breaker. Overall Rank is the snapshot rank and is not renumbered when a visitor filters or paginates.
P. Interpretation Limitations
Explicit citation frequency does not reveal every source a platform retrieved or used internally. AI platforms have different citation interfaces, retrieval systems, and source-display behavior. A citation does not necessarily represent endorsement or positive sentiment. A highly cited source may contain inaccurate, outdated, duplicated, or company-supplied information. These results are a snapshot and may change when platforms, search indexes, products, pages, or source environments change. This index is not a product ranking and should not be used to determine which AI Search and GEO Agencies option is best for an individual consumer.
Q. Affiliation Disclosure
AI Marketing Consensus Index and LLM Authority Index are affiliated publications. AI Marketing Consensus Index produces the underlying consensus studies, while LLM Authority Index aggregates and analyzes the citation records. Companies cannot purchase inclusion, ranking position, or favorable treatment in either index.
