Methodology
A citation-readiness audit should show its work.
GetCitedByAI maps a confirmed page question to simulated follow-up questions, tests exact source passages in isolation, and reports what is covered, broken, or absent. It does not reproduce a provider’s hidden algorithm or predict whether a system will retrieve, rank, or cite a page.
See the method in an example audit, or check a page.
Citation pipeline
Section titled “Citation pipeline”Every AI system that might cite a page moves it through the same conceptual stages before a citation can happen. GetCitedByAI’s audit sits at the end of that chain: it cannot see or change the stages before passage extraction, and it does not claim to.
- Discovery. Can the system find that the page exists at all?
- Crawl / access. Can the system’s crawler fetch and read the page’s content?
- Index / corpus availability. Has the page been stored somewhere the system can retrieve it from?
- Retrieval. For a given query, does the system pull this page as a candidate?
- Source selection. Among retrieved candidates, does the system choose this page’s passage?
- Passage extraction. Can a specific passage be lifted out of the page and still make sense on its own?
- Attribution / citation. Does the system name or link the source when it uses that passage?
These are conceptual states, not a claim that every provider implements the same index, corpus, or sequence. Providers differ in how they discover, crawl, index, retrieve, and select sources.
GetCitedByAI does not determine why a provider discovered, indexed, retrieved, selected, or attributed a source. It evaluates whether the supplied page contains answers that remain usable when tested outside their original context.
From a source to a bounded finding
Section titled “From a source to a bounded finding”- Scope. One page and one confirmed head query define the audit. The report names that scope before showing a finding.
- Acquire the source. We accept a public URL or pasted text, record how it was acquired, and retain an immutable snapshot for the run.
- Simulate questions. The system proposes plausible follow-up questions for the audience. They are an explicit simulation, not a provider’s hidden query decomposition.
- Find evidence. Each question is matched to candidate passages from the supplied source before the evaluator gives advice.
- Test in isolation. A passage is checked as if it were lifted out of the page: can the answer stand on its own, stay in scope, and avoid invented facts?
- Validate the report. Quote matching, classification checks, and prohibited-output checks block incomplete or inconsistent reports from being delivered.
Three classifications, no mystery score
Section titled “Three classifications, no mystery score”- Covered. The answer is present and remains usable when read on its own.
- Not Extractable. The answer is present, but the passage loses meaning outside the page. The finding includes a concrete rewrite direction.
- Absent. The source does not answer the question. The report identifies the missing facts; it does not invent them.
Counts are descriptive. They are not weighted into a grade, percentage, probability, ranking forecast, traffic forecast, or citation promise.
Comparable re-audit or fresh audit?
Section titled “Comparable re-audit or fresh audit?”Comparable re-audit freezes the original head query, question wording and order, locale, methodology, and evaluator profile whenever possible. It tests the same questions against a new snapshot and names transitions such as Resolved, Improved still open, Still open, Maintained, and Regressed.
Fresh audit is a new definition. It may discover a different question set and is not presented as a strict before/after comparison.
What this method cannot tell you
Section titled “What this method cannot tell you”- It cannot guarantee retrieval, ranking, or citation by any AI system.
- It is not a generic SEO, readability, or conversion audit.
- It cannot supply missing subject-matter facts that are absent from the source.
- Its simulated questions are useful probes, not evidence of a provider’s private process.
Every report carries a methodology version
Section titled “Every report carries a methodology version”The rules, evaluator profile, acquisition mode, and retrieval timestamp stay attached to the report so a later comparison can explain what changed.