Citation Integrity: The Complete Guide to Dependable Evidence for AI

Why did AI answers change what "publishing well" means?

For twenty years, publishing well online largely meant being findable. A search engine returned a list of links and a person decided what to trust. AI assistants changed the shape of that decision. The system now reads sources, decides which ones to rely on, composes an answer, and — sometimes — attributes it.

That shift moves the burden. Whether your organisation appears in an AI answer depends less on how well a page is optimised and more on whether the evidence behind it can be relied upon, verified and correctly attributed. Research on attributed generation makes the distinction explicit: a citation is only useful if the cited material actually supports the statement it is attached to. A 2023 evaluation of generative search engines found that a substantial share of generated statements were not fully supported by the sources cited alongside them.

NIST published its Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (AI 600-1) in July 2024. It describes confabulation — confidently stated content that is not grounded in fact — and information-integrity risks as characteristics that AI actors are expected to manage. Systems built to reduce those risks are, by design, selective about what they lean on.

Citation Integrity™ is our name for the property those systems are selecting for.

What does Citation Integrity™ cover?

Citation Integrity™ is the degree to which published information can be relied upon and correctly attributed. It concerns the evidence, its provenance, the standing of whoever published it, and whether a machine can find, parse and attribute it without guessing.

It is deliberately broader than a content checklist. A well-written page with no verifiable evidence behind it has low Citation Integrity™. So does a rigorously evidenced page that a crawler cannot reach, or one published by an entity that cannot be identified.

What are the eight dimensions?

CiteAbility™ assesses Citation Integrity™ across eight public dimensions. Each dimension is a question rather than a score to be gamed.

We publish the dimensions and the questions they ask. We do not publish underlying signals, weightings or thresholds, because a published formula becomes a target and stops measuring anything real.

How do the dimensions relate to each other?

They are not independent scores to be averaged in your head. Three relationships matter most.

Evidence depends on identity. Evidence & Citations and First-Hand Experience only carry weight once Source Authority and Entity Authority establish who is speaking and why they would know. Anonymous rigour is hard to rely on.

Standing conditions reliance. Organizational Probity is about material verified and dated information concerning conduct and standing — an official finding, an adjudicated sanction, a company's own disclosure. Absence of such information is neutral. It is not a mark against an organisation, and an unresolved allegation is not treated as an established fact.

Accessibility is a gate, not a virtue. Technical Accessibility can prevent everything else from counting. If a page cannot be retrieved, parsed or attributed, its evidence is invisible. But being perfectly machine-readable adds nothing on its own.

Integrity Analysis is the step that considers the other seven dimensions together and asks the question that matters in practice: should this be relied upon?

What is Citation Integrity™ not?

• It is not SEO. Ranking in a results page and being relied upon inside a generated answer are different outcomes with different requirements.
• It is not "AI SEO", GEO or AEO. Those describe tactics for influencing AI surfaces. Citation Integrity™ describes a property of your evidence. The two are not interchangeable.
• It is not Google's E-E-A-T. E-E-A-T is a set of quality concepts used in search rater guidance. Citation Integrity™ is an assessable framework aimed at machine reliance and attribution, and it includes dimensions — notably Organizational Probity and Technical Accessibility — that E-E-A-T does not address in those terms.
• It is not fact-checking. Fact-checking adjudicates individual claims. Citation Integrity™ assesses whether a body of published information, and the organisation behind it, can be relied upon.
• It is not brand monitoring. Counting mentions tells you what happened. Citation Integrity™ tells you what your evidence deserves.

How do tested evaluation and observed citation differ?

Two things get conflated constantly, so we keep them separate everywhere in our products and writing.

A model saying it would cite you is not evidence that it has, or that it will.

How does CiteAbility™ apply the framework?

• The Citation Integrity™ Audit examines what an AI system can actually retrieve from a website and where the evidence falls short.
• The Citation Integrity™ Score tracks assessed position across the eight dimensions over time.
• Citation Integrity™ Monitor re-checks pages so a change or a regression does not go unnoticed.
• Citation Integrity™ Certified is a lifecycle assessment against confidential criteria, with public verification of active status.
• The framework overview sets out the dimensions in product terms, and research and testing transparency explains how we test.

None of these promise placement in an AI answer. They assess and improve the evidence, which is the part you control.

How should you use this cluster?

Start with the definition, then work outward:

• What Citation Integrity™ is — the focused definition, its scope, and the boundary against adjacent terms.
• Why AI systems are selective — what the research says about attribution and support.
• The eight dimensions — what each one asks and what good looks like.
• Evidence in practice — provenance, primary sources, and claims that survive checking.
• Technical accessibility — retrieval, parsing and attribution.
• Measurement and honesty — what can be assessed, what can only be observed, and what cannot be known.

Each supporting article is written to be read on its own. This page exists to show how they fit together.

What are the honest limits?

Citation Integrity™ is an assessment framework, not a guarantee mechanism. AI systems change their retrieval and attribution behaviour without notice, they differ from one another, and none of them publish their selection criteria. What can be done reliably is to make published evidence genuinely dependable, verifiable and reachable — and to be honest about the difference between that and a promised outcome.

This guide reflects our framework as at September 2026 and is revised as the research and the products develop.

For the focused definition, including where citation support ends and editorial completeness begins, read What Is Citation Integrity™?.