Domain authority

Domain authority is a third-party predictive score that estimates how likely a whole domain is to rank, computed by an SEO vendor from its own crawl of the link graph. It is a vendor's model of a search engine, not an output of one. Several vendors publish their own version under their own name, and the scores are not interchangeable.

That distinction is the entire practical content of the term. A domain authority number is evidence about a vendor's index. It is not a number Google or any assistant computes, publishes, or consumes.

What search engines actually document

Google publishes a list of its ranking systems. It names BERT, crisis information systems, deduplication systems, the exact match domain system, freshness systems, link analysis systems and PageRank, local news systems, MUM, neural matching, original content systems, removal-based demotion systems, the passage ranking system, RankBrain, reliable information systems, the reviews system, the site diversity system, and spam detection systems. No site-wide authority score is among them.

What it says about links is deliberately unquantified: "We have various systems that understand how pages link to each other as a way to determine what pages are about and which might be most helpful in response to a query."

The underlying idea is old and public. The PageRank patent — US 6,285,999 B1, "Method for node ranking in a linked database", inventor Lawrence Page, filed January 9, 1998, priority January 10, 1997, granted September 4, 2001 — describes ranking a document by the ranks of the documents that cite it, computed iteratively as a probability distribution over a random surfer, with a damping factor for the chance of jumping to an arbitrary page. Note what that describes: a score per document. The unit was never the domain.

Two further documented systems cut directly against domain-level thinking. The site diversity system "works so that we generally won't show more than two web page listings from the same site in our top results, so that no single site tends to dominate all the top results" — a ceiling on how far one domain's standing can carry it in a single result set. And the exact match domain system exists because the domain string itself was being gamed: Google's ranking systems "consider the words in domain names as one of many factors", but the system "works to ensure we don't give too much credit for content hosted under domains designed to exactly match particular queries".

Why the domain is the wrong unit for citation

An assistant that answers with links fetches documents while answering and writes from what it retrieved. The selection happens over passages — see RAG — so what competes is a section of one page, not an average across a hostname.

This changes what "authority" has to mean operationally. A domain-level score cannot tell you which of your pages was retrieved, whether the retrieved passage supported the claim it was attached to, or whether the fetch succeeded at all. Those are per-page facts, and the first of them is an access question before it is a quality question: a page an agent may not fetch is not a candidate regardless of the domain it sits on. See crawling and indexing.

The property that does travel across a site is more specific than a score: consistent, checkable claims, stated with their qualifiers, on pages that are cheap to fetch and unambiguous about who published them. That is the substance behind source citation and the practical subject of GEO.

Failure modes

• Reporting a vendor score as a search engine's opinion. It is a prediction produced by a different index than the one serving the result.
• Comparing scores across vendors. Different crawls, different models, different scales; the comparison has no defined meaning.
• Chasing the score instead of the input. Because these scores are trained to predict rankings, they follow visibility at least as much as they cause it.
• Assuming domain strength transfers to every page. The site diversity system alone bounds how much of one result set a single site occupies.
• Buying links to move a number. The number may move in a vendor's index while the search engine treats the pattern as spam.
• Extending the concept to assistants. No assistant documents a domain-level authority input, and retrieval operates on passages.

How to use the term honestly

Treat a domain authority score as a rough, directional comparator between sites inside one vendor's dataset, over time, and nothing more — useful for triaging a link prospect list, useless as a target. When someone reports it as a reason a page will or will not be cited, the question to ask is which page, which passage, and whether the agent could fetch it.