Hubs and authorities: the HITS algorithm and what an authority site is
HITS is Jon Kleinberg's 1998 link analysis method, which scores each page twice for a topic, once as an authority and once as a hub. An authority site, in everyday SEO talk, is a site widely trusted and cited in its topic. Google says it holds no single authority score for a site.
In short
- HITS, presented by Jon Kleinberg in 1998, gives each page an authority score and a hub score for one topic, where good hubs point to good authorities and good authorities are pointed to by good hubs.
- PageRank gives a page one score that does not depend on the query, while HITS works on a small set of pages gathered for one query, about 1,000 to 5,000 in Kleinberg's experiments.
- Kleinberg warned in the paper that many links confer no authority, naming navigation links and paid advertisements as examples.
- Google's John Mueller was reported by Search Engine Journal in June 2019 as saying that Google does not evaluate a site's authority, so "authority site" is a description and not a score Google holds.
- A link from a curated hub page places you beside the recognised sources in your topic, which is worth more than a link from a strong site that never writes about your subject.
HITS, short for Hyperlink-Induced Topic Search, is a link analysis algorithm that Jon Kleinberg presented in 1998. For any one topic it gives every page two scores. The authority score is high when many good hubs link to the page, and the hub score is high when the page links to many good authorities. An authority site, in everyday SEO talk, is something looser: a site that is widely trusted and cited within its topic. The two ideas share a history, but only the first is a defined measurement, and nothing Google has published describes a hub score or an authority score of Kleinberg’s kind. What a link builder can take from HITS is the reasoning behind it, which is that authority belongs to a topic and that the pages which curate links play a different part from the pages which receive them.
The problem Kleinberg set out to solve
Kleinberg’s paper, “Authoritative Sources in a Hyperlinked Environment”, starts from what he calls the abundance problem. A broad search such as “java” or “censorship” matches far more pages than a person can read, so a search engine needs a way to pick out the few that are most definitive. Text alone cannot do that. His example is the query “Harvard”: more than a million pages used the word at the time, and the university’s own homepage was not the one that used it most often or most prominently. A second example is sharper still. The homepages of the main search engines of the day did not contain the phrase “search engines”, just as the homepages of car makers have no reason to contain “automobile manufacturers”. The best answer to a broad query often does not describe itself in the words people search with.
Links offered a way round this. Kleinberg wrote that hyperlinks encode a considerable amount of latent human judgment, because the creator of a page who links to another page has in some measure conferred authority on it. He was careful about the weak points of that claim. Many links, the paper says, are created for reasons that have nothing to do with conferring authority, and it names links made for navigation and links that are paid advertisements. That caveat sat in the founding paper of the field, long before anyone ran a link marketplace. He also showed why counting links is not enough. If you take every page that contains the query and rank by the number of inbound links, a universally popular page wins every topic it happens to mention, and the true authorities that never use the phrase are left out altogether.
How the algorithm works
The method runs when someone searches, on a small sample of the web built for that one query. It begins with a root set: about 200 of the top results for the query from a text search engine, which in Kleinberg’s experiments was AltaVista. The root set is then widened into a base set by adding the pages that the root pages link to and a limited number of pages that link to them, capped at 50 per root page in his tests. The reasoning is that a strong authority on the topic may not appear in the text results at all, yet it is very likely to be linked from at least one page that does. The base sets in his experiments held between 1,000 and 5,000 pages. Links between pages on the same domain are discarded before scoring, on the grounds that they mostly exist for navigation.
Scoring rests on what the paper calls a mutually reinforcing relationship. In Kleinberg’s words, a good hub is a page that points to many good authorities, and a good authority is a page that is pointed to by many good hubs. The definition is circular, so the algorithm resolves it by repetition. Every page starts with the same two weights. Each page’s authority weight is then replaced with the sum of the hub weights of the pages linking to it, each page’s hub weight is replaced with the sum of the authority weights of the pages it links to, and both sets of weights are scaled back to a fixed total. After enough rounds the numbers settle, and the pages with the highest authority weights are returned as the answer. A popular page that has nothing to do with the topic tends to drop out, because the hubs for the topic do not point to it together.
How HITS differs from PageRank
Both methods appeared in the same period and both treat a link as a vote, which is why they are often confused. The first difference is when the score is worked out and what it depends on. PageRank is calculated across the whole link graph in advance and gives a page one number that is the same whatever the search. HITS gives a page a score only in relation to a topic, so the same page can be a strong authority for one query and absent from the results for another. Kleinberg drew this contrast himself: PageRank was applied to an index of 24 million pages and then used to order later text searches, while his method first ran a text search and then scored a small subgraph built from the results.
The second difference is the hub. Kleinberg described PageRank as a model in which authority passes directly from authorities to other authorities, with no hub pages in between. HITS assumes that the leading sources on a subject often do not link to one another, and the paper gives the homepages of competing search engines as its example, and that what ties them together is the set of pages that link to all of them. That makes the hub a thing worth scoring in its own right. It also exposes a weakness that follows from the definitions. An authority score depends on links other people give you, but a hub score depends on links you publish yourself, which means anyone can write a page that points to every recognised source on a topic. A search engine that leaned on raw hub scores would have to deal with that, and it is one plausible reason, though only an inference, why the hub side of the model is not something you see reported anywhere.
What Google says about authority
Google’s guide to its ranking systems lists “link analysis systems and PageRank” and says only that it has various systems that understand how pages link to each other. It does not name HITS, and Google has not said that it uses hub and authority scores. The same guide describes reliable information systems that help surface more authoritative pages, and Google’s public explanation of how ranking works says that one of the factors used to judge quality is whether other prominent websites link or refer to the content. So the word “authoritative” is part of Google’s own vocabulary. What Google has declined to confirm is that it boils this down to one figure per site.
Search Engine Journal reported in June 2019 that John Mueller, asked in a webmaster hangout, said Google does not evaluate a site’s authority and that this was not something it would be applying. Google has denied using Domain Authority, and the scores sold by tool vendors are their own estimates, as the lesson on Domain Authority and Domain Rating explains. There is evidence of a different kind for trusted starting points, which is covered in the lesson on TrustRank and seed sites and is not repeated here. The fair summary is that Google talks about authoritative pages and sources, that links from prominent sites are a stated input, and that the mechanics are not public.
What “authority site” means in practice
Since no outside party can read a score, “authority site” is a judgment, and the useful way to make that judgment is the one HITS suggests: ask who points to the site and for which subject. A cardiology journal is an authority on heart disease and has no standing on mortgage rates. A site earns the description in a topic when the curated pages in that topic, the reading lists, professional bodies, trade press and reference works, cite it without being asked. You can check this with a backlink tool by looking at the linking pages and not at the totals. If a good share of the referring pages are themselves about the subject and link out to the other sources you would expect to see, the site is an authority in the sense that matters. This is a working method and not a measurement, and two careful people can reach different answers for the same site.
The limits of the label show up fastest in sales material. Vendors describe almost any site with a high metric as an authority site, and a metric can be inflated or faked without a single respected source ever linking to the domain. A large general news site is an authority on the news and may have no particular standing in your niche, which is the old argument about relevance and authority in another form. Authority also sits at page level as much as site level. A forgotten page deep inside a respected domain, with no internal links and no readers, borrows the name of the site and little else. When a seller uses the phrase, the sensible reply is to ask which topic the site is an authority on and who says so, then run the checks in how to vet a publisher.
Hubs and authorities in link building
Hub pages are easy to recognise once you look for them. Resource pages, curated reading lists, association directories of members, “useful links” pages kept by libraries and universities, and comparison articles that review the main options in a category all exist to point readers at the best sources. A link from one of these does two jobs. It is a link in the ordinary sense, and it puts your page on the same list as the established names in your topic, which is the pattern described in the lesson on co-citation and co-occurrence. Picture a small site that publishes detailed guides to keeping bees. A link from a regional beekeeping association’s resources page, sitting beside the national association and two university extension departments, says more about what the site is and how far it is trusted than a link from a large lifestyle magazine’s gift guide.
That is why several standard tactics are aimed at hubs. Broken link building works largely on resource pages, because a curator who keeps a list has a reason to replace a dead entry. Getting included in best-of lists is hub link building under another name. A link gap analysis often surfaces hubs without calling them that: a page that links to three of your competitors and not to you is, nearly by definition, a hub for your topic. The quality test is whether the page is curated. A hub that lists whoever pays or whoever submits a form has the shape of a hub and none of the judgment, and Google’s link spam policy names low-quality directory links as an example of link spam.
Links from the authorities themselves are scarcer, and for the reason Kleinberg gave: the leading sources in a field seldom link to each other, and still less to newcomers. When they do link, it is usually a citation of something specific, such as a dataset, a tool or a piece of original research, which is the case for building linkable assets before asking anyone for anything. It is also worth being a good hub yourself. Linking out to the best sources on your subject is what a useful page does, and Google’s link guidance says that linking to other sites can help establish trustworthiness. No evidence shows that outbound links raise your own rankings directly, so treat it as good publishing and not as a ranking tactic. Terms used here, including HITS algorithm and authority site, are defined in the glossary.
Common questions
What is the HITS algorithm?
HITS stands for Hyperlink-Induced Topic Search. It is a link analysis method by Jon Kleinberg that scores pages on one topic as authorities, which many good hubs link to, and as hubs, which link to many good authorities.
How is HITS different from PageRank?
PageRank gives each page a single score worked out across the whole web before anyone searches. HITS gives each page two scores and works them out for one topic, on a small set of pages gathered for that query.
Does Google use HITS?
Google has not said that it does. Its ranking systems guide names PageRank among its link analysis systems and says only that it has various systems that understand how pages link to each other.
What is an authority site?
It is a site that is widely trusted and cited within its topic. The phrase is a description used by SEOs and link sellers. Google staff have said Google does not give a site one authority score.
Are links from hub pages worth getting?
A curated hub that readers use, such as a university reading list or an industry resource page, is a sound link. A page that lists anyone who pays or asks is a hub in shape only and adds little.

