How search engines evaluate links

Not every link counts the same, and many count for nothing. Search engines weigh the authority of the linking site, how relevant it is, the anchor text and where the link sits on the page.

In short

  • Three things decide what a link is worth: the authority of the source, its relevance to your topic, and the anchor text.
  • Since 2019 Google treats nofollow, sponsored and ugc as hints, not commands. It may still use those links.
  • Google now mostly ignores links it judges to be spam instead of punishing them, so the common cost of a bad link is wasted money.
  • No primary source supports a safe anchor text ratio or a safe number of links per month. Any figure you see is an industry rule of thumb.

A search engine has to decide, for every link it finds, whether to count it and how much. The exact weights are not public. What is known comes from Google’s own documentation, patents, the internal documents leaked in May 2024 and exhibits from the US antitrust trial. Put together, those sources point to a short list of things a search engine looks at: how much authority the linking site has, how relevant it is to your topic, what the anchor text says, and where on the page the link sits. Each of the four is explained below in the order a beginner is likely to meet them, along with what happens to links that fail the test and which popular rules have no source behind them at all.

Authority: who is linking

PageRank still exists. It is Google’s system for scoring pages by the links that point to them, and it passes value from page to page, so a link from a site that itself has strong links is worth more than one from a site nobody links to. That chain effect is the heart of the idea. A recommendation is only as good as the standing of whoever gives it, and a search engine measures that standing by looking at who recommended the recommender. An exhibit from the antitrust trial describes the modern version as a signal relating to distance from a known good source, which means a link is worth more the fewer steps separate the linking page from sites Google already trusts.

The 2024 leak also lists a site-level authority field and shows that Google stores links in quality tiers, which suggests links from pages in its top index tier are treated as more valuable. An index tier is a quality or freshness layer of Google’s index where a page is stored, so the suggestion is that a link counts for more when the page it sits on is kept in the best layer. The leak documents fields, not weights, so how much each one matters is not confirmed. That limit is worth holding on to. The leak shows that Google has a place to store a piece of information, and it cannot show how heavily that information counts in a ranking, or whether it counts at all. For practical purposes the lesson is modest. A link from a page that is itself well linked, well visited and clearly valued by Google is the kind most likely to carry weight, and a link from a page that nothing points to is the kind most likely to carry none.

Relevance and anchor text

A link from a site about your topic carries more meaning than one from an unrelated site. Say a shop sells running shoes. A link from a marathon training blog fits the subject, and a reader of that blog might well follow it, whereas a link from a site about mortgage rates has no obvious reason to exist. The leaked documents include an anchor mismatch demotion, which points to Google checking whether the anchor text fits the page it points to. Anchor text is the clickable wording of a link, and a mismatch would be a link whose words promise one subject while the page behind it covers another. So relevance appears to be checked at two levels: whether the linking site belongs near your topic, and whether the words of the individual link describe the page honestly.

The words of the link tell the engine what the target page is about. That makes anchors powerful and risky. They are powerful because a link reading “lightweight trail running shoes” is a far clearer description than a link reading “this website”, and they are risky because the clear version is exactly what someone trying to manipulate rankings would write, over and over. The Penguin update of 2012 was built to catch profiles stuffed with keyword anchors, and the leak includes fields for spikes of spammy anchor phrases. A natural set of links is untidy. People who link of their own accord tend to use a brand name, a bare web address or a vague phrase, and only a few happen to use the exact words a site owner would have picked. When a large share of a site’s links all carry the same commercial phrase, the pattern itself becomes the evidence that the links were arranged.

Placement and attributes

A link in the main body of an article, where a reader would click it, is treated as stronger than one in a footer or sidebar. This idea comes from Google’s “reasonable surfer” patent, and the name explains the reasoning: a link is weighed by how likely a reasonable person reading the page would be to click it. A link inside a sentence that discusses the subject is likely to be clicked. A link in a block of small print at the bottom of every page is not, and treating the two alike would reward exactly the kind of placement that is easiest to buy in bulk. For anyone arranging a link, this means asking where on the page it will appear before agreeing to it, since a mention in the body of a relevant article and a line in a site-wide footer are different things even when they come from the same website.

Links can also carry three labels, which are short attributes in the page’s code that visitors never see. The first, nofollow, says the site does not vouch for the target. The second, sponsored, says the link was paid for. The third, ugc, stands for user-generated content and says the link was added by a user, as in comments. Since 2019 Google reads all three as hints. It decides for itself whether to use the link. Many people still talk as though a labelled link is always set aside. Under the hint model a labelled link may count or may not, and nobody outside Google can tell which. That is one reason not to dismiss a link from a strong news site because it carries a nofollow label, as many such sites only give labelled links, and those still bring visibility and brand mentions.

In most cases, nothing visible. Google’s SpamBrain system is designed to neutralise links it considers manipulative, which means they are ignored. To neutralise a link is to treat it as if it were not there, so the site that received it neither gains nor loses. A manual action, where a human reviewer penalises a site, still exists for clear patterns. It is the more serious outcome, because a penalty takes something away where neutralising only withholds a gain. Google has also said that once link value is neutralised it cannot be regained, which means there is no later repair that brings a discounted link back to life.

The practical point follows from that: the usual price of a poor link is the money spent on it. This changes how a beginner should think about risk. The older fear was that one bad link could sink a site, and under a system that mostly ignores spam that fear is overstated. The quieter danger is paying month after month for links that are doing nothing, with no warning that anything is wrong, because an ignored link looks identical to a counted one in every tool. The only signs are indirect, such as rankings that fail to move despite a growing list of links. Judging a link before buying or requesting it, on the grounds of authority, relevance and placement described above, is therefore worth more than any attempt to clean up afterwards.

What the evidence does not support

Two popular ideas have no primary source behind them. One is a safe ratio of anchor types. The other is a safe number of new links per month. You will meet both stated with great confidence, often as precise figures, and it is worth knowing that no published Google source gives either. The common advice to keep exact-match anchors, meaning anchors that repeat the target keyword word for word, at 1 to 5 percent is a heuristic drawn from practitioner experience. That does not make it useless, but it does mean it is a rule of thumb and should be labelled as one whenever it is repeated.

Both depend on the market, so the working method is to study the profiles of the sites that already rank for your terms. The reasoning is that what looks normal differs from one field to another. A news publisher may gain hundreds of links in a day when a story travels, while a local accountant gaining links at that pace would look strange. The same goes for anchors, where a brand whose name happens to contain its main keyword will have a profile that would look manipulated on anyone else. The sites that already rank show what Google currently accepts in your market, which is better evidence than a universal number. The same caution applies to the authority scores that tools publish. Domain Rating and Domain Authority are third-party estimates, not Google metrics, and they can be inflated. Real organic traffic and topical relevance are harder to fake and say more about a site.

Other search engines

Most of this lesson is about Google because most of the public evidence is about Google, but the other engines are not identical and the differences matter if you work in their markets. Bing uses links in a similar way and removed its disavow tool in October 2023. A disavow tool lets a site owner ask a search engine to disregard particular links pointing at the site, so removing it fits the same direction Google has taken, in which the engine decides for itself what to ignore.

Yandex’s leaked code from 2023 shows factors for link age and penalties for sudden bursts. That is a rare case where the idea of pace has something concrete behind it, though it comes from a different engine and cannot simply be carried across to Google. Naver barely uses links at all, so link building has little to offer there. The general point for a beginner is that “how search engines evaluate links” has no single answer. The principles of authority, relevance, wording and placement travel reasonably well, but how much weight links carry against everything else is a decision each engine makes for itself.

Common questions

What is a good Domain Rating for a link?

Domain Rating and Domain Authority are third-party estimates, not Google metrics, and they can be inflated. Real organic traffic and topical relevance are harder to fake and say more about a site.

Do nofollow links help?

They can. Google has treated nofollow as a hint since 2019, so it may count a nofollow link. Many of the strongest news sites only give nofollow links, and those still bring visibility and brand mentions.

Is there a safe percentage of exact-match anchors?

No published Google source gives one. The common advice of 1 to 5 percent is a heuristic drawn from practitioner experience. Compare your profile with the sites that rank in your market instead.

Vendors to look at

  • BazoomEditor's pick

    Sponsored content and link marketplace, managed service

  • MotherlinkEditor's pick

    Backlink services, guest posts, niche edits, full SEO