Anchor text

By Bill Hartzer · Hartzer Consulting · 25+ years in search, domains and expert witness work

What anchor text is and why it matters

Anchor text is the visible, clickable text of a link. In the sentence "see the tactic reference for detail", the words tactic reference are the anchor. When the link is an image, the image's alt attribute serves the same function.

It matters because it is a description of the destination written by somebody other than the destination's owner. That makes it a genuinely independent signal about what a page is about, and search engines have used it since the beginning — the original PageRank work noted that anchor text often describes a target page more accurately than the page describes itself. It is also why a page can rank for a phrase it does not contain: enough people linked to it using that phrase.

And because it is valuable, it became the first thing manipulated. If pointing a hundred links at a page with the anchor "cheap car insurance" helps that page rank for cheap car insurance, then people will point a hundred links at it. That arms race is most of what Penguin was built to address, and it is why anchor text is now the fastest way to identify a manufactured profile from the outside.

The six types

• Brand anchors. Your company or site name — Hartzer Consulting, billhartzer.com. Overwhelmingly the most common anchor in any real profile, because that is how people refer to organizations in prose.
• Exact-match anchors. The precise phrase you want to rank for — link building course. Powerful, and the type that gets profiles into trouble.
• Partial-match anchors. The target phrase inside a longer natural expression — this free link building course from Hartzer. Common in genuine editorial writing.
• Naked URLs. The address itself as the text — https://example.com/page. Extremely common in forums, comments, citations, footnotes and anywhere someone pasted a link without formatting it.
• Generic anchors. Click here, read more, this article, source. Universal in real writing, useless as a description, and a signal of nothing except that a human wrote the sentence.
• Image anchors. A linked image, where the alt attribute does the descriptive work. Frequently the alt is empty, which produces a link with no anchor text at all.

A seventh category is worth naming because it dominates real data and no article ever mentions it: empty anchor text. Linked images with no alt, linked whitespace, buttons, icons and markup artifacts. It is enormous, and I will come to the number in a moment.

What a natural distribution actually looks like

Almost every article about anchor text distribution presents a target percentage table — 50% brand, 20% naked URL, 5% exact match, and so on. I have never seen anyone show the data those figures came from. So here is a real profile, measured: 1,301,839 links from 22,260 referring domains, accumulated over twenty years, on a site where no link was ever bought.

Three findings.

The top anchor was the brand name, at 272,856 links. Not a keyword. Not a service phrase. The name of the site. By a very large margin.

There was no commercial exact-match anchor anywhere near the top of the list. None. On a site that has published continuously for two decades in a commercially competitive field, the phrases a link buyer would target simply do not appear in volume. This is the single most useful calibration point I can give you, and it is not what the target-percentage tables imply.

37.1% of referring domains gave empty anchor text. Over a third of all linking domains contributed no anchor text at all — linked logos, linked images with no alt, icons, widget artifacts. If your model of a natural profile does not have a very large empty bucket in it, your model is wrong.

Add to that the concentration effect: one domain sent 219,159 links from a single sitewide placement, and the top hundred domains accounted for 84.3% of all links. A sitewide link repeats one anchor tens of thousands of times, which distorts any link-level percentage. This is another reason to measure anchor distribution at the referring domain level rather than the link level.

Over-optimization, and what it looks like

Over-optimization means an anchor distribution that could not have arisen from people writing about you naturally. It is not a threshold you cross at a specific percentage; it is a shape.

The signatures I look for when auditing a profile:

• Commercial exact-match phrases in the top anchors. On a natural profile the top of the list is the brand, the domain, and generic connectives. If best plumber in Dallas is the third most common anchor across distinct domains, somebody chose that.
• Uniformity across unrelated domains. Thirty different sites in different countries independently choosing the same six-word phrase does not happen.
• Missing empty and generic buckets. Built links have deliberate anchors, because somebody typed them. Nobody builds a link with an empty alt attribute on purpose. A profile with almost no empty anchors has been assembled.
• Anchor and page mismatch. Exact-match anchors pointing at deep commercial pages while the homepage receives only brand anchors is a distribution that arises from a strategy, not from readers.

The remedy, when you find this on a site you now own, is not to build a pile of brand anchors to dilute the ratio. That is treating the symptom, and the ratio manipulation is itself detectable. The remedy is to stop producing the pattern and, where a specific set of links is both manipulative and identifiable, deal with those links directly.

Measuring your own distribution

If you want to know what your anchor profile looks like, measure it properly, because the default export will mislead you.

Pull anchor text at the referring domain level, not the link level. A single sitewide placement repeats one anchor across every page of a site, and on the reference profile that effect was extreme — one domain contributed 219,159 links, all carrying whatever anchor sat in that template. Counted per link, that one decision would dominate the entire distribution. Counted per domain, it is one opinion among 22,260.

Then group the anchors into the types above and look at the shape rather than the percentages. The questions worth asking are: is the brand at the top? Is there a substantial empty and generic bucket? Do any commercial phrases appear across many unrelated domains? Does anything in the top twenty look like it was requested rather than written?

Do this before you start any campaign, so you have a baseline, and repeat it quarterly. The value is not the numbers themselves — it is noticing a change in shape early, particularly on a site where someone else is also buying links you did not authorize.

The mistake of controlling anchors at all

Here is the part that gets people into trouble, and it is a mindset rather than a technique.

The moment you decide what the anchor text should say, you have converted an independent signal into a self-authored one. Every link where you specified the anchor is a link that carries your intent rather than an editor's judgment. Do it once and it is invisible. Do it two hundred times and you have written a pattern across two hundred sites that reads, to anything looking for patterns, exactly like what it is.

Genuinely earned links come with anchors you did not choose, and the anchors are usually worse than what you would have picked — your brand name, the article title, the word "here", the bare URL. That variety is not a cost of earning links honestly. It is the evidence that they were earned.

Where anchor text is entirely yours to control, and where I would spend the effort instead, is internally. Your own site's internal links let you describe your pages precisely and consistently, at any scale you like, with no risk and no permission needed. It is the highest-leverage anchor text work available and almost nobody does it deliberately.

Lesson 1.4 of the course defines the vocabulary here, and Module 8 covers reading a profile for unnatural patterns including anchor distribution.