Anchor Text Optimization and Distribution: A Link Builder's Guide

Anchor text is the clickable text inside a link. Link to you with the words "payroll software," and those two words become the anchor, one of the few places where somebody else gets to describe your page in their own words.

Optimization is deciding what those words say. Distribution is the mix you end up with once you sort every link pointing at a page by type.

You don't dial in that mix. It's a readout of how you acquire links, and it shifts whenever your acquisition mix does:

So when you dislike a profile, don't chase a number. Change the process generating it.

This guide covers what search engines pull from an anchor, the types worth knowing, whether anchors still move rankings, how to read a distribution, the internal anchors you fully control, an audit sequence, and when anchor text is the wrong problem to chase.

In HTML, the anchor is the text between the opening and closing tags, and the destination sits in the href attribute:

<a href="payroll" target="_self">https://example.com/payroll-software">payroll software</a>

Google's own standard is shorter and less demanding than the industry treats it. Good anchor text is descriptive, reasonably concise, and relevant to both the page it sits on and the page it points to. The same documentation warns against generic text like click here and against keyword stuffing, and it ties that second habit to spam policies rather than any ranking bonus.

Two mechanics matter more than the wording inside the tag.

Linked images work the same way in a different container. The alt attribute on a linked image does that link's job, so an empty alt on a linked logo hands the crawler nothing.

Naming the types is not academic. An audit stays unreadable until you can classify a whole profile in one pass, and most arguments about ratios are really arguments about where the boundaries sit.

Ten types, and most audits only ever look for one, the exact match everyone's trained to spot. The other nine are where the real signal, and the real risk, actually sits. 

The compound anchor survives scrutiny. It names the company, carries the topic, and reads like something a human editor would write, so it rarely looks manipulative in aggregate. It is also the type most under-used by teams chasing exact match.

The mismatched anchor is the one nobody classifies. A link anchored to payroll software pricing pointing to a general product homepage isn't over-optimized by any ratio, but it promises something the page doesn't answer. Community analysis of the 2024 Content Warehouse leak surfaced attribute names such as anchorMismatchDemotion and phraseAnchorSpamDays. Those are leaked field names, not confirmed mechanics, and treating them as proven behavior is a mistake. They suggest that mismatch and phrase repetition over time are worth watching in your own data.

Weakly on its own, strongly as a risk. Both halves of the industry get this wrong in opposite directions, so the split is worth stating carefully.

Anchors were once powerful enough to be absurd. In 2004, enough people linked the phrase miserable failure to a White House biography that the page ranked first for it, even though those words appeared nowhere on it. Google took three years to ship a fix.

The modern picture is far quieter. The largest public correlation work comes from Ahrefs, which analyzed 384,614 pages and found a Spearman correlation of 0.1436 between exact-match anchors and rankings. That is a weak relationship. In the same dataset, the median number-one result carried 3.7 percent exact-match anchors and 8.3 percent partial-match anchors. The pages winning competitive terms are sitting nowhere near the aggressive ratios that link vendors quote when they sell packages.

So the upside is thin. The downside is not.

Google's spam policies define "links with optimized anchor text in articles, guest posts, or press releases distributed on other sites" as link spam. The exact tactic sold as anchor optimization is written into the policy as the example of the violation.

The consequence runs deeper than most teams assume. When Google neutralizes links it has classified as spam, the ranking benefit those links were carrying is gone and does not come back. No cleanup restores it. You are not penalized back to neutral; you are returned to the position you would have held without the links, even though you paid for them.

That asymmetry is the whole argument. A perfectly tuned anchor profile buys you a correlation of 0.14. A profile that trips the classifier costs you every point of value those links ever passed. Optimizing hard against a weak signal to court a permanent loss is a bad trade, and it is the trade most ratio-driven campaigns are quietly making. A link's value is set by the authority and placement behind it, not the words on top.

Anchor text distribution is the share of your inbound links falling into each type. Every guide publishes a target table; no two tables agree, and that disagreement is the most useful fact in this section.

Put five of them side by side, and they fall apart.

Partial match runs from 1 percent to 60 percent depending on which page you happen to open. That is a sixty-fold spread on one variable. These are not variations on a single finding; they are separate agency heuristics presented with the confidence of research.

No credible fresh benchmark exists for average anchor distribution across ranking pages. That absence reflects the state of the evidence, not a gap in this guide. Any table handed to you with clean round percentages was assembled from someone's client work, so it describes their link acquisition, not yours.

A more useful model sorts your links by how much say you have over the anchor.

A healthy profile is mostly tier three by volume. When a distribution looks unnaturally tuned, the cause is almost always tier two swallowing the profile, which is a procurement problem wearing an anchor problem's clothes.

Rate of change reads worse than any static ratio. Forty exact-match anchors accumulated across four years looks like a brand people cite. The same forty acquired in six weeks looks like a campaign, because it is.

Nobody publishes a safe velocity number, and you should not trust one that appears. Track your own instead. Pull new referring domains by month with their anchor types, then watch whether the mix in the last quarter resembles the mix in the four before it. A profile that swings sharply toward one phrase in one quarter is the pattern worth investigating, regardless of what the overall percentages say.

If the published tables describe someone else's link acquisition, build yours from the pages you are trying to beat. This takes about an hour per target page.

Take the top five ranking URLs for your target keyword. Not the domains, the exact URLs. In Ahrefs, Semrush, or Majestic, open the backlink report for each and export the anchors.

You want page-level data because that is the level the ranking system works at. A domain-wide export mixes your competitor's homepage brand mentions into the numbers for the page you are racing.

Sort each competitor's anchors into the types above. This is manual, and it is fast, because anchor lists repeat themselves. Most exports collapse into a handful of distinct strings with counts attached.

Count referring domains rather than raw links. Two hundred sitewide footer links carrying the same anchor are one voice, not two hundred, and counting them as two hundred will skew your target badly.

Now you have five real distributions instead of one invented one. Look at where they overlap. If exact-match anchors sit between 2 and 9 percent across all five, that band is your working range.

Watch the outliers too. A page ranking second with 22 percent exact match tells you this SERP tolerates more than you assumed. A page ranking first with almost no exact match tells you anchors aren't what wins this keyword, which matters before you spend a quarter on them.

Homepages run heavily branded, because people link to companies by name. Deep commercial pages carry more descriptive anchors, because people link to them for what they do. Blog posts collect page titles and generic anchors, because that is how writers cite articles. Apply the range you measured to the matching page type, since reading a homepage distribution onto a product page is the most common way this analysis goes wrong.

A target distribution describes a destination and says nothing about the order you arrive in. That order is exactly where new pages get into trouble.

Take a page published in March whose first three backlinks all carry exact-match anchors. Until link four shows up, that page sits at a 100 percent exact-match profile. If links four through ten take two months to land, the page spends those two months looking like the most obviously engineered thing on the internet. The end-state plan might be reasonable. The path there is not.

Phase it instead:

Established pages can absorb a different order, since a page with 150 referring domains doesn't swing on the next three links it gets. New and newly commercial pages have no such cushion. Either way, exact-match anchors should never be the first thing a page receives.

External anchors are negotiated. You write internal ones in your CMS, and you can change them this afternoon. Most teams spend their anchor attention on the half they can only influence.

Google has been consistent that internal anchors do not carry the same over-optimization risk, because you control both ends and a search engine has never read your own wording as evidence of manipulation. Our topical authority work found that pages with at least one exact-match internal anchor pulled more than five times the search traffic of pages with none. Exact match is dangerous when you buy it and free when you write it.

Anchor dilution happens when a single destination collects a scatter of unrelated anchors from across your site. Three pages linking to your integrations page with integrations, see our tools, and connect your stack hand the crawler three different summaries of one URL.

Pick one dominant descriptive anchor per target and repeat it, with natural variation around it rather than in place of it. Competing links are the sibling problem, where two of your own pages target the same query and link to each other with the same phrase, leaving the search engine to choose between them.

Generic anchors fail an accessibility requirement, not only an SEO preference. WCAG requires that the purpose of a link be determinable from its text, or from that text plus its immediate context. Someone tabbing through a page with a screen reader hears the links stripped of the prose around them, and a page of read more links tells them nothing.

The rewrites are trivial. "To see our pricing, click here" becomes "See our pricing for teams under 50 seats." "More info here" becomes "Read the full integration setup guide." Same character count, better anchor, and one less compliance risk.

Run this in sequence. Skipping to remediation before reading the pattern is how teams disavow links that were doing nothing wrong.

Export every referring domain, along with its anchor text and destination URL, from your backlink tool, and pull the Search Console links report alongside it for a second view. Classify each anchor by type, at domain level rather than link level.

You are looking for concentration. One phrase appearing across many unrelated low-quality domains. A cluster acquired inside a short window. A destination whose anchors describe something it does not cover. Single odd links are noise, and every real profile has them. Running the same pattern-based audit on your internal links usually surfaces the sibling problem — two of your own pages targeting the same query and linking to each other with the same phrase — which is more common than most teams realize and easier to fix than an external profile problem."

One number is worth checking directly: what share of your profile do your ten most-used anchors take up? Somewhere between 30 and 50 percent is ordinary. Well above that means a handful of strings are doing all the talking, and concentration describes a footprint better than any per-type percentage.

Since Penguin has worked at page level and in real time since 2016, the domain average is the wrong unit. A site with 2,000 referring domains can post a healthy overall distribution while one money page sits at 45 percent exact match, and the site-wide report will never show it.

Rewrite internal anchors, update your own profiles and directory listings, and clean the anchors on properties you control. It costs an afternoon and moves a real share of the profile without touching a link you do not own. After that, dilute: build new links carrying branded, naked, and generic anchors at the affected page. On a page with 40 links, twenty new branded anchors move the number further than most people expect.

Reflex disavowing is the most common overreaction. If the pattern came from links Google has likely already classified, the benefit is gone whether or not you file anything, and a disavow strips value from the borderline links that were still counting. Reserve it for a manual action or a documented negative SEO campaign, and read what makes a backlink toxic first, because the honest answer is a much shorter list than most audit tools suggest.

Most sites worrying about anchor ratios have no ratio worth reading in the first place. A few common situations get mistaken for an anchor problem when the real issue lies elsewhere.

Only the last row calls for real work. The other three get mistaken for anchor problems constantly, and treating them that way just burns time on a spreadsheet instead of the page or the link surface that actually needs it.

Answer engines do not parse a link the way a crawler ranks one. They read the paragraph it sits in and build an association between a company and a topic rather than passing a weighted score along an edge.

That shifts which anchors matter. Analysis of AI visibility found branded web mentions correlating 0.664 with AI Overview presence and branded anchor text 0.527, while raw backlink counts came in at 0.218. The signals describing who you are outrank the signals counting how many people pointed at you, which reverses a decade of link-building instinct.

The practical read is unglamorous. Compound and branded anchors, sitting in sentences that state plainly what your company does, feed entity disambiguation. Exact-match anchors on generic phrases feed nothing, because the model already knows what payroll software is and is trying to work out who you are.

The industry still treats anchor text distribution as a control surface, something you tune toward a published table until the chart looks natural. That belief survives because a table is easy to sell and easy to invoice against. The truth is less comfortable: if the profile is an output, the fix isn't a spreadsheet. It's changing where your links come from, and that's slower, pricier, and impossible to show off in four weeks.

Here's what actually works:

If your distribution looks wrong, the distribution isn't the problem.

There is no universal ratio. Distributions differ by industry, page type, and how a site earns links. A working range comes from exporting the anchor profiles of the top five URLs for your target keyword and finding where they overlap. In Ahrefs data, the median number-one result carried 3.7 percent exact-match anchors.

Not by itself. Google's spam policies name links with optimized anchor text in articles, guest posts, or press releases distributed on other sites as link spam. The trigger is the placed or paid pattern, not the phrase. Exact-match anchors earned editorially, or written internally on your own site, carry no penalty risk.

Google's own standard is descriptive and reasonably concise, which in practice means roughly two to eight words. Long enough to state what the destination covers, short enough to sit inside a sentence. Single-word anchors often lose the context, and full-sentence anchors read as spam to human editors before any algorithm sees them.

Only by asking the publisher. You have no technical control over anchors on sites you do not own, which is why the three tiers of control matter. Editors will sometimes update an anchor for a factual reason, such as a rebrand or a moved URL. They rarely change one because you would prefer a different keyword.

Yes. When an image is wrapped in a link, its alt attribute does that link's job, and search engines read it that way. A linked image with an empty alt attribute passes no descriptive signal at all. Write alt text that describes the destination page, not only the image, whenever the image is a link.

Google treats nofollow as a hint rather than a directive, so a nofollowed link may still inform how a page is understood. What it will not reliably do is pass ranking value. Nofollowed anchors still count toward how your profile looks in aggregate, so heavy exact-match anchoring on nofollowed placements is not a safe workaround.

Dilution is when one destination collects many different unrelated anchors, leaving search engines without a consistent summary of what the page covers. It most often shows up in internal links written ad hoc by different authors. The fix is to pick one dominant descriptive anchor per target page and vary around it rather than replacing it.

Export the backlinks for a specific URL in Ahrefs, Semrush, or Majestic and use the anchors report, counting referring domains rather than raw links. Google Search Console's links report shows your top linking text straight from Google. Check page by page, since domain averages hide over-optimized individual pages.

Working through an anchor profile you inherited, or building one from scratch? Anchor distribution is one of the few parts of link building where the honest answer is changing how you acquire links, not a spreadsheet. Our link building team does this work daily, from backlink audits to acquisition programs built for profiles that read as earned. Tell us what your link surface looks like, and we will tell you what it is saying.