Anchor Text Optimization: How to Audit, Balance and Fix Your Anchors

Quick quiz

Which way of finding customers online fits you best?

Question 1 of 425%

How do your customers find businesses like yours?

They search on Google or ask ChatGPTThey discover them on social mediaSomeone recommends them

This guide goes further than the definition. If you only need the basics (what an anchor is, the main types), start with our anchor text glossary entry. Here you get the decision rules, the audit method with copy-ready formulas, the four ways to set a target distribution, and the mistakes that cost rankings.

What anchor text optimization really controls

The anchor text is the visible, clickable text inside an <a href> element. Search engines read it as a short label for the destination page. When dozens of pages describe a URL with similar words, that label becomes a strong hint about the page's topic.

Optimization is not "put the keyword in the link". It is managing three things at once: what the anchor says, how the anchors pointing to one page add up, and whether the pattern looks like it came from real editors.

The three signals carried by a link

• The anchor itself: the words between <a> and </a>. For an image link, the image's alt attribute plays this role.
• The surrounding text: the sentence and paragraph around the link give context, especially when the anchor is short or generic.
• The link profile: the sum of all anchors pointing to a URL. One exact-match anchor is normal. Sixty percent exact-match anchors on a commercial page is a pattern.

What Google says, word for word

Google's documentation on making links crawlable (Google Search Central) gives the official rule: good anchor text is "descriptive, reasonably concise, and relevant to the page that it's on and to the page it links to". It also suggests a simple test: read only the anchor text, out of context, and check that it still makes sense. The same page flags "click here" and "read more" as too generic, and warns against cramming every keyword into the link.

The other side of the rule sits in Google's spam policies, which list as link spam "links with optimized anchor text in articles, guest posts, or press releases distributed on other sites". Read both together: descriptive anchors are encouraged, manufactured keyword anchors on links you placed yourself are not.

The 10 anchor types and when each one fits

Every audit starts with a classification. The usual lists stop at five or six types. In practice you need ten, because "partial match" and "branded" hide very different cases. Take an example target page: sorank.com/seo-tool/backlink-analysis, a free backlink checker, with the target keyword "backlink checker".

Type	Example for the target page	Where it is natural	Risk if overused
Exact match	backlink checker	Internal links, a few editorial backlinks	High on backlinks, low on internal links
Phrase match	free backlink checker tool	Internal links, reviews, tool lists	Medium
Partial match	check the backlinks of any domain	Everywhere, the most natural editorial form	Low
Topical (entity)	referring domains and link profile	Contextual mentions in related articles	Very low
Branded	Sorank	Press, partners, directories, reviews	None
Brand + keyword	Sorank's backlink checker	Reviews, comparisons, guest content	Low
Naked URL	sorank.com/seo-tool/backlink-analysis	Forums, citations, resource lists	None
Generic	this tool, here, read more	Occasional backlinks	None for Google, poor for users
Long-tail sentence	a free tool that lists every link pointing to your site	Editorial articles, documentation	Very low
Image or empty	alt="Backlink checker results"	Logos, buttons, cards	None, but no text signal if alt is empty

Internal anchors and backlink anchors are not the same game

The single most useful distinction in this topic: you write 100% of your internal anchors, but you only influence a small share of your backlink anchors. Google treats links you control on your own site as navigation. It treats keyword anchors on links you placed on other sites as a possible manipulation. So the rules are opposite:

• Internal links: be precise. Descriptive and exact-match anchors are fine, as long as they vary and each anchor points to one page only.
• Backlinks: be relaxed. Branded, URL and natural sentence anchors should dominate, because that is what real editors write.

For the full method on the internal side (priority pages, link depth, hubs), see our internal linking strategy guide. And do not confuse anchor text with anchor links, which are jump links to a section of the same page.

What the data says about anchors and rankings

Most "ideal ratio" tables you will find online come from no study at all. Two public datasets are worth knowing, one on backlinks and one on internal links. Both are correlational, which means they show what ranking pages have in common, not proof of cause.

Backlink anchors: a weak but real correlation

The anchor text study by Joshua Hardwick at Ahrefs (updated April 3, 2020) looked at the top 20 results for 19840 keywords, or 384614 pages. It chose keywords where the top 10 pages had similar URL Rating, to isolate the anchor variable. Result: exact-match anchors had a "relatively weak correlation" with rankings, partial and phrase matches a "very weak" one, and random anchors "effectively no correlation". The author also notes that the exact-match median across positions is zero: many top pages have no exact-match backlinks at all.

How to read it: relevant anchors help a little, but a page can rank without a single exact-match backlink. That is the data behind the advice "do not chase keyword anchors on backlinks".

Internal anchors: variety wins

The internal link side is more striking. Cyrus Shepard's 23 million internal links study (Zyppy, updated February 23, 2026) matched 23 million internal links on 1800 sites with Google Search Console clicks. The strongest finding: the more anchor text variations a URL receives from internal links, the more Google clicks it gets. Pages with at least one exact-match internal anchor had "at least five times more traffic" than pages without, and pages with a URL anchor saw "almost 50% more traffic".

The same study saw traffic start to decline beyond roughly 45 to 50 internal links per URL, which the author attributes mostly to sitewide navigation links pointing to weaker pages. Practical takeaway: more varied, contextual internal anchors on a page you care about, not more menu links.

The algorithm that changed the rules

The 2016 change matters for anchor optimization. In Gary Illyes' announcement of real-time Penguin (Google Search Central, September 23, 2016), Google says Penguin "now devalues spam by adjusting ranking based on spam signals, rather than affecting ranking of the whole site". In plain words: a manipulative anchor pattern today mostly means your links stop counting. Manual actions still exist for blatant schemes, but the common outcome is a page that stops progressing despite new links.

Four ways to set a target anchor distribution

There is no official ratio. There are four methods to decide what "normal" looks like for a given page, each with its own logic. Pick one deliberately instead of copying a percentage table.

1. Fixed ratio grid

Principle: you apply a preset split, for example a majority of branded and URL anchors, a minority of partial match and a small slice of exact match. When to use it: a new site with almost no links, where you need a simple rule for every campaign. Advantages: easy to brief to a team or a provider. Limits: the numbers come from habit, not data, and they ignore your niche. A local plumber and a SaaS do not get the same anchors naturally. Example: a brief that says "no more than one keyword anchor out of ten new backlinks on any commercial page".

2. Competitor benchmark per page

Principle: for one target keyword, you export the anchors of the pages ranking in the top 10, classify them, and take the median share of each type as your guardrail. When to use it: competitive keywords where you have budget for analysis. Advantages: grounded in what Google already rewards for that query. Limits: takes time, and competitors may themselves be over-optimized and about to drop. Use the median, never the most aggressive competitor. Example: if 8 of 10 ranking pages have under 5% exact-match anchors, a page at 30% is the outlier to fix.

3. Editorial-first (no target)

Principle: you never request an anchor. You earn links with content and digital PR and accept whatever editors write. When to use it: brands with strong content and PR, or after a period of over-optimization. Advantages: the most natural profile possible, zero risk. Limits: you lose a lever. Many editors default to your brand name or "this study", so the target page gets few topical signals from backlinks and must rely on internal links. Example: a data study picked up by 40 media sites, almost all linking with the brand or the study's title.

4. Topical and entity-based anchors

Principle: instead of one keyword, you build a vocabulary of the topic (synonyms, sub-topics, entities, questions) and spread it across anchors. When to use it: internal linking at scale, and the anchors you suggest to partners. Advantages: matches what the Zyppy data rewards (variety), covers long-tail queries, reads naturally. Limits: needs a real topic map, otherwise you drift into vague anchors. Example: for a backlink checker page, anchors like "see who links to a site", "referring domains report", "audit your link profile", "free backlink checker".

Method	Best for	Effort	Risk of over-optimization	Topical signal
Fixed ratio grid	New sites, simple briefs	Low	Medium (ratios can be wrong for the niche)	Medium
Competitor benchmark	Competitive money pages	High	Low if you use the median	Medium to high
Editorial-first	Strong brands, recovery periods	Low on anchors, high on content	Very low	Low from backlinks
Topical / entity anchors	Internal links, partner content	Medium	Low	High

Our recommendation: combine 4 for internal links, 3 or 4 for backlinks, and keep 2 as a check on your most valuable pages.

Step-by-step: audit and optimize your anchors

This is the method we use on a site, from export to fixes. Plan two to four hours for a site of a few hundred pages.

Step 1: export your backlink anchors

Google Search Console gives a free starting point. In the Links report, the "Top linking text" table shows the link text used in backlinks to your site, grouped and sorted by count. Links without text, such as image links, appear as "(empty)". Tables stop at 1000 rows, and the Export button downloads up to 100000 rows of link data.

Search Console aggregates anchors at site level, so you also need a URL-level export from a backlink tool. Our free backlink checker lists the links pointing to any domain with their anchors.

Step 2: crawl your internal anchors

Run a crawler on the site and export every internal link with three columns: source URL, target URL, anchor (or alt text for images). The free version of Screaming Frog crawls up to 500 URLs and shows the aggregated and per-link anchor text of all links.

Step 3: classify each anchor

Do it per target page, because "exact match" depends on that page's keyword. In Google Sheets, with the anchor in column C, the brand in cell $H$1 and the target keyword in $H$2, this formula labels each row:

=IFS(
  C2="", "Empty / image",
  REGEXMATCH(LOWER(C2), "^https?://|^www\.|\.(com|fr|io|net|org)"), "Naked URL",
  REGEXMATCH(LOWER(C2), "^(click here|here|read more|learn more|this|website|link|source)$"), "Generic",
  AND(REGEXMATCH(LOWER(C2), LOWER($H$1)), REGEXMATCH(LOWER(C2), LOWER($H$2))), "Brand + keyword",
  REGEXMATCH(LOWER(C2), LOWER($H$1)), "Branded",
  LOWER(TRIM(C2))=LOWER($H$2), "Exact match",
  REGEXMATCH(LOWER(C2), LOWER($H$2)), "Phrase match",
  TRUE, "Partial / other"
)

Then review the "Partial / other" rows by hand: that is where you separate true partial matches (some keyword words) from topical anchors and long sentences. For large exports, the same logic in Python:

import re
import pandas as pd

BRAND = "sorank"
KEYWORD = "backlink checker"
GENERIC = {"click here", "here", "read more", "learn more", "this", "website", "link", "source"}

def classify(anchor: str) -> str:
    a = (anchor or "").strip().lower()
    if not a:
        return "Empty / image"
    if re.match(r"^(https?://|www\.)", a) or re.search(r"\.(com|fr|io|net|org)\b", a):
        return "Naked URL"
    if a in GENERIC:
        return "Generic"
    has_brand, has_kw = BRAND in a, KEYWORD in a
    if has_brand and has_kw:
        return "Brand + keyword"
    if has_brand:
        return "Branded"
    if a == KEYWORD:
        return "Exact match"
    if has_kw:
        return "Phrase match"
    kw_words = set(KEYWORD.split())
    if kw_words & set(a.split()):
        return "Partial match"
    return "Topical / other"

df = pd.read_csv("anchors.csv")  # columns: source, target, anchor
df["type"] = df["anchor"].fillna("").map(classify)
split = (df.groupby(["target", "type"]).size()
           .groupby(level=0).apply(lambda s: (s / s.sum() * 100).round(1)))
print(split)

Step 4: compute the split per URL

A domain-wide distribution is almost useless: the homepage collects branded anchors and drowns the signal of the commercial pages. Build one pivot table per priority URL, with the count and the share of each type. Flag any page where one non-branded anchor, word for word, represents a large share of all backlinks to that page. That repetition, more than the keyword itself, is what looks engineered.

Step 5: benchmark against the pages that rank

For your three to five most valuable pages, export the anchors of the top 10 results for the target keyword and run the same classification. Note the median share of exact and phrase match. If your page is far above it, you have found your problem. If it is far below and you rank on page 2, your internal anchors are probably too vague.

Step 6: fix internal anchors first

This is where you get results without asking anyone. Rewrite the links you control so that each priority page receives several descriptive variants, from different source pages, inside the body text. Example of a rewrite in HTML:

<!-- Before: generic, no topical signal -->
<p>To check your links, <a href="/seo-tool/backlink-analysis">click here</a>.</p>

<!-- After: descriptive, readable out of context -->
<p>Start with a <a href="/seo-tool/backlink-analysis">free backlink checker</a> to list the domains that link to you.</p>

<!-- Image link: the alt text is the anchor -->
<a href="/seo-tool/backlink-analysis"><img src="/img/checker.png" alt="Backlink checker report"></a>

Step 7: adjust future backlinks and monitor

Do not mass-email old linkers to change their anchors: it is time-consuming, and an editor suddenly switching a branded link to a keyword link is exactly the kind of change you want to avoid. Instead, brief future partners: brand name, page title or a natural phrase, and let them choose. Re-run the classification every month on your priority pages and log the changes. Sorank's natural backlink acquisition follows the same logic: links are placed inside articles from sites in your niche, never a direct competitor, the dashboard shows each link's anchor, and anchors vary from brand names to natural phrases.

Writing a good anchor, sentence by sentence

Ratios matter less than the quality of each anchor. Here are the rules we apply when we write or review a link.

Weak anchor	Why it fails	Better anchor
click here	Says nothing about the destination	free backlink checker
SEO	Too broad, could be any page	on-page SEO audit
best cheap SEO backlink checker tool free online	Stuffed, unreadable	backlink checker
our article about how internal links can help your rankings in many ways	Whole clause, no focus	internal linking strategy
pricing (pointing to a features page)	Broken promise	product features

Length and scope

Aim for two to six words, the shortest phrase that still names the destination. One noun phrase is ideal: "keyword clustering tool", "list of cheese types". If you need a full clause to make sense, link the key part of it, not the whole sentence. Google itself shows a "weirdly long" anchor as a bad example.

Match the promise to the destination

The anchor is a promise. If it says "pricing" and lands on a features page, users bounce and the label teaches Google the wrong thing. Before saving a link, open the target and check that its title and first paragraph answer the anchor.

One anchor, one destination

Never use the same anchor for two different pages of your site. If "backlink audit" points to a guide in one article and to a tool in another, you send two conflicting labels for the same words, a form of keyword cannibalization at link level. Keep a simple sheet: anchor phrase, target URL.

Let the surrounding text work

The sentence around the link carries context too. A partial-match anchor in a precise sentence ("compare the types of backlinks by value and risk") is often better than an exact match in a vague one. Write the sentence first, then choose the anchor inside it.

<!-- Crawlable -->
<a href="/seo-tool/keyword-clustering-tool">keyword clustering tool</a>

<!-- Not reliably crawlable: no href -->
<span onclick="goTo('/seo-tool/keyword-clustering-tool')">keyword clustering tool</span>
<a onclick="goTo('/seo-tool/keyword-clustering-tool')">keyword clustering tool</a>

Common mistakes, how to spot them and how to fix them

These are the errors we see most often in audits, from the most expensive to the most frequent.

1. Exact-match overload on money pages

What it is: a commercial page whose backlinks mostly repeat the target keyword, usually because the links were bought, swapped or placed through guest posts with imposed anchors. How to spot it: in your per-URL split, one keyword anchor stands far above the competitor median, and the page stalls despite new links. How to fix it: stop the source of these links, build new links with branded and natural anchors, and strengthen the page through internal links. If some links come from obvious networks, review them as described in our guide to toxic backlinks, and reserve the disavow file for real manual-action risk.

2. Measuring at domain level

What it is: concluding "my profile is 60% branded, all good" while one product page is entirely keyword anchors. How to spot it: you only have one distribution chart for the whole site. How to fix it: recompute per target URL, starting with the pages that bring revenue.

3. Generic internal anchors

What it is: "click here", "read more", "this article" in your own content. How to spot it: filter the crawl export on the generic list from the formula above. How to fix it: rewrite each one with the destination's topic. It is the fastest win of the whole audit, since you control every link.

4. Same anchor for two pages

What it is: one phrase linking to different URLs across the site. How to spot it: in the crawl export, group by anchor and count distinct targets. Any anchor with two or more targets is a conflict. How to fix it: decide which page owns the phrase, repoint the links, and give the other page its own vocabulary.

=QUERY(A:C, "select C, count(B) where C is not null group by C label count(B) 'links'", 1)
-- then, for distinct targets per anchor, in a pivot table:
-- Rows: anchor (column C) | Values: COUNTUNIQUE of target (column B) | Filter: value > 1

5. Sitewide links doing all the work

What it is: the only links to an important page come from the menu or footer, with a one-word label like "Tools". How to spot it: the page has hundreds of inlinks but only one or two distinct anchors. How to fix it: add contextual links from relevant articles, each with a different descriptive anchor. The Zyppy data ties variety, not volume, to clicks.

6. Empty anchors and missing alt text

What it is: image links, logos and card links with no text and no alt. How to spot it: "(empty)" in Search Console, empty anchor cells in the crawl. How to fix it: write a short descriptive alt on linked images, or wrap the card title in the link.

7. Keyword-stuffed long anchors

What it is: anchors like "best cheap SEO backlink checker tool free online 2026". How to spot it: sort anchors by length; anything over eight words deserves a look. How to fix it: keep the core noun phrase, move the rest to the sentence.

8. Rewriting old backlinks in bulk

What it is: asking every site that links with your brand to switch to a keyword. How to spot it: an outreach campaign whose only goal is anchors. How to fix it: drop it. Only request an edit when an anchor is factually wrong or points to a dead page, which is a link reclamation task, not an optimization one.

Special cases: new sites, ecommerce, multilingual, recovery

Situation	Main risk	Priority action
New site, under 50 backlinks	Each keyword anchor weighs heavily	Branded and URL backlinks, topical internal links
Ecommerce categories	One short label from menus and filters	Descriptive links from guides and posts
Multilingual site	Anchors in the wrong language	Translate anchors with the content, check hreflang targets
Over-optimized profile	Links devalued, page stalls	Stop the source, dilute with natural links, fix internal anchors

New site with few backlinks

With 10 or 20 backlinks, each one weighs heavily in the split. Two keyword anchors already make a visible pattern. Keep early links branded or URL-based, and put your topical effort into internal links and content.

Ecommerce category and product pages

Category pages often receive anchors from filters, breadcrumbs and menus, all with the same short label. Add descriptive links from buying guides and blog posts ("waterproof hiking boots for women") so each category has a vocabulary, not one word.

Multilingual sites

Anchors must be in the language of the page that holds the link. A French article linking to a French page with an English anchor sends a weak signal. When you translate content, translate the anchors too, and check that they point to the matching language version.

Recovering from over-optimization

Since Penguin runs in real time and devalues rather than demotes, recovery is mostly about dilution and rebuilding trust: stop the manipulative links, earn new natural ones, and strengthen internal anchors. There is no instant switch, but there is no fixed waiting period either.

Anchor text in the AI search era

AI assistants that browse the web, and the crawlers that feed them, read links too. A descriptive anchor tells any reader, human or machine, what the destination contains before it is fetched. Generic anchors give no such hint. The practical consequence is simple: the rules that serve Google (descriptive, concise, consistent per page) also make your site easier to map for AI systems. There is no separate "AI anchor strategy" to build.

Link as written	What a reader or crawler knows before following it
Read more	Nothing, except that the page continues somewhere
this tool	That the target is a tool, not which one
free internal link analyzer	The type of tool, its function and that it is free
sorank.com/seo-tool/internal-link-analyzer	The site, the section and the tool name from the URL

What changes is the weight of your own site's structure. Clear internal anchors around a topic help show that your site covers it in depth, which is the kind of signal a topic cluster is built on. If you plan clusters, our topic cluster generator groups pages by topic and flags overlaps, which makes the "one anchor, one destination" rule easier to keep.

Your action plan for this week

You do not need a full audit to start. Here is the shortest path to results:

1. List your five most valuable pages (revenue or leads).
2. For each, crawl the internal links that point to it and fix every generic, duplicated or empty anchor.
3. Add two or three contextual links from relevant articles, each with a different descriptive anchor.
4. Export backlink anchors for the same pages and classify them with the formula above.
5. Compare with the median of the top 10 for each target keyword and note the outliers.
6. Brief your next link partners with branded and natural anchors only.
7. Put a monthly reminder to re-run the classification.

Anchor text optimization is less about the perfect ratio than about consistency: every link should say clearly where it goes, and the sum of your links should look like what editors would write anyway. Control what you own, stay natural on what you don't, and measure page by page.