A correlation between anchor text and search rankings does not mean that switching to a particular type of anchor text will improve rankings. In research published in a 2019 article, Ahrefs found only a weak positive correlation between the proportion of exact-match anchor text and rankings. Correlations for phrase-match and partial-match anchors were even weaker, while generic anchor text showed almost no correlation. These were observations from a specific sample of English-language searches, not a current ranking formula, and they cannot establish an optimal anchor-text ratio.
The study compared anchor-text proportions in external links, using a screened sample
Study 1 analyzed the top 20 results for 19,840 keywords, covering 384,614 unique pages. The keywords consisted of two to four English words and had monthly search volumes of 2,000 to 5,000. Queries containing special characters or consisting entirely of numbers were excluded. The study also required the top 10 results to have similar URL Ratings (UR), in an attempt to reduce the influence of differences in link strength.
The findings therefore cannot be applied directly to Chinese-language searches, all search-volume ranges, or internal links. These selection criteria make comparisons easier, but they do not mean that all factors affecting rankings were controlled for.
The findings for these four types are not a recipe for better rankings
The following are the Spearman correlation coefficients reported in the study, with the mean followed by the median. They do not represent the size of any ranking improvement:
- Exact match: 0.1436 and 0.1869, indicating a relatively weak positive correlation.
- Phrase match: 0.1057 and 0.1393, indicating a very weak correlation. For example, the target keyword might be “SEO tool” and the anchor text “best SEO tool.”
- Partial match: 0.1076 and 0.1393, also indicating a very weak correlation.
- Generic text: 0.0161 and 0.0130, indicating virtually no correlation. Examples include “click here” or “this article.”
The study’s definition of partial match also included exact and phrase matches. The categories therefore overlap, so the results cannot be treated as a comparison of four independent strategies.
Keep average proportions and correlation coefficients separate
In the study, the median proportion of exact-match anchor text was zero at every ranking position, indicating that many pages had no backlinks of this type. The article illustrated this with a hypothetical example: if four of five pages have a proportion of zero and one has a proportion of 100%, the average is 20%, but that does not represent most pages.
This “median proportion” is different from the “median correlation coefficient” discussed above. Looking only at trends in average proportions can lead readers to overestimate the relationship. Even an observed positive correlation does not prove that adding exact-match links causes rankings to rise.
Conversely, a weak correlation does not mean anchor text has no effect. The article noted that John Mueller’s explanation did not specify how much weight the algorithm gives anchor text.
Describing a link’s purpose has more support than pursuing a fixed ratio
A 2020 Semrush article recommended natural, descriptive anchor text, advised against creating large numbers of links with identical text, and warned that unnatural manipulation could risk penalties. This was practical advice offered at the time. It does not mean the study validated any safety threshold, nor should it be treated as current official guidance.
Internal links are a different context. In an article on site structure published the same year, Semrush used the example of a “dog toys” category linking to product pages to explain how anchor text and the relationship between pages can help clarify a link’s purpose. But this example cannot establish a causal relationship between anchor text in external links and rankings.
Anchor text and rankings: perspectives from across the web
Does using a brand name as anchor text guarantee better rankings?
No. One forum response suggested that branded anchor text helps rankings and recommended using different wording for variety. However, this was a personal opinion, and the supplied material included no research verifying the effect of branded anchor text. A brand name is one option for link text, but that does not establish that using it will improve rankings.
Do two links with identical anchor text have the same value for rankings?
Not necessarily. A response on a webmaster forum suggested that the relevance of the linking source and the source page’s ranking for that term could both affect a link’s value. Two links with identical text could therefore carry very different weight. This was a forum contributor’s assessment, not a verified, fixed weighting or an official ranking rule.
Does surrounding content still help when the anchor text is generic, such as “click here”?
It may provide context, but an effect on rankings is not guaranteed. An Ahrefs author cited a patent filed in 2004 that described using information or anchor text to the left and right of a link to determine its context. On that basis, the author suggested that surrounding text might help search engines understand the destination page. Neither the patent nor the author’s inference proves that search engines necessarily work this way today.
References
- Link building suggestion to improve 2k postswww.warriorforum.com — Discusses: Does using a brand name as anchor text guarantee better rankings?
- How Do Search Engines Rank Combined Keywords?webmasters.stackexchange.com — Discusses: Do two links with identical anchor text have the same value for rankings?
- Anchor Text: A Data-Driven Guide (384,614 Web Pages Studied)ahrefs.com · 2019-02-26T20:07:54+00:00 — Discusses: Does surrounding content still help when the anchor text is generic, such as “click here”?
- What are backlinks in SEO & how do I get them?www.semrush.com · 2020-04-16T21:51:00Z — Source used in the article above.
- What Is Website Architecture? (& How to Improve it for SEO)www.semrush.com · 2020-12-22T14:44:00Z — Source used in the article above.



