Does readability affect search engines?
What is Readability?
Google “Reading Level” search filter
Was Google using common readability formulas?
How Google “Reading Level” correlated with readability formulas?
Was/is Google using its “Reading Level” for ranking?
Is Google still using “Readability Level” under the hood?
Does “Readability” in general help ranking?
Asking Google
Non-organic benefits of easy to read content: articles virality
There is not only ranking
Conclusions
Does the ease of reading have any impact on search engines?
Search engines are machines processing and ranking a large variety of (hyper) texts.
It comes natural to wonder if they use – among all other things – an estimation of the text Readability.
What is Readability?
Readability has been researched during the last 70 years, with studies funded by both governments and private companies. There are tens of readability formulas to assess it.
In some US states reading ease is a legal requirement for many documents, and is a common editorial policy of newspapers and magazines to impose a minimal readability score.
Traditional readability indexes do not say whether texts are correct, state the true, or make any sense. They are simple formulas based on the statistics of the language.
Google “Reading Level” search filter
Not many people know Google used to optionally show (and filter by) reading level on the English SERP. Google announced the feature on Sep 2010 and dropped it around May 2015.
Google Readability Levels
Web pages were classified using three levels: Advanced, Intermediate, Basic.
As a Google researcher explained: “Basic” meant elementary level texts, “Advanced” roughly the level you find in Google Scholar, and “Intermediate” everything in between.
Was Google using common readability formulas to compute the “Reading Level”?
Apparently not. They developed their own based on statistical models.
Most reading score systems were developed 70-30 years ago. The well known Flesch formulas were first developed in 1948 (Flesch Reading Ease Score – FRES) and 1975 (Flesch-Kincaid Index). Even if they still are largely in use, they were conceived in a time when computing power was much inferior to today. Artificial intelligence and machine learning were still to be born.
Google decision is not a surprise given their sheer amount of data, processing power and engineering talent. Also Microsoft experimented statistical models for similar tasks.
How Google “Reading Level” correlated with common readability formulas?
I found surprisingly few studies comparing readability indexes for English language to Google score. The only quasi-formal study I’m aware of attempted to compare Google “Reading Level” to Flesch Reading Ease Score and Flesch-Kincaid grade level. It didn’t conclude much. The study had several flaws, first of all a too small sample of data. Google only showed three levels and that doesn’t help trying to match values from a larger scale.
Unfortunately not being the Reading Level visible any longer, we are unable to make an independent study.
Was/is Google using its “Reading Level” for ranking?
Since the filter appeared, there has been much debate on whether Readability where a so-called ranking factor.
While several people claimed having observed a positive correlation between Reading Level and positioning, we couldn’t find any formal study. Again, we cannot make our own, having Google dropped the search filter.
Is Google still using “Readability Level” under the hood?
The fact the search filter was dropped doesn’t mean the level is not still computed. It could be like the “Toolbar PageRank” which was dismissed, but the real PageRank is of course still used. Google has invested a lot developing its own score system; my bet is it still is doing it and likely has improved it significantly, probably also for other languages.
Does “Readability” in general help ranking?
This is an area where more studies can be found. The most prominent is SearchMetrics Ranking Factors 2015 which examined the correlation between Google search position and the Flesch Reading Ease Score. The study seems to demonstrate how higher ranking pages generally have higher readability score; on top of that, this trend increased since 2014.
SearchMetrics 2015 study maps average Flesch score against positioning
It can’t be so simple, of course. Every niche must have its own averages, tales for kids and academic material cannot aim to the same readability levels.
Correlation does not imply causation, but there are circumstantial evidences to suggest Google is taking some sort of readability estimation into account when ranking.
A notable case was the debate in September 2014: Google systematically ranked a low quality site for queries about web technologies better than the authoritative Mozilla Developer Network.
Mozilla blamed the competitor of using unfair “SEO tricks”, but on a Hacker News thread Google engineer Ryan Moulton (“moultano”) gave an alternative possible explanation. He did not explicitly use the terms “Readability” and “Reading Level”, nevertheless his statements give more than a clue: he explained that the search engine often favors texts easier to understand. That is exactly what Readability is all about.
I have to point out that the thread dates to a time when “Reading Level” was still optionally visible in the SERP. The Google engineer could have been describing an inner working now no longer in use.
Asking Google
There is another way of getting a piece of information: asking to the source. That’s what BEM Research co-founder Mariachiara Marsella did on Twitter. She didn’t get a direct answer, but as she recounts (in Italian) it seems she had an indirect confirmation: John Mueller didn’t answer directly, but “loved” a tweet by Andrea Pernici in reply to her question. Andrea stated in his opinion anything improving User Experience improved a site visibility. Up to you judging: John may have not given a direct answer for any reason, and his “love” could mean anything.
Non-organic benefits of easy to read content: article virality
Common sense suggests easier to read content is also easier to be shared.
Someone went a little further and inspected the Readability of the most shared Buzzfeed articles comparing them to the most shared articles of two popular on-line newspapers.
scribblrs.com study on viral posts
They took care to average the four common grade-level estimation formulas to get a more accurate assessment of the article reading levels.
The key finding is: highly shared articles are highly readable. What they missed to do is providing a graph mapping readability and number of shares; it would have indeed needed a much larger data set.
There is not only ranking
Search engine might use readability evaluation for other purpose: for example many professionals have pointed out badly spun articles tend to score badly on readability scores. Being readability scores quite simple and fast to be computed, they could be used to trigger an initial alarm for spam detection.
Conclusions
We miss a smoking gun to directly relate Readability to organic search ranking.
Google has for sure invested huge resources to develop their own readability score, without resorting to last century readability formulas. Such traditional readability formulas seem to be demonstrated to have a good correlation with organic ranking, and they likely on average correlate well with Google “reading level”.
While at the same time there are several elements to suspect Google is using also readability as part of the so-called ranking factors, high readability seems also to be an element of highly shared contents. Since highly shared contents are also normally highly linked, readability could be an indirect factor: increasing the page chances to be linked, thus making it gain more PageRank.
We need more and better studies and experiments. SearchMetrics’ study is a good start, but concentrated only on a readability formula which is nearly 70 years old (even if still largely accepted and in use), and has all the limitations of correlation studies.
My opinion:
I strongly suspect Google is still using its Reading Level assessment, or an evolution of it. I think the three levels were just an approximation of a larger scale (like the “Toolbar PageRank” was a scalar approximation of the value of the real PageRank), and that such value was and still is used as a direct ranking factor (perhaps not for all languages).
This is just a personal opinion, a strong suspect. I cannot demonstrate it.
And what’s your take? Do you think readability to be important for SEO?