Study: How Often Google Ignores and Rewrites Our Meta Descriptions

Evan Hall, SEO Strategist

Since search engines started using meta descriptions in their search result snippets, they’ve always had the challenge of deciding what to display when the meta description won’t work or isn’t relevant. Usually, a search engine will take an excerpt from the page when a meta description tag isn’t viable. To prevent search engines from taking bizarre excerpts from pages to make a snippet, SEO practitioners have put great attention into making sure meta descriptions are present, unique, and descriptive.

This worked great for a while. An SEO could rely on the search engines to use the meta descriptions we wrote the majority of the time, but that has been changing. In the last few years, Google has been displaying content excerpts in the snippet over meta descriptions more and more. And this is distressing for SEOs because we spend a lot of time making sure our meta descriptions are good!

So a question arises: how often does Google ignore our meta descriptions? How often is Google displaying some other text in their SERP and ignoring our hard work? Recently, Ahrefs published some original research on the topic and found that Google disregards a page’s meta description 63% of the time for pages in the top 10 results.

I thought that figure was pretty high. I believe it, but I don’t want to believe it. So, to corroborate Ahref’s research, I pitched an idea to Matthew Henry, our SEO Fellow and technical wizard: let’s do our own version of this research and see if we get a similar result. If Google is fairly consistent and both Ahrefs and we have a comparable sample of keywords, we should get roughly the same results. Isn’t replicating research the heart of the scientific method?

Methodology

Before we get into the results, we need to talk about where we got our data. We used a list of 30,000 keywords taken from our existing and previous clients with brand terms filtered out. We entered the keywords into STAT Search Analytics and pulled rankings for both desktop and mobile search results. Using STAT’s full HTML SERP export, Matthew scraped the display snippet descriptions from Google’s HTML after laboriously parsing Google’s wonky HTML structure.

Matthew then wrote a custom scraping script to go through each search result URL and find the first listed meta description tag. The SERP descriptions and the meta description tags were added to an SQL database and checked to see if the text Google displayed was included in the meta description tag. If the SERP snippet text wasn’t included in the meta description text, we counted it as a case of Google ignoring the meta description.

We didn’t render JavaScript, and we didn’t try to bypass clever firewalls. If we couldn’t retrieve a page or a meta description, we didn’t include it in the final data.

Results

Below are our research results, which set out to analyze both desktop and mobile meta description rewrite rates based on the following conditions:

  • First-page ranking
  • Search volume
  • Query length

We also took the opportunity to use this data to evaluate meta description display lengths and rewritten snippet display lengths.

First Page Rewrite Rate

We found the rewrite rate for meta descriptions on the first page to be 71% in mobile search results and 68% on desktop. In other words, we should expect Google to use our meta description tag for the snippet around 30% of the time when we rank on the first page.

This rewrite rate is a little higher than Ahref’s number (63%), and that’s not due to featured snippets being included in these figures; we took them out. This difference between our overall rewrite rates could be explained by us including more keywords that are rewritten more often. There were some dimensions with high variance in the rewrite rate, and we could have selected for those unknowingly.

The rewrite rate isn’t uniform by position, though. Here’s the rewrite rate by position for the first three pages of results.

""

""

The data includes featured snippets in the first position, so that explains the spike there. But do notice the “bump” from positions 4 to 6. Why would Google rewrite meta descriptions in those positions more than others?

I speculate that since positions 1-3 get the most click-through rate, Google might be trying to boost the relevance for 4-6 to get more clicks before users leave the page or search for something else.

The other interesting thing in this data is how the rewrite rate tends to increase with position. I’ve noticed that the results on pages 2 and 3 aren’t the most relevant query, so maybe Google is trying to pull a more relevant excerpt from the page. I think that’s plausible because the results on pages 2 and 3 might not even be targeting my query.

Rewrite Rate by Search Volume

I expected a more uniform rewrite rate over our keyword set when we bucketed them by search volume, but there is a clear trend with mobile having a higher rewrite rate.

""

I’m fairly confident of the relationship between search volume and rewrite rate: the higher the search volume of the keyword, the less likely Google is to rewrite the meta description. It’s not a linear relationship, though; our x-axis is log scale here.

So why do we see this relationship? I think it’s because SEOs tend to focus on writing meta descriptions for head terms more than the long-tail. If you’re ranking on the first page for a keyword with 1 million searches per month, you are likely aware of that and put significant effort into your meta description. You’re probably not hyper-focusing on the 10 searches per month terms, and they probably vary too much to even target with any one description.

What about query length? Would we also see a similar trend for queries with more characters since they tend to be long-tail queries?

Rewrite Rate by Query Length

Just like search volume, long-tail queries tend to have search result snippets that don’t use the page’s meta description.

""

Desktop and mobile results are very similar, with mobile having a slightly higher rate of Google ignoring the meta description. We didn’t have many queries in our data set with more than 45 characters, so we don’t know much past that.

Meta Description by Display Length

As long as we had scraped search results, we thought it would be nice to have recent numbers on meta description display limits. Even though meta description display space is limited by pixel width, targeting an ideal pixel width is difficult with meta descriptions because there is often a gap of unusable space at the end of the first row. This makes character counts easier to work with and communicate.

""

""

There are a few things going on with displayable meta description length. On desktop, displayed characters peak at 156 and then rapidly decline after 165, but also notice the other “bump” that peaks at 142 characters. There’s a similar bump on mobile too. I think that’s due to pages with a date in the snippet shortening the meta description able to be displayed.

In this screenshot, you can see the date snippet Aug 13, 2020 placed at the beginning of the meta description, taking up 12 characters.

The spike at 87 characters is caused by a single domain using the same meta description on every page. They must be over-represented in our keywords for them to have such a large effect.

Mobile results show a similar pattern as desktop with the two display limits that depend on the publication date’s presence in the snippet. Luckily, we’ve segmented our data by whether the snippet uses the date to see if the snippet date is causing a difference in meta description display length.

Meta Description Length With Publish Dates

Here, we’ve separated the desktop results where the meta descriptions that have the publishing date rich snippet feature from those that don’t. Side by side, there is a clear difference between the two, with fewer displayable characters when the publish date is in the snippet.

""

""

Without a date present, fully-displayable meta descriptions are going to be between 150 and 160 characters. If you’re feeling lucky, you might get away with 165 characters.

If dates are robbing us of some displayable space, what length should we aim for with pages that have dates listed, like blog posts? This chart suggests 138 to 148 characters is a safe target if we want a fully-displayable meta description on desktop with a date in the snippet.

The difference between search results with snippet dates and without on mobile is a little different than on desktop. If there is a date feature in the snippet, we can get just as many characters to display, but we’re much less likely to.

""

""

Our mobile search results show that we have less room to work with for meta descriptions than on desktop. Displayable characters for snippets without a date rich snippet feature peak at 118, and then rapidly drop off after 121. This means that a safe target character count for mobile results without a date is under around 120. That’s about 25% fewer characters than the non-date target amount for desktop. That’s significantly less room to work with!

Mobile snippets with the publish date featured snippet look a little weird. It appears that the most common snippets are going to display between at most 95 to 105 characters, but there is a smaller possibility Google will show between 112 and 126 characters. Maybe Google is allowing some mobile snippets another line of text when the date is present? Either way, content with a publish date on the page, like blog posts, don’t have a lot of guaranteed display space.

Rewritten Snippet Display Length

So how much space does Google give themselves to write a snippet description? It turns out it’s a little bit more than the space they give us.

""

""

In desktop results without the rich snippet date feature, Google’s descriptions peak at 160 characters and then rapidly decline after 167. With dates, the peak is at 147 and declines quickly after 149. Not a big difference from snippets using a meta description tag, but it would be nice if we had just as much room to write a nice description for our content.

Also, notice the frequency of snippets around 225 characters and 325 characters. Remember back in December 2017, when Google increased the meta description length to 320 characters and then rapidly changed their minds? Looks like the 320 character display limit is still there, but only for snippets Google writes.

""

""

In mobile search results, Google gives itself around 114 to 121 characters to work with for snippets without the date feature, with the peak at 118. When the date is in the snippet, the sweet spot appears to be between 99 and 105, with the peak at 100.

Just like desktop snippets, there are a few extended snippet ranges. Those extended snippets appear to be around 150, 175, 200, 250, and 320 characters long. That’s considerable space to take up for a mobile device.

So What Do We Do With This Information?

I’m somewhat disheartened about how infrequently our meta descriptions are actually displayed in the SERP, and how few characters we’re likely to utilize. But I don’t think that means we should give up on writing good meta descriptions. We just need to be more clever about the process.

With the rewrite rates being lowest on short, high volume keywords, we should continue to be ambitious about which keywords we’re writing for, but we should keep mobile display lengths in mind. I’m still going to recommend writing between 150 and 160 characters for desktop-focused meta descriptions, but I’m going to add the caveat that the first 120 characters should include the kernel of the message for mobile search results. If a client’s qualified traffic is primarily mobile, I’ll treat the extra desktop space as a bonus.

The publish date rich snippet feature means we need to be aware of what content type we’re writing meta descriptions for. If I’m auditing descriptions for blog posts, I’m going to recommend a target of 138 to 148 characters for blog posts and other pages with publish dates. Since the majority of snippets on mobile with dates are cut off after 105 characters, the most important part of the description should be in the first 100.

I’m also going to start experimenting with adjusting content that Google is using in their excerpts. One way to view Google’s excerpt snippets is that there are snippet descriptions distributed around the page, and they can be optimized just like a meta description tag. I’m pretty sure Google selects content to use for the snippet in the same way as featured snippets.

Future Research

There are definitely ways to improve our process for this kind of research. One way that sticks out is a keyword list that’s more representative and uniform over dimensions like query length and search volume. There might be a time-efficient way to categorize keywords by vertical. That would be fun to look at.

Naturally, title tags and featured snippets are good choices to analyze too. How many characters do we get for paragraph featured snippets? How often does Google ignore our title tags?

As Matthew Henry can attest, doing this kind of research is hard to perform, but I look forward to others attempting to answer these questions. As Google changes the layout of the SERPs and introduces new features, we’ll need this research to make sure our recommendations and practices are current.

Start call to action

See how Portent can help you own your piece of the web.

End call to action
0

Comments

  1. This would seem to support the continuing, consistent and long-standing recommendation from Google to “write great content for the user, not for the search engine”. A good meta-tag should be a result of the content, and and indicator to what to find on the page at a generic level. As search engines continue to become more semantic, they will seek to provide users with answers to their questions from the page. This should not be disheartening to a content and user focused SEO, but those who focus on technical issues primarily may struggle.

    1. ^^ Yes, what Ted said! I just did google searches for various site pages. 15/15 display my written description. It pays to listen to Google’s recommendations.

    2. TOTALLY AGREE WITH TED! It’s about the End-User and THEIR experience. it’s not all about ranking. I think focusing soley on rank – is short sided. I understand you did test the data. But I can tell you I’ve been doing organic and local SEO for 15+ years and I’ve never had a single meta re-written. Ever.

  2. Good point Evan about shortening meta descriptions for mobile devices. Interesting that Google rewrites so many of them. I guess their AI knows more about my blog posts than me! LOL

  3. Thanks for sharing the research. It would be interesting to study pages with optimized meta descriptions by themselves. Many non-optimized pages will have meta tags that pull automatically from the page, or which will not be written in a way that truly reflects the content of the page. When we write meta descriptions we try to keep both things in mind. I’m wondering if Google is less inclined to do rewrites on deliberate, crafted meta descriptions. Perhaps something to study in future.

  4. Great research! Especially the correlations between ranking and search volume.

    Maybe I missed it, but how many pages had no meta desc to begin with? Cuz that’s enough reason for google to rewrite them.

    1. The rewrite rate is based only on those pages that have meta descriptions (which accounts for about 75% of the total pages crawled). Pages that had no meta description were not counted when calculating rewrite rate. In other words, the rewrite rate is defined as the number of pages that had a meta description that was not used, divided by the total number of pages that had meta descriptions.

  5. Hey guys, such a great study! A quick question: the “Search Query Lenght” parameter is in words or characters? If in words, why not focusing on short head queries with only one or two terms?

    1. Lengths are based on character counts, so the leftmost edge of the chart (which has a value of 5) shows results for five-character queries, which are probably all single words.

  6. Thank you, yes I had too experienced the same problem, but I think these days google is not taking meta description seriously even from a ranking perspective.

  7. Great Article, i always look forward to read more like this. really helpful. keep it up. Thanks for sharing great SEO tips. Keep posting awesome and helpful content for increasing our knowledge. I love your blog.

  8. Here is what I would like to see next.

    Either Google makes up the replacement or it uses content on the page. Does it do both?
    If it uses content from the page, does it snag sequential content from the page?
    Does it snag content from schema, og/twitter cards, headers, paragraphs or spans?
    Does the length of the content it replaces matter?

    It would appear that Google is showing content that is of a fairly stable length? Is it just looking on the page for the content that both resembles the most searched for queries and length of content?

    Should we move the content that it snags for the replacement higher on the page?

    1. Hi Marston,

      Google takes an excerpt from the page when it chooses not to use the meta description. I’ve never seen it invent snippet text. The excerpts are usually sequential. I had to dig through a few pages of results with a few keywords to find an instance of non-sequential excerpt text.

      Google generally puts emphasis on content users can possibly see, so I doubt text that is only in structured data markup, Open Graph, or Twitter Card is eligible.

      We didn’t look at rewrite rate by meta description length. That is something we could look into for the update to this research.

      Google must have some mechanism for detecting when a query deserves a long snippet, since it’s not too frequent it must not be very often. I think Google is using an NLP approach like BERT to find passages of text that either resemble an answer to the query or lead to higher click-through rate. Text that’s relevant to a query should contain some of the words in the query, or variations that mean the same thing.

      We didn’t look at where Google takes excerpts from on the page, and we thought that would be a fairly difficult thing to detect and measure. Google might have looked to see if excerpt position is correlated with user satisfaction, but my hunch is that relevance is a stronger driver of user satisfaction. In any case, would a user be better served if the answer to their query is toward the beginning of a page instead of the end?

  9. Just a few days ago I spotted that Google showed weird caption from general description as meta description even if there is set specific sentence for it. Now I read this post and it’s giving me idea what could possibly to trigger it. Thanks for Post

  10. Great share. This will really help me work on my new website SEO. Thanks for writing such a great post.

  11. Thanks for sharing the research. It would be interesting to study pages with optimized meta descriptions by themselves

  12. I like your blog because your content is very informative and knowledgeable. keep share with us more ideas.

  13. This is the best article i like it very much. This is very informative and useful content. please keep sharing with us.Thanks for this great post!

  14. I really love your post, thanks for sharing this and looking forward to seeing more from you.

  15. Again a valuable post. I haven’t read any case study on Meta description with this in-depth insights. However, I have seen many case studies about how to write meta description properly but this is truly amazing and thank you so much for sharing it…

    Thanks

  16. This is very well thought out! I enjoyed reading your article .
    Thank you and keep these good articles coming.

Leave a Reply

Your email address will not be published. Required fields are marked *

Close search overlay