How Many Outbound Links Per Word or Page? [Original Study]

on under Myth-Busting, SEO Auditing.

If you google for an answer to how many outbound links you should be seeing on your pages, most answers fall into one of these categories:

  • However many is most useful for users (okay, yeah, there’s truth there, but it’s not very useful, and everybody knows it’s not what you’re looking for)
  • Less than 100, because ages ago that was the maximum Google could crawl (which is no longer true)
  • None, ever, because you will bleed your PageRank or because Matt Cutts once said that outbound links can be a spam signal
  • Link ALL the LINX!!!1!, because Matt Cutts once said that there are aspects of the algorithm that reward linking to good sites
  • Some arbitrary figure pulled out of thin air

Well, I don’t like those answers. Neither does my boss, Corey Northcutt, who wants our rigorous audits, with their zillion-point checklists, to be based on hard data and fact-based opinions.

And so, to alleviate the problem, I conducted some original research. Here’s what I found.

Methodology

Before I share the results, let’s talk about where my data is coming from:

  1. I used this random word generator to generate an arbitrary but commonly used word. (I quickly realized when using a different tool that it was important for the word to be somewhat common, otherwise I was getting dictionary results, which would dramatically skew the kinds of pages I was looking at.)
  2. I then took the word to the Google Analytics Keyword Planner and sorted the keyphrase recommendations by least popular, and used the least popular query. Why? Because these are the most natural results. Popular search queries are competitive, so the top ranking pages are more likely influenced by SEO “best practices” and other marketing disciplines. I’m interested in learning about the algorithm here, not in what other marketers are doing. A study like this is still only looking at mere correlation, but where I can weed out noise from marketers, I will. (This is a major issue I have with other SEO correlation studies.)
  3. I ignored any queries that were related to videos, games, local, and any remaining queries that skewed toward dictionary results. I didn’t want the results tainted by influences from local algorithms, and since videos and games aren’t going to use many words, they’re of a completely different category when it comes to things like words per outbound link, and would serve as distracting outliers. Same again for the dictionary results.
  4. I used a private browser window (a new one for each query) to pull the top ten search results, ignoring any non-web results like images, knowledge graph, etc.
  5. I ran each of the pages through Screaming Frog and captured the word count, outlinks, and external outlinks
  6. I repeated this for a total of 20 queries, or 200 total search results.
  7. The histograms and scatterplot were made in Google Sheets (with the red bars highlighted in, yes, MS Paint)
  8. The ranges, interquartile ranges, medians, correlation, and statistical significance were found in Minitab 16.1.1 (I’ll elaborate on these, don’t worry.)

Now, the results

How Many Outbound Links Should You Use?

Our list of ranking factors mentions:

But we wanted to get more specific.

Let me cut to the chase by saying that if your average page is falling outside of these ranges, it’s worth asking yourself if your outbound linking strategy makes sense:

  • All outbound links per page: 56 to 171
  • All external outlinks per page: 5 (or 0) to 42 (More on this later.)
  • Number of words per outlink: 6 to 23
  • Number of words per external outlink: 29 to 232
  • Also, a teaser: if you think external links are hurting your rankings, you might want to consider the possibility that the opposite is true.

Where did those numbers come from? Well, I looked at 200 search results: the top 10 for 20 “random” search queries. (“Random” here being defined above, in the Methodology section.) The middle 50% of those pages all fell within these ranges. Since these were all pages that made it to the top 10 search results, we know Google isn’t booting these pages out for falling in this range.

Now for a bit more depth. Let’s start by taking a look at all outbound links per page:

histogram of outlinks

 

This is a histogram. Each bar represents a chunk of pages 30 outlinks wide. So the first bar counts all of the pages with 0 to 30 outlinks on them. The next bar counts all of the pages with 60 to 90 outlinks on them. And so on. As you can see, once you start to get out to about 450 outlinks, things are starting to look pretty lonely.

This contains the complete range of total outlinks I found for each of those pages. Not surprisingly, the minimum was 0 outlinks. The maximum was 1033. So, yes, you can make it to the front page of Google with 0 outlinks, and you can make it with 1033 outlinks. Good to know, but that still doesn’t tell us very much.

See that red section in the middle? That’s what statisticians like to call the “interquartile range.” It contains everything from the 25th percentile (called Q1) to the 75th percentile (called Q3).

The Q1-Q3 range is much narrower than the full range. It goes from 56 to 171. So, 50% of the pages that I looked at, relatively natural pages sitting on the front page of Google, had between 56 and 171 outlinks on them. The median, which separates the bottom half from the top half, was 101.

If you have fewer than 56 outlinks, or more than 171, you should probably ask yourself why. It’s not going to keep you from making it to the front page of Google. Like I said, everything here did. But if you’re outside of that center 50% range, you might want to ask yourself if it’s for good reasons.

Words Per Outlink

Since some pages have more words than others, relying strictly on the number of links per page isn’t necessarily a good idea. Another way to think about it is to focus on the number of words per link:

histogram of words per link

The total range goes from literally more links than words all the way up to all words and no links. More usefully, the Q1-Q3 range goes from 6.1 to 22.6 words per link. If you’re outside that range, you might want to ask yourself why. The median was 10.9.

The above is for all outbound links. What about outbound external links? You know, links that point to other sites?

External Outlinks

How did things look for external links?

histogram of external outlinks

The total range for external outlinks per page was from 0 to 254. The Q1-Q3 range goes from 5 to 42. The median, dividing the top 50% and bottom 50%, was 19.

That’s considerably more external links than you might be expecting.

So, a caveat before we move on. Screaming frog looks at all “href” links, not just followed links. That means plugins, comment links, and so on, that point to external sites are getting counted here. However, it’s important to remember that, as far as PageRank is concerned, nofollowed links divide PageRank just like any other link.

Now, the question is, should you be concerned if you have less than 5 external links just because that puts you outside of the Q1-Q3 range? The answer is…maybe.

By no means should every page on your site have more than 5 external links. However, if your sitewide average is fewer than 5 external links per page, it’s worth asking why. Is it merely because you don’t use any plugins? Is it because your site consists entirely of landing pages with no blog style content? If so, it’s likely not an issue. In fact, external links on those kinds of pages could be considered a spam signal.

However, if your blog style content isn’t, on average, starting to approach at least 5 external links per page, you might deviate somewhat from what is natural, and what seems to be ranking well for natural queries.

Words Per External Outlink

Now let’s look at that from the perspective of words per external link:

histogram of words per external link

Incredibly, the minimum was 2.7 words per external link. The maximum was, of course, an entire document with no external links. The Q1-Q3 range went from 29.8 to 231.5 words per external outlink. So, another way of addressing the question above is to look at your pages and see if they go more than 230 words or so without an external link. Shorter pages don’t necessarily need any. Longer pages start to look a bit odd if they don’t have any external links.

Now, why am I concerning myself so much with the possibility of having too few external outlinks?

External Outlinks Per Page Is Showing A Statistically Significant Correlation With Rankings

My research found a positive correlation of 16.1% between rankings and external outlinks. That correlation was statistically significant (for you stats geeks, p=0.027).

Now, this is by no means a knock-it-out-of-the-park result. It’s statistically significant, but it’s still a small study, and yes, it’s correlation, not causation.

For the visually inclined, here’s what the relationship looks like:

scatterplot of external outlinks vs rankings

Now, I could make things look a lot cleaner by showing you a bar graph reporting the “average” number of external links for each ranking, but I want to prove a point here. Like all search engine relationships, this relationship is messy. Still, as you can see from that best-fit red line, the predicted number of external outlinks in position 1 is about 50, while the predicted number of external outlinks in position 10 is closer to 25.

Is this evidence that external outlinks can help your rankings?

On its own, not really. It could be because higher ranked pages get more comments, and therefore more external outlinks in the comment section. It could be because content producers who link out are doing a lot of other things that Google likes. It could be that more people reference these pages because they’re well sourced, which is why they have more external outlinks.

Of course, back in 2009, Matt Cutts said:

In the same way that Google trusts sites less when they link to spammy sites or bad neighborhoods, parts of our system encourage links to good sites.

So, does this mean everybody should start loading their pages with external outlinks?

It’s a possibility, but let’s not jump to that conclusion yet.

Instead, I plan to run a quasi-controlled experiment of my own to see what happens if I make some changes to one of my own sites.

Now I want to hear from you. Are these results surprising? How seriously do you think we should take the correlation between outbound externals and rankings?

  •  
  •  
  •  
  •  
  •  
  • Like many things (most?) in SEO, this is a grey area. That said, I love that you actually dug deep and looked for correlations… and found them! SEO is testing testing and more testing. Either done personally or by others and then followed. Great contribution, thank you!

    • Carter Bowles

      No problem Aaron, and thanks. Yeah, a gray area for sure. SEO gets complicated fast, but that’s why it’s so important to tie everything you can to measurable results.

  • John Greving

    Wow. An exhaustive study to say the least. Thanks for publishing this. I also love that your intro just gets right into it the meat of the matter.

    • Carter Bowles

      Thanks John. Yep, I wanted to make sure it was easy to refer back to.

  • Colby

    Carter, if you still have the data…

    I would love to see how this data stands out compared to results on page 10 for the same queries. Would we find the same spreads and stats, or would they vary somewhat significantly?

    • Carter Bowles

      I agree, that could be interesting to see. I wouldn’t trust the data to remain accurate after this much time has passed. It might be worth investigating in a second study. I’ll consider it.

      • Adam Melson

        I’d definitely share that comparison post. Thanks again for spending the time and sharing and your quick reply here.

  • Adam Melson

    Do you have any plans on running the test again with another randomly generated word? Replicating this, especially with a larger test, would help. Also, if you shared the word used in the initial test, that would help clear some confusion.

    The test focused on the least competitive terms (based on the randomly generated word) and the external links pointing from the sites in search results. If you were to take the same results, see how many websites have that word (or another randomly generated word) as the first word in their title tag, then correlates to rankings, what would we find?

    Kudos for spending the time to test and letting the industry learn from the test.

  • Wusch El Gehirn

    Once again a great data-driven article from Northcutt. Thumbs up!!

    First, as already pointed out by Adam Melson, lots of other factors (e.g., keywords in title, headings etc.) add huge variance to the data. This can or rather must be taken into account in the statistical model. Once you do that, you will very likely end up with higher or at least more trustworthy levels of significance and correlations for the factors of interest. To get started, google for “ANOVA random factor” or email me.

    Second, I strongly support your phrase “it’s correlation, not causation.” However, at least we can exclude that the number of links were an effect caused by SERP rank. Hence, left aside that the level of significance is weak, and further ignoring that many popular statistical tools (ANOVA, t-test) are only trustworthy for data with Gaussian distribution (not the case here!), I’d conclude that SERP rank is at least indirectly caused by something related to the number of outlinks.

    So, the next questions to investigate could be: Which factors correlate with the number and impact of outlinks (e.g., link placement in main content or in navigation header/sidebar/footer, overall wordcount, page rank of linked site etc.), and which of them contributes most to SERP rank. My guess would be that you will find at least one factor with even higher correlation to SERP rank than the mere number of outlinks. What I like about this approach is that one doesn’t need to consider the huge number of SEO factors which have (some effect on SERPs but) no effect on the number or impact of outlinks.

    PCA or SVD might already do the job, even though there are likely more adequate methods. If you can collect some serious data, I’d be proud to offer some assistance during the analysis…

    Helge Lüddemann
    contact@truefidelity.de

    • Carter Bowles

      Nice to hear from a fellow stats geek! I’m aware of the statistical techniques you’re recommending. At this point the sample size is small, and the distribution is definitely not Gaussian, so I’m considering this a pilot study even though I did find statistical significance. Should I get the opportunity, I’m more interested in investigating this with a semi-experimental approach if I find the time to investigate further. Even looking at correlated variables, there’s only so much you can do with an observational study. Thanks for the recommendations!

      • Wusch El Gehirn

        Thanks for this clarification. As I see, you had good reasons not to go further than you did with the analysis. Just in case anybody misunderstood me: What you call a “pilot study” deserves more kudos than the “data” and “conclusions” other self-proclaimed SEO experts share with the public. I am *very* curious about your upcoming projects and reports. Let me know then if I can help to make good things even better.

  • Niranjan Pande

    Thanks for publishing Outbound Links topic, i learnt a lot about this article

  • That’s a lot of info in a single post:) I think it’s a delicate balance you have to get a feel for and it can vary from website to website. I’ll never dispute actual data and research though, a great read thanks!