Thread by @willcritchlow

I wanted to talk this week about one of the most advanced SEO subjects there is: internal linking

The big takeaways from our testing are:

1. Internal links are important (these tests are some of our most consistently statistically significant)

2. We have surprisingly often found big benefits to the pages where the links are added, not just to the targets of those links

3. In fact, we have seen cases where recipient page impacts were undetectable or even negative, while the linking pages saw a benefit

4. Some of our hardest-to-predict test results have also been internal linking tests!

Here’s the breakdown of some specific @SearchPilot internal linking tests:

1. Firstly, and hopefully reassuringly, we have a test showing that, on a website without systematic cross-linking among regional pages, adding cross-linking was strongly positive:

searchpilot.com/resources/case…

2. Sometimes, the benefit doesn’t come from exactly where you think it will (or only from where you think it will).

This test showed a 90% confidence level uplift in total, but when we broke the analysis down, we found that the bigger uplift / most confidence was the benefit to the linking pages rather than the linked-to pages:

searchpilot.com/resources/case…

The recipient pages saw no detectable uplift, and in fact, if anything, the change was net negative for those pages

3. That wasn’t the only test to show benefit to the linking pages, though this one had benefits for both the linking pages and the target pages:

searchpilot.com/resources/case…

4. We recently published a surprising test result showing an inconclusive (but, if anything, negative impact) result from pre-rendering links that were previously only present in JavaScript:

searchpilot.com/resources/case…

It’s hard to be sure, but if that impact was negative, it is most likely due to those links not having previously been rendered by Google, and when they were included in the HTML, complex iterative link equity effects reducing the authority of the linking pages

5. If you’re interested in how SEO A/B testing works, my colleague @craigbradford wrote up a useful post here:

searchpilot.com/resources/blog…

But for internal linking tests, there is extra complexity because the impact & effect might be not only on the pages where the changes were made, but also potentially on the page types receiving additional links, or even elsewhere on the site

We manage this with what we call “cross site section measurement” that can run our neural network model over different groups of pages and site sections:

searchpilot.com/resources/blog…

This can require a lot of custom analysis, which is one of the benefits of our professional services team:

searchpilot.com/features/profe…

It’s worth the effort, though, because internal linking tests are some of our most consistent winners on very large sites

Finally, a bit of theory...

You might be wondering why I said that this was one of the most advanced SEO subjects there is. Internal linking doesn’t sound hard. We know that links are good. More links are better than fewer links. We control our internal links, so let’s put some more links around the place.

The problem is that

1. The equity a page has to distribute is divided among its outlinks and

2. Link authority algorithms are iterative, and changes flow through the system in complex and hard-to-understand ways

As a result, it’s impossible to be sure in advance things like A. Whether a new outlink from an important page will benefit that page itself

Or B. Whether the benefits of distributing link equity to a new group of pages will outweigh the relative loss of equity to others

We have seen this in our SEO testing @SearchPilot, where internal linking tests have shown some of the biggest proportions of our respondents getting it wrong (see for example this test, where 2/3 of respondents got it wrong: searchpilot.com/resources/case… )

A few years back, I wrote up some thoughts on the challenges of talking coherently about internal linking, and how we might work towards making better recommendations:

searchpilot.com/resources/blog…

Even if we did crack that problem, we’d still be left with all the normal “SEO is hard” problems - most specifically in this case, that we don’t know what Google’s link algorithms reward, nor how they are weighted:

searchpilot.com/resources/blog…

which is why we're all-in on testing

Want more of the kinds of insights I've shared here, based on statistically-controlled tests? We have a low-volume email list for you:

marketing.searchpilot.com/seo-abtesting-…