Why your linking tests suck

Every now and then someone in the SEO community plucks up the courage to say something completely irresponsible and stupid about links. Usually it’s something like, “SEO is all about links.” Sometimes it’s more along the lines of, “Here is a great place to get links.”

Lately, however, people have been “testing” links to see if a 2nd link will pass value. That is, if you embed two links to the same destination on a page and each uses different anchor text, will the second anchor text be passed to the destination as well as the first? Some people are claiming you won’t see the second link pass value.

To which I say: Oh, really. I guess all those 2nd links I’ve been getting value from don’t count. Of course, how silly of me.

SEOs in general conduct poor quality tests. I mean, I’ve written about how SEO tests usually don’t work because of basic SEO incompetence and how SEOs do their tests wrong.

SEOs love to test nofollow links and make extravagant false claims and SEOs can’t test nofollow links.

Every good SEO should understand that link analysis can be misled by many factors. That is, link analysis sometimes sucks.

There are fundamental concepts that should govern all attempts at link analysis and proven concepts about link analysis that help you understand which links should work and why they should work as well as why they sometimes don’t work as expected.

But you can boil it all down to a few points:

• SEOs don’t know how to isolate factors in their tests (you pretty much have to do it by statistical analysis)
• SEOs don’t provide adequate allowance for other factors in their test results
• SEOs don’t include temporal factors in their testing
• When it comes to testing, most SEOs don’t have a clue

A bad SEO test consists of placing a link on a page, or setting up a few pages that are “optimized” for some unique expression. The worst tests, in fact, are usually conducted on nonsense expressions with little to no content. Without an environment that replicates natural search results, your testing won’t show you anything useful for search engine optimization analysis.

For example, if you want to know whether an H1 header helps improve your relevance score, creating five pages that each emphasize relevance in a different way invalidates your test. If you decide, “Well, I can jumble up the text and just put an H1 on a page” that may seem to make your test more valid but you’re not considering the way search engines evaluate proximity and word order.

The most reliable test model for evaluating on-page factors is to position pages in natural search results. You need to use a test set of at least 5 to 10 pages (the more the better, but larger test sets require more time and effort). Move the pages as high into the search results without the on-page factors you want to test as you reasonably can.

Then walk away for three months. Give the search results time to stabilize. If your pages drop in ranking you either have to make adjustment or just wait until they hit whatever ranking they’ll reasonably, naturally settle to. You should see a page hit its natural ranking somewhere in the 3 to 6 month timeframe.

When you’ve recorded 4 weeks of approximately consistent search ranking (no week varies from the previous or next by more than 2 positions), you’re ready to conduct your test. Now add you the on-page factor you want to test to your test pages.

Wait 1 month. If you cannot see the page’s new on-page content in the search engine cache, your test may be pointless for lack of crawling/indexing. In Google’s Supplemental Results Index, for example, pages may be crawled and indexed quickly and not recrawled for a very long time (although Google claims to have been working on this lag time, there are other problems with Supplemental Index pages).

If you see a change in results within 4 weeks, you need to remove your on-page factor from your test pages and wait 1 more month. If you see a change in results within 4 weeks, you need to add the on-page factor back to the test pages. Wait another 4 weeks. If you see the same change in results as the first time you add the on-page factor, it’s now safe to reasonably conclude that the on-page factor influences search results.

Real SEO testing takes time and resources and most SEOs don’t invest either time or resources in the testing process. You can engage in parlor trickery and SEOfoolery. A lot of people do that, usually with their linking tests.

Simply embedding a link on a page doesn’t tell you anything useful, even if you see the link appear to pass anchor text. Why? Because you allow other factors to pollute your test environment. Let’s look at possible linking relationships. NOTE: I call a page trusted if it is capable of both passing and receiving value. I call a page trustworthy if it’s unindexed (probably new) but on a site that has a lot of indexed/trusted pages (more than 50%). Just because I say a page is untrusted does not mean you should assume it won’t pass or receive value.

1. You can link from an indexed trusted page to an indexed trusted page
2. You can link from an indexed trusted page to an UNindexed trustworthy page
3. You can link from an indexed trusted page to an indexed untrusted page
4. You can link from an indexed trusted page to an UNindexed untrustworthy page
5. You can link from an UNindexed trustworthy page to an indexed trusted page
6. You can link from an UNindexed trustworthy page to an indexed untrusted page
7. You can link from an UNindexed trustworthy page to an UNindexed trustworthy page
8. You can link from an UNindexed trustworthy page to an UNindexed untrustworthy page
9. You can link from an indexed untrusted page to an indexed trusted page
10. You can link from an indexed untrusted page to an indexed untrusted page
11. You can link from an indexed untrusted page to an UNindexed trusted page
12. You can link from an indexed untrusted page to an UNindexed untrusted page
13. You can link from an UNindexed untrusted page to an indexed trusted page
14. You can link from an UNindexed untrusted page to an indexed untrusted page
15. You can link from an UNindexed untrusted page to an UNindexed trusted page
16. You can link from an UNindexed untrusted page to an UNindexed untrusted page

You don’t want to test all these different conditions at once (although if you want to try, you’ll need — in my opinion — at least 80 test pages but you cannot be sure what’s trusted and untrusted in your indexed pages).

You should never use a page to test more than one hypothesis at a time. That is, don’t perform two tests on a page at the same time.

Nurturing test pages is a bit like grooming a line of laboratory test rats over many generations. You have to know the pedigree of your test subjects. If you don’t know that then your conclusions are invalid because you don’t have any way of knowing why a test produces a certain result.

You have to be certain that you’re changing a stable page. You have to have a firm idea (based on solid evidence) of whether the pages you’re linking to can receive value, are stable, and don’t have other factors contributing to their possible relevance for a particular expression.

Just dropping links on pages at random tells you nothing. If you look at 10 different queries and use links to affect the results in 6 or more of of those queries (at least twice — remember the do-undo-redo component is vital for SEO testing), you have a statistical basis for concluding that your test results are predictable. In SEO testing it is absolutely mandatory that you statistically isolate the factor you’re testing in order for your test to be valid.

When you conclude that a cause-and-effect relationship exists, the established predicatability doesn’t mean you’ll always see the same results in natural search behavior. After all, in natural search behavior you’re looking at many different factors coming together at once. The search engines and the searchers have to figure out how they want the results to look. All you can do is put enough factors together (both on-page and off-page) and hope that you’ve met the challenge with sufficient resources.

Anyone can slap a link on a page and point to apparent results and say, “See! This proves my theory!” But they’re not testing or proving theories. They’re just practicing SEO parlor tricks, which serve no useful purpose (except maybe to attract links from gullible bloggers).

And that’s not all. But I’ll leave the rest of the discussion for another time. I’m sure there will be plenty of opportunities to comment again on faux SEO link tests in the future.