SEO Halo Effect

Question

I have a new SEO company doing some work with offpage optimization. They came up with several keyword phrases that they will be building backlinks to. All of the keywords have "0" monthly searches. They say it's ok that they have "0" monthly searches because what they are trying to do ..is to get the search engines to crawl my site and move some of my existing keywords (that do have monthly searches) up in Google and other search engine rankings. They say this is called the Halo Effect. This is the first time I am hearing of this and I have done some of my own SEO, worked with other companies, and nothing like this was ever mentioned. I thought a major determining factor in selecting keyword phrases was the monthly searches. They also said that Google Keyword Planner data monthly searches is only counting PPC advertisng? Is the Halo effect really true? Will what they are trying to do really work? Input appreciated. Thanks.

Answer 1

Yeah. This is soooooo 2010. Fire your SEO.

This is based upon a premise that is only partly true. The fact that your SEO is talking about keywords tells me that they do not know what they are doing.

Case in point.

Google has never made direct search term matches. Any apparent match is incidental to the process of semantic analysis which is very advanced these days. It only takes a short read of the original research paper to know that matching terms was never the process. What appears as keyword matches is simply the process of highlighting terms used in the search query as the final step of the result process while building the HTML for display. Nothing more.

How it all works briefly.

Google fetches a page from your site. It breaks down the page into HTML DOM elements. As more pages are fetched, Google looks for patterns that distinguish the content from templated elements such as headers, footers, and sidebars. The DOM elements that are content elements as well as a few tags such as the title tag, description meta-tag, image tags, link tags, etc. are semantically analyzed. All HTML DOM objects receive an ID that allows them to be reconstructed in order as they appear in the original HTML. This not only allows Google to order elements but also relate them in a parent child relationship. Parents maintain a significance above child elements meaning that they are more important.

Semantics, explaining it simply, is the understanding of the written language. Several algorithms are applied to each element and scores assigned per algorithm into a matrix for that element. In some cases, boosts in the scores are assigned depending on criteria. For example, the single most important HTML element on a page is the title tag. Next is the description meta-tag, and then the one and only h1 header tag. Each of these will boost the semantic scores of the semantic analysis for that HTML element according to weights given for the element. As well, the order of the HTML elements, relationship of the HTML element, position to important sections of content such as the first paragraph following a header tag, the first paragraph, the last paragraph, etc., are all given additional weight to the semantic scores.

In a semantic analysis, sentences are broken down into (again) elements. Semantics analyzes sentences much like you did in high school. The simplest example would be Bob threw the ball. where Bob is the subject, threw would be the predicate, and ball would be the object. From more complex sentences, the sentences are broken down into segments where elements are weighed according to precedent. For example Bob threw the ball to Sally would represent two sets of subjects, predicates, and objects, however, Bob would be seen as more important than Sally. As each HTML DOM element is analyzed, the terms found, the element type (subject, predicate, object, and there are more than just these) are indicated, and a score for each algorithm assigned in a table much like a spreadsheet. Included in the matrix, potentially, are terms that are relationally similar such as synonyms, plural and non-plural versions, etc. Also included are topic scores, industry scores, experience level scores, grade level scores, etc.

Even more interestingly, content is broken down into content segments. A simple example would be the paragraphs between header tags would belong to the header tag directly above them. Again order plays a role in importance making the first paragraph following a header tag more important than the second. This applies to sentences. As well term positions from these key points and from the beginning of the content are maintained so that term relationships are also retained. This would be recognized as phrases typically. Matrix scores are overlaid to further analyze content. For example the matrix for the h1 tag would be overlaid over the paragraph immediately following it with the matching scores added. This allows semantic analysis to understand the similarities and dissimilarities between related content segments.

This is what is stored in the search index for the page.

When someone searches for something using Google, RankBrain does nearly the exact same process for the search query. It breaks the search query into a matrix that is then matched against scores within the index. What takes precedence is not always terms, but topics, expertise, industry, grade level, etc. It is not terms that are being matched. It is the value of the content itself.

So what does this have to do with links?

Simple. Links are one of the top semantic indicators of content. It is fairly reliable.

Okay explain this more.

Semantic analysis has value when there is something to analyze. It prefers a sentence. If, for example, Bob ball was all you had, then the question would be What about Bob?, and What about ball? Full sentences, even simple ones yield better results. In the case of Bob ball, there is almost nothing that can be determined.

Have you ever wondered why some sites and pages can rank for terms that do not appear? Simple. Semantics. Where this becomes important is that semantics allows an engine to not only make intelligent linguistic link graphs, but also determine what is missing from content and where another content relates in that regard. Over simplifying it completely, a search for car for sale will also find automobiles. But it goes much further. It can also analyze relationships between named characters within content and analyze the relationships not specified. For example, Bob's cousin Rebecca has a sister Sally. What is not said is that Bob has a cousin Sally. As these relationships are graphed, they can be searched upon even when they do not exist.

Back to links and Halos.

What your SEO is trying to do, without actually knowing what they are doing, is boost term weights for a page, and yes all full pages of content also have a single matrix, by influencing important semantic clues for that page. Each link is analyzed and a matrix indexed. This sounds good right? Yeah. Wait.

What matrices can also do.

Analyzing content into semantic weights allows an engine to extremely quickly determine what does not fit. It also allows an engine to determine terms that are overly used, or synonyms that are overly used, language that is not natural, etc. In other words, links that appear to be on the margins of similarity are very likely to be determined to be unnatural. Holy sheep dip Batman! Are you telling me that Google can separate B.S. from honest content? Yes Robin. I am.

Any link created must have a subject, predicate, and object to have any value. Single terms only yield one sets of scores. Sad really. Two terms, well, not much better. However, where semantics can sink it's teeth into analysis is where the magic really begins.

The term long-tail is largely junk.

There really is no such thing as long-tail keywords. As you have probably figured out already, the whole notion of keywords is silly. Your dog makes simple word associations, do you think Google with all of it's PhD's could do better? I hope so. Long-tail, simply means terms that are less searched, but more common. What they really are are semantic clues that are being utilized between analysis and search. Nothing more. So the next time someone mentions long-tail, have some fun and ask them to explain it.

So what is the danger?

Unnatural weighting of semantic clues that would otherwise have yielded un-muddied search results. Simple. They are actually doing more harm than good unless they can explain all that I have about semantics and how what they are doing semantically boosts how your page is found. Are we taking bets at this point??

Answer 2

Ranking for related searches works - but not for the reasons they've stated. There's a semantic engine in Google that breaks down the meaning of your content and relates it to keywords. For an example of how this works, you can try IBM's Alchemy API:

http://www.alchemyapi.com/products/demo/alchemylanguage

Look at the difference between 'keywords' (literal word content) and 'entities'(or subjects). The relation between these is the art of semantics. You could theoretically have an entire article about 'a famous black president' without using 'Obama' once - yet the engine will piece together the keyword patterns and figure out what your subject is. This is how Watson answered all those Jeopardy questions and what the Alchemy API is based on.

Google does the same but on a grander scale. Attempting to rank for related queries is well intentioned but not entirely correct. It's more important to add lengthy, wordy, well informed content that's directly related to your main focus. Running a site on Mattresses? Write an article on bed bugs or turning mattresses. Writing an article focused on the keyword 'bed frame materials' and attempting to back link it more heavily than your main content will look instantly unnatural and could actually lower our rankings.

Edit: And as if by magic, Moz.com has just done an article that includes my link.