How Search Engines Operate

1. Crawling the Web
   Search engines run automated programs, called "bots" or "spiders", that use the hyperlink structure of the web to "crawl" the pages and documents that make up the World Wide Web. Estimates are that of the approximately 20 billion existing pages, search engines have crawled between 8 and 10 billion.
2. Indexing Documents
   Once a page has been crawled, its contents can be "indexed" - stored in a giant database of documents that makes up a search engine's "index". This index needs to be tightly managed so that requests which must search and sort billions of documents can be completed in fractions of a second.
3. Processing Queries
   When a request for information comes into the search engine (hundreds of millions do each day), the engine retrieves from its index all the document that match the query. A match is determined if the terms or phrase is found on the page in the manner specified by the user. For example, a search for car and driver magazine at Google returns 8.25 million results, but a search for the same phrase in quotes ("car and driver magazine") returns only 166 thousand results. In the first system, commonly called "Findall" mode, Google returned all documents which had the terms "car", "driver", and "magazine" (they ignore the term "and" because it's not useful to narrowing the results), while in the second search, only those pages with the exact phrase "car and driver magazine" were returned. Other advanced operators (Google has a list of 11) can change which results a search engine will consider a match for a given query.
4. Ranking Results
   Once the search engine has determined which results are a match for the query, the engine's algorithm (a mathematical equation commonly used for sorting) runs calculations on each of the results to determine which is most relevant to the given query. They sort these on the results pages in order from most relevant to least so that users can make a choice about which to select.

Although a search engine's operations are not particularly lengthy, systems like Google, Yahoo!, AskJeeves, and MSN are among the most complex, processing-intensive computers in the world, managing millions of calculations each second and funneling demands for information to an enormous group of users.

Speed Bumps & Walls

Certain types of navigation may hinder or entirely prevent search engines from reaching your website's content. As search engine spiders crawl the web, they rely on the architecture of hyperlinks to find new documents and revisit those that may have changed. In the analogy of speed bumps and walls, complex links and deep site structures with little unique content may serve as "bumps." Data that cannot be accessed by spiderable links qualify as "walls."

Possible "Speed Bumps" for SE Spiders:

• URLs with 2+ dynamic parameters; i.e. http://www.url.com/page.php?id=4&CK=34rr&User=%Tom% (spiders may be reluctant to crawl complex URLs like this because they often result in errors with non-human visitors)
• Pages with more than 100 unique links to other pages on the site (spiders may not follow each one)
• Pages buried more than 3 clicks/links from the home page of a website (unless there are many other external links pointing to the site, spiders will often ignore deep pages)
• Pages requiring a "Session ID" or Cookie to enable navigation (spiders may not be able to retain these elements as a browser user can)
• Pages that are split into "frames" can hinder crawling and cause confusion about which pages to rank in the results.

Possible "Walls" for SE Spiders:

• Pages accessible only via a select form and submit button
• Pages requiring a drop down menu (HTML attribute) to access them
• Documents accessible only via a search box
• Documents blocked purposefully (via a robots meta tag or robots.txt file)
• Pages requiring a login
• Pages that re-direct before showing content (search engines call this cloaking or bait-and-switch and may actually ban sites that use this tactic)

The key to ensuring that a site's contents are fully crawlable is to provide direct, HTML links to each page you want the search engine spiders to index. Remember that if a page cannot be accessed from the home page (where most spiders are likely to start their crawl), it is likely that it will not be indexed by the search engines. A sitemap (which is discussed later in this guide) can be of tremendous help for this purpose.

Measuring Relevance and Popularity

Modern commercial search engines rely on the science of information retrieval (IR). That science has existed since the middle of the 20th century, when retrieval systems powered computers in libraries, research facilities, and government labs. Early in the development of search systems, IR scientists realized that two critical components made up the majority of search functionality:

Relevance - the degree to which the content of the documents returned in a search matched the user's query intention and terms. The relevance of a document increases if the terms or phrase queried by the user occurs multiple times and shows up in the title of the work or in important headlines or subheaders.

Popularity - the relative importance, measured via citation (the act of one work referencing another, as often occurs in academic and business documents) of a given document that matches the user's query. The popularity of a given document increases with every other document that references it.

These two items were translated to web search 40 years later and manifest themselves in the form of document analysis and link analysis.

In document analysis, search engines look at whether the search terms are found in important areas of the document - the title, the meta data, the heading tags, and the body of text content. They also attempt to automatically measure the quality of the document (through complex systems beyond the scope of this guide).

In link analysis, search engines measure not only who is linking to a site or page, but what they are saying about that page/site. They also have a good grasp on who is affiliated with whom (through historical link data, the site's registration records, and other sources), who is worthy of being trusted (links from .edu and .gov pages are generally more valuable for this reason), and contextual data about the site the page is hosted on (who links to that site, what they say about the site, etc.).

Link and document analysis combine and overlap hundreds of factors that can be individually measured and filtered through the search engine algorithms (the set of instructions that tells the engines what importance to assign to each factor). The algorithm then determines scoring for the documents and (ideally) lists results in decreasing order of importance (rankings).

Information Search Engines Can Trust

As search engines index the web's link structure and page contents, they find two distinct kinds of information about a given site or page - attributes of the page/site itself and descriptives about that site/page from other pages. Since the web is such a commercial place, with so many parties interested in ranking well for particular searches, the engines have learned that they cannot always rely on websites to be honest about their importance. Thus, the days when artificially stuffed meta tags and keyword-rich pages dominated search results (pre-1998) have vanished and given way to search engines that measure trust via links and content.

The theory goes that if hundreds or thousands of other websites link to you, your site must be popular, and thus, have value. If those links come from very popular and important (and thus, trustworthy) websites, their power is multiplied to even greater degrees. Links from sites like NYTimes.com, Yale.edu, Whitehouse.gov, and others carry with them inherent trust that search engines then use to boost your ranking position. If, on the other hand, the links that point to you are from low-quality, interlinked sites or automated garbage domains (aka link farms), search engines have systems in place to discount the value of those links.

The most well-known system for ranking sites based on link data is the simplistic formula developed by Google's founders - PageRank. PageRank, which relies on a mathematical formula (based around finding a given document in a random pattern of clicking on links), is described by Google in their technology section:

PageRank relies on the uniquely democratic nature of the web by using its vast link structure as an indicator of an individual page's value. In essence, Google interprets a link from page A to page B as a vote, by page A, for page B. But, Google looks at more than the sheer volume of votes, or links a page receives; it also analyzes the page that casts the vote. Votes cast by pages that are themselves "important" weigh more heavily and help to make other pages "important."

Google uses a PageRank “proxy” value, which logarithmically translates the actual PageRank of a document to a value between 1 and 10, to rank Web sites listed in its directory (which offers a PageRank order or an Alphabetical order for listings) and in its toolbar (below).

Google's toolbar (available here) includes an icon that shows a PageRank value from 0-10

PageRank is, in essence, a rough system for estimating the value of a given link based on the links that point to the host page. Since PageRank's inception in the late '90s, more subtle and sophisticated link analysis systems have taken the place of PageRank. Thus, in the modern era of SEO, the PageRank measurement in Google's toolbar, directory, or through sites that query the service is of limited value. Pages with PR8 can be found ranked 20-30 positions below pages with a PR3 or PR4. In addition, the toolbar numbers are updated only every 3-6 months by Google, making the values even less useful. Rather than focusing on PageRank, it's important to think holistically about a link's worth.

Here's a small list of the most important factors search engines look at when attempting to value a link:

• The Anchor Text of Link - Anchor text describes the visible characters and words that hyperlink to another document or location on the web. For example, in the phrase "CNN is a good source of news, but I actually prefer the BBC's take on events," two unique pieces of anchor text exist - "CNN" is the anchor text pointing to http://www.cnn.com, while "the BBC's take on events" points to http://news.bbc.co.uk. Search engines use this text to help them determine the subject matter of the linked-to document. In the example above, the links would tell the search engine that when users search for "CNN", SEOmoz.org thinks that http://www.cnn.com is a relevant site for the term "CNN" and that http://news.bbc.co.uk is relevant to "the BBC's take on events". If hundreds or thousands of sites think that a particular page is relevant for a given set of terms, that page can manage to rank well even if the terms NEVER appear in the text itself (for example, see the BBC's explanation of why Google ranks certain pages for the term "Miserable Failure").

• Global Popularity of the Site - More popular sites, as denoted by the number and power of the links pointing to them, provide more powerful links. Thus, while a link from SEOmoz may be a valuable vote for a site, a link from bbc.co.uk or cnn.com carries far more weight. This is one area where PageRank (assuming it was accurate) could be a good measure, as it's designed to calculate global popularity.

• Popularity of Site in Relevant Communities - In the example above, the weight or power of a site's vote is based on its raw popularity across the web. As search engines became more sophisticated and granular in their approach to link data, they acknowledged the existence of "topical communities"; sites on the same subject that often interlink with one another, referencing documents and providing unique data on a particular topic. Sites in these communities provide more value when they link to a site/page on a relevant subject rather than a site that is largely irrelevant to their topic.

• Text Directly Surrounding the Link - Search engines have been noted to weight the text directly surrounding a link with greater important and relevant than the other text on the page. Thus, a link from inside an on-topic paragraph may carry greater weight than a link in the sidebar or footer.

• Subject Matter of the Linking Page - The topical relationship between the subject of a given page and the sites/pages linked to on it may also factor into the value a search engine assigns to that link. Thus, it will be more valuable to have links from pages that are related to the site/page's subject matter than those that have little to do with the topic.

These are only a few of the many factors search engines measure and weigh when evaluating links. For a more complete list.

Link metrics are in place so that search engines can find information to trust. In the academic world, greater citation meant greater importance, but in a commercial environment, manipulation and conflicting interests interfere with the purity of citation-based measurements. Thus, on the modern WWW, the source, style, and context of those citations is vital to ensuring high quality results.

The Anatomy of a Hyperlink

A standard hyperlink in HTML code looks like this:

<a href="http://www.seomoz.org">SEOmoz</a>

In this example, the code simply indicates that the text "SEOmoz" (called the "anchor text" of the link) should be hyperlinked to the page http://www.seomoz.org. A search engine would interpret this code as a message that the page carrying this code believed the page http://www.seomoz.org to be relevant to the text on the page and particularly relevant to the term "SEOmoz".

A more complex piece of HTML code for a link may include additional attributes such as:

<a href="http://www.seomoz.org" title="Rand's Site" rel="nofollow">SEOmoz</a>
 

In this example, new elements such as the link title and rel attribute may influence how a search engine views the link, despite its appearance on the page remaining unchanged. The title attribute may serve as an additional piece of information, telling the search engine that http://www.seomoz.org, in addition to being related to the term "SEOmoz", is also relevant to the phrase "Rand's Site". The rel attribute, originally designed to describe the relationship between the linked-to page and the linking page, has, with the recent emergence of the "nofollow" descriptive, become more complex.

"Nofollow" is a tag designed specifically for search engines. When ascribed to a link in the rel attribute, it tells the engine's ranking system that the link should not be considered an editorially approved "vote" for the linked-to page. Currently, 3 major search engines (Yahoo!, MSN, & Google) all support "nofollow". AskJeeves, due to its unique ranking system, does not support nofollow, and ignores its presence in link code. For more information about how this works.

Some links may be assigned to images, rather than text:

<a href="http://www.seomoz.org/randfish.php"><img src="rand.jpg" alt="Rand Fishkin of SEOmoz"></a>
 

This example shows an image named "rand.jpg" linking to the page - http://www.seomoz.org/randfish.php. The alt attribute, designed originally to display in place of images that were slow to load or on voice-based browsers for the blind, reads "Rand Fishkin of SEOmoz" (in many browsers, you can see the alt text by hovering the mouse over the images). Search engines can use the information in an image-based link, including the name of the image and the alt attribute to interpret what the linked-to page is about.

Other types of links may also be used on the web, many of which pass no ranking or spidering value due to their use of re-direct, Javascript, or other technologies. A link that does not have the classic <a href="URL">text</a> format, be it image or text, should be generally considered not to pass link value via the search engines (although in rare instances, engines may attempt to follow these more complex style links).

<a href="redirect/jump.php?url=%2Fgro.zomoes.www%2F%2F%3Aptth" title="http://www.seomoz.org/" target="_blank" class="postlink">SEOmoz</a>

In this example, the redirect used scrambles the URL by writing it backwards, but unscrambles it later with a script and sends the visitor to the site. It can be assumed that this passes no search engine link value.

<a href="redirectiontarget.htm">SEOmoz</a>

This sample shows the very simple piece of Javascript code that calls a function referenced in the document to pull up a specified page. Creative uses of Javascript like this can also be assumed to pass no link value to a search engine.

It's important to understand that, based on a link's anatomy, search engines can (or cannot) interpret and use the data therein. Whereas the right sort of links can provide great value, the wrong sort will be virtually useless (for search ranking purposes).

Keywords and Queries

Search engines rely on the terms queried by users to determine which results to put through their algorithms, order, and return to the user. But, rather than simply recognizing and retrieving exact matches for query terms, search engines use their knowledge of semantics (the science of language) to construct intelligent matching for queries. An example might be a search for loan providers that also returned results that did not contain that specific phrase, but instead had the term lenders.

The engines collect data based on the frequency of use of terms and the co-occurrence of words and phrases throughout the web. If certain terms or phrases are often found together on pages or sites, search engines can construct intelligent theories about their relationships. Mining semantic data through the incredible corpus that is the Internet has given search engines some of the most accurate data about word ontologies and the connections between words ever assembled artificially. This immense knowledge of language and its usage gives them the ability to determine which pages in a site are topically related, what the topic of a page or site is, how the link structure of the web divides into topical communties, and much, much more.

Search engines' growing artificial intelligence on the subject of language means that queries will increasingly return more intelligent, evolved results. This heavy investment in the field of natural language processing (NLP) will help to achieve greater understanding of the meaning and intent behind their users' queries. Over the long term, users can expect the results of this work to produce increased relevancy in the SERPs (Search Engine Results Pages) and more accurate guesses from the engines as to the intent of a user's queries.

Sorting the Wheat from the Chaff

In the classic world of Information Retrieval, when no commercial interests existed in the databases, very simplistic algorithms could be used to return high quality results. On the world wide web, however, the opposite is true. Commercial interests in the SERPs are a constant issue for modern search engines. With every new focus on quality control and growth in relevance metrics, there are thousands of individuals (many in the field of SEO) dedicated to manipulating these metrics in order to control the SERPs, typically by aiming to list their sites/pages first.

The worst kind of results are what the industry refers to as "search spam" - pages and sites with little real value that contain primarily re-directs to other pages, lists of links, scraped (copied) content, etc. These pages are so irrelevant and useless that search engines are highly focused on removing them from the index. Naturally, the monetary incentives are similar to email spam - although few visit and fewer click on the links (which are what provide the spam publisher with revenue), the sheer quantity is the decisive factor in producing income.

Other "spam" results range from sites that are of low quality or affiliate status that search engines would prefer not to list, to high quality sites and businesses that are using the link structure of the web to manipulate the results in their favor. Search engines are focused on clearing out all types of manipulation and hope to eventually achieve fully relevant and organic algorithms to determine ranking order. So-called "search engine spammers" engage in a constant battle against these tactics, seeking new loopholes and methods for manipulation, resulting in a never-ending struggle.

This guide is NOT about how to manipulate the search engines to achieve rankings, but rather how to create a website that search engines and users will be happy to have ranking permanently in the top positions, thanks to its relevance, quality, and user friendliness.

Paid Placement and Secondary Sources in the Results

The search engine results pages contain not only listings of documents found to be relevant to the user's query, but other content, including paid advertisements and secondary source results. Google, for example, serves up ads from its well-known AdWords program (which currently fuels more than 99% of Google's revenues), as well as secondary content from its local search, product search (called Froogle), and image search results.

Below is a screenshot of Google's search engine results page. Hover on any of the areas of the image to reveal the source of the content:

The sites/pages ranking in the "organic" search results receive the lion's share of searcher eyeballs and clicks - between 60-70%, depending on factors such as the prominence of ads, relevance of secondary content, etc. The practice of optimization for the paid search results is called SEM, or Search Engine Marketing, while optimizing to rank in the secondary results requires unique, advanced methods of targeting specific searches in arenas such as local search, product search, image search, and others. While all of these practices are a valuable part of any online marketing campaign, they are beyond the scope of this guide. Our sole focus remains on the "organic" results, although links at the bottom of this paper can help direct you to resources on other subjects.

     Back             Next