How does Google Pick Search Snippets for Your Pages to Show in Results?

There are often three pieces of information about pages displayed in search results to searchers in response to a search:

  1. Page title,
  2. The URL where that page can be found, and;
  3. A summary of the page in the form of a snippet or snippets, taken from either a meta description tag, or a description of the page from a directory like the DMOZ, or actual text from the page itself.

One mystery involving search engines involves how a snippet might be generated when it is taken from a page.

Snippet Folklore

I’ve heard a mix of ideas from commentators on the subject of snippet selection, including the search engine grabbing text from the first mention of a keyword, or paying more attention to whether the keyword is used in a heading, or is bolded, or draws attention in some other manner.

It’s hard to say if any of those things is really a consideration to search engines.

New Google Patent on Snippets

A patent granted to Google today describes some of the process behind the choosing of text from a page to summarize the content of that page in relation to the keywords that it was found for in a search.

It doesn’t tell us everything about snippets – another unpublished Google patent is referenced that may describe more. But it does talk about some of the assumptions behind choosing a snippet, and some of the ways that they may be created and chosen.

Methods and systems for generating textual information
Invented Niniane Wang and Stephen R. Lawrence
Assigned to Google
US Patent 7,310,633
Granted December 18, 2007
Filed March 31, 2004

Abstract

Methods and systems for generating textual information are disclosed. In one exemplary embodiment, a method of generating textual information is disclosed that comprises identifying a plurality of candidate summaries related to textual information based at least in part on a document, determining first and second attribute values based at least in part on the candidate summaries, and determining an optimal candidate summary based at least in part on the first and second attribute values.

Importance and Size of Snippets

How important are snippets to the search process? The authors of the patent tell us that:

As efficient as a search engine may be, its value to a user may be limited by the manner in which the search engine provides a summary of search results to a user. For instance, search engines generally provide summaries (sometimes referred to as “snippets”) of documents or websites located in response to a query.

A user browses such summaries, and typically selects a link associated with a summary that best matches the search criteria to view the entire document or to navigate to the desired web page.

Choosing the right text to use to describe a page can be essential to how well the search engine delivers results. One issue involved in the selection of text in a snippet is the size of the snippet itself.

Summaries that provide too much information can consume output (e.g., display) resources and can overwhelm a user with extraneous information, which can slow down the user’s search.

Summaries that provide too little information may not provide the user with sufficient information to identify relevant documents. In either case, such summaries are generally ineffective in aiding a user’s search for desired information.

Search Engine Snippet Creation Considerations

Some things to keep in mind in the snippet creation process:

Snippet Size – One of the main goals in the creation of a snippet is that it not be too long that people won’t read it, or too short that it isn’t helpful.

Smaller Snippets for Mobile Devices Possible – A search engine might keep a snippet down to a minimum size, and might even generate a shorter snippet if the screen being used is small, like on a smart phone or PDA connected to the Web.

Some Snippets Possibly Pre-Generated – For some queries, snippets might be created ahead of time and stored, with the numbers of candidate snippets based upon the size of the documents. Others might be generated dynamically, so that the optimal snippet can be created just before being presented in search results.

Searcher Control over Snippet Size – The patent filing mentions the possibility of a searcher deciding on how large or small a snippet they might want to see – that could be interesting if a search engine added a feature like that.

Snippet Storage – When a snippet is pre-generated, a larger document generally will include a greater number of storage locations for candidate snippets than a comparatively smaller document, and a query with more keywords will generate greater number of storage locations for candidate summaries. It’s also possible that the number of storage locations may be limited to a predetermined number regardless of the size of the document.

Keywords in Snippets – The textual information in a snippet will depend upon the keywords found in a document. The character string presented to searchers may include an entire sentence or portions of the sentence including the keyword. It might also include a predetermined number of words or characters disposed before and after the keyword.

Things get a little more interesting when there is more than one keyword, or the words in a keyword phrase are not close to each other. Candidate text for keywords may be combined together.

Selecting Snippets from Candidate Summaries

When snippets are taken from text on pages, amongst the possible snippet candidate summaries, the first and second highest-weighted combined candidate summaries are selected, and can be merged together to show to searchers (three might also be combined).

They are weighted possibly using one or more factors, which may be further described in the unpublished patent application, Methods and Systems for Processing Textual Information.

This patent does describe some ways that the weighting of candidate snippets could be done, but the descriptions aren’t extremely clear. Here are a couple of them:

– A weight might be adjusted based at least in part on the percentage of keywords included in a candidate snippet, so that a combined candidate which includes a greater number of the query keywords can have a higher adjusted weight than a combined candidate summary which includes a lesser number of query keywords.

– A weight might be adjusted by a number of words in the candidate snippet, so that possible snippets with more words can have a higher weighting.

So, a snippet might be chosen based upon percentage of keywords, or by the amount of meaningful textual information that might be contained within the candidate snippets.

While this patent offers a tantalizing peek at snippet creation, it doesn’t go into a great deal of depth on the topic. It’s quite possible that the related, and unpublished, patent filing that it refers to a couple of times will tell us more. Guess we will have to wait and see…

Conclusion

A paper jointly created by A9 and Yahoo provides some research that they conducted on snippets, and is worth a look. I wrote about it in Search Result Snippets and the Perception of Search Quality.

Google also discussed what makes a good snippet for a product review, including topics like readability and punctuation of text, in another patent application on generating product descriptions. It’s also worth considering. I wrote about that in Google Reviews: Reputation + Quality + Snippets + Clustering.

A takeaway from this patent may be that you should pay attention carefully to the text that surrounds and supports phrases on your pages that you think might be terms that people will search for to find those pages, and that may rank well and show up in search results that people will see.

My post on product description snippets provides some ideas on how to try to do that well.

Share

12 thoughts on “How does Google Pick Search Snippets for Your Pages to Show in Results?”

  1. Finally Google shows us how it does that. This could get interesting, especially since we know the specifics now it may be easier to control what Google shows for a site. Imagine the possibilities!

    Thank You

  2. You’re welcome, Diddy1.

    I will be keeping an eye out for that other patent filing. Hopefully we’ll see it sometime soon, and it will give us a little more substance.

    I do like how this patent lays out the competing objectives of using snippets with lots of keywords versus snippets that are longer and possibly better descriptions of content that appears upon a page.

  3. Thanks, Liz.

    There is a rich culture of talk and discussion about search engines. Folklore seemed like a good choice of words when I was giving that section a title. I think I’ll be using it more.

    I’m always pretty excited when I come across something in a patent filing or whitepaper that reinforces some of the observations and assumptions that I’ve seen made on forum posts or blogs.

  4. Sheesh, I just had the most pathetic thing happen to me with a Google Alert for my site. My META Description Tag said “Your body text has to persuade your target market as well as the search engines to believe your site is the most relevant to the query.”

    But, the snippet in the Google Alert said “Body text. Making it keyword rich There are many well written pages on the web. Some with truly captivating text. You know the kind. It’s the ones that have you screaming, “Beat me, whip me, take away my charge cards!”

    I was thinking “Great move Kimberly. That ought to bring a mature audience”…….

  5. What a great article! This is a question that we get a lot from our customers as they demand that certain things been displayed in their natural results. For the most part, all we can really tell them is to create the best possible descriptions and the search engines do the rest. However, with this information it is a little more apparent as to how these snippets might be extracted.

  6. It is absolutely fascinating how Google and other such search engines pick the snippets of code. Often times, I’ll perform a search just to look at the results. Many times I can’t even find the snippet of text that Google pulled within the page, which just cracks me up! Many of my customers also ask me about what helps them rank, and how the search engines pull these “snippets?” It just goes to show what an inexact science our industry can truly be! Loved the article FYI…

  7. Hi Jacob,

    Thanks.

    I agree, it is fascinating. Google may pull that snippet from the page, or the meta description, but doesn’t have to. But there are so many possibilities, and mysteries. That’s part of why I like looking through the patent filings. Sometimes they provide some interesting clues.

Comments are closed.