How Scoring Each Candidate Answer Passage (CAP) Is Done

Understand how to win featured snippets by utilizing the information inside Google’s patent - context scoring adjustments for answer passages.

Patent Overview

Google's patent - context scoring adjustments for answer passages describes systems used to find the most likely candidate to answer a user’s query. Once a group of candidate websites has been gathered, they are compared and weighed against one another according to the systems described in the patent. The candidate with the highest score will win the featured snippet.

Important Terms & Phrases This Patent Mentions

What Is a Candidate Answer Passage

A candidate answer passage (CAP) is a passage in your article that directly answers a user’s question. You can have tens, twenties, dozens, or hundreds of them in each article depending on how complex and in-demand your topic might be.

What Is Passage Coverage Ratio

Your PCR or passage coverage ratio is a metric used by Google to determine how much “fluff” your article contains. In this context, your PCR focuses on how much fluff is in your CAP. If you can effectively answer a user’s query in two sentences, your CAP should be 2 sentences. The more unnecessary words you add, and the more off-topic you begin to wander, the more you will hurt your rankings.

What Is The Adjusted Answer Score

An adjusted answer score is the score once the candidate answer passage has been evaluated by the methods described in this article. Once all of these scoring methods have been carried out for all candidate answer passages, the candidate with the highest adjusted answer score will win the snippet.

Scoring Based On Heading Vectors

There are three ways heading vectors are scored:

1. Scoring based on the vector depth (how deep the candidate answer passage is found in the heading hierarchy)
   1. The deeper the parent heading of the CAP, the larger the scoring bonus the document will receive (contrary to popular belief)
2. Scoring based on how similar the user’s query is to the actual text of the headings in the heading vector
3. Scoring for best-matched sets of headings

According to the patent itself, the deeper in your heading structure the answer to a user's query is - the greater the increase in your adjusted answer score. This seems contrary to popular belief that you should answer a user's query immediately under an H1 or H2. The reason this seems to be the case, is because the additional layers in your heading structure will provide additional context and relevance.

Scoring Based On Passage Coverage Ratio

The patent details a process in which the passage coverage ratio (PCR) evaluates the text surrounding the CAP and determines how well it covers topics relevant to the user’s query. To make it more simple and to provide an actionable step, you should avoid useless filler words, fluff, or other irrelevant terms and phrases. Strive to make every sentence, phrase, word, and character as concise but thorough as possible.

If some text in your document doesn’t really add context, doesn’t really expand context, or doesn’t really improve a user’s understanding of the topic at hand, you probably shouldn’t include it. The way the patent describes PCR, the more words required to fulfill a user’s query, the more your PCR will be lowered. The candidate webpage with the highest PCR will have the largest boost to its context score.

Scoring Based on Other Features

Now, these are three different scoring “schemes” discussed in the patent. Not all will be used at once supposedly. The patent states that “Three example features” are to be described. This implies there are potentially more features apart from the three we cover below.

Distinctive Text:

Distinctive text is described in US Patent 9,959,315 B1 as any text “formatted to be visually different from other text of a passage from which a candidate answer is selected.”

Meaning, underlined text, bold text, text wrapped in a span and highlighted in a different color, etc. This is considered distinctive text.

It goes on further and says that if distinctive text is discovered, it will add it to the heading vector, and that it will thereby increase your heading depth by 1 (if the distinctive text scheme is being used to score your CAP).

Basically what this means then, is that you should strive to make your distinctive text an expansion - in terms of context - of the parent heading. So for example:

(H2) What Dog Biscuits Should Diabetic Dogs Eat?
‍
Dogs suffering from diabetes should eat dog biscuits high in protein so that they can prevent insulin spikes.

This callout doesn’t depict the perfect example of distinctive text, but it gives you an idea of how it should be used to further expand the context of its parent heading.

You should try your best to use the distinctive text to lead in to your CAP so that it's depth can be expanded by 1 - further increasing your adjusted answer score and improving your odds of winning the snippet.

Preceding Questions:

A preceding question is described in the patent as a question in the text that precedes your CAP. In the dog biscuit callout example given above, an example of a preceding question would be the mach heading “(H2) What Dog Biscuits Should Diabetic Dogs Eat?”

When you have a header that is most relevant for your preceding question, make sure your CAP is subordinate to that header. By subordinate, I mean that the CAP is in the text underneath that header. Any text in the same section as the header is considered subordinate. So any H3s that might follow that H2 would technically be considered subordinate. But, the text underneath each of those H3s would be subordinate to those H3s, not the H2.

Lists

If lists are detected, this patent states that there are systems that will detect the list quality. The list quality is detected by two concepts.

1. If the list is in the center of the page
   1. Typically lists off to one side or another of a page are indicative of some sort of navigational (reference) content, as opposed to informational content.
2. If a high percentage of the text in the list consists of links it will be disqualified and therefore be unable to benefit from this boost
   1. Tables of contents are an example of lists that would be disqualified
   2. From what the patent says, you can have some links in your list and still have it qualify for this boost, but you should ensure that the overwhelming majority of the content in the list is NOT link text

Key Takeaways

Based on the ending conclusions of the patent, the PCR is by far the most important metric in this adjustment. So basically, if you gather nothing else from this, make sure you avoid using filler words when writing your content. Stay concise, on topic, and make content that builds upon itself to further expand and break down the topics being covered on the page. Below we'll provide a list of actionable steps that can be utilized to win featured snippets based on the information given by this patent:

1. Scoring Based on Heading Vectors - Create logically structured and ordered headings that follow an order that is helpful even for a complete novice on the topic
   1. ‍If there is a primary question you need to answer, include it further down your heading structure (H3, H4, instead of H1 or H2)
2. Passage Coverage Ratio - Avoid filler words - get to the point and be direct. Make sure each sentence - and ideally each word - directly relates to the answer you are providing
3. Distinctive Text - For each section, provide your answer in bold text
4. Preceding Questions - Practice using questions for your headings as opposed to statements.
5. Lists - Practice utilizing lists whenever applicable. Avoid using lists that are filled with links to other pages or resources, as the patent mentions that these lists are not counted.
   1. You can include links in your list, but ensure that the links don't make up the majority of text in the list.

One of the biggest reasons you should be reading patents just like this is to expand your understanding of how search engines work. If you want to learn more about this patent and read it for yourself, you can download it here.