Related questions are questions that Google shows next to the results of a query, chosen from questions users have already searched and matched to the topics of the top results. Patent US 9,213,748 B1, "Generating related questions for search queries," describes how they are selected. Google filed it on March 14, 2013, and the patent was granted on December 15, 2015.
What does patent US 9,213,748 describe?
Patent US 9,213,748 describes a question engine that selects, ranks and displays questions related to a search query. Its 5 inventors are Yossi Matias, Dvir Keysar, Gal Chechik, Ziv Bar-Yossef and Tomer Shmiel. The patent holds 18 claims and 6 drawing sheets, and a continuation with the same title, US 9,679,027 B1, was granted on June 13, 2017.
The patent's example is the query "Lichen planus": the results page shows 2 search results and a block of related questions under a title. Each question is a link that runs a search for a query derived from the text of the question. The questions can be sent with the search results or in place of them, presented as an answer box ("a formatted presentation of content that is relevant to the query"), or shown only after the user selects a "show frequently asked questions" element. The patent does not use the name "People Also Ask"; the block it describes resembles that feature.
Where do the related questions come from?
The related questions come from a question database. The database holds "search queries that have previously been submitted by users" and "that have been determined to be in question form." A query is in question form when it contains question terms (interrogative words, other function words used to ask a question, or a question mark), or when it matches a question template such as "why is [X] used."
The database can keep only queries submitted "more than a threshold number of times." It can include questions from other sources, such as "question and answer websites," where phrases are identified as questions in the text.
What is a topic set?
A topic set is a past query that led users to click a given result page. The patent calls these queries "qualified search queries": a query is qualified for a page when "at least one user has selected a search result that identifies the search result resource" after submitting it. A selection can be a click, a voice selection or a tap on a touch screen, and "selection data may not be tethered to the identity of individual users."
The system builds topic sets in 3 steps:
- Identify the qualified queries for each selected result page.
- Rank them by the number of times each query was submitted, or by the number of times users selected the page after submitting it.
- Keep the highest-ranked queries as the topic sets of the page.
The system can filter out a topic set selected for fewer than a threshold number or share of the result pages. In the patent, a page's topic sets therefore come from the searches after which people select it, not from the text of the page.
How does Google select the related questions?
Google selects the related questions in 6 steps:
- Take the top results of the query, using the scores provided by the search engine: a predetermined number of the highest-scoring pages, or every page whose score exceeds a threshold.
- Compute their topic sets from the qualified queries.
- Find matching questions: any question in the database "that includes each term included in a given topic set."
- Rank the questions by the number of times each one has been submitted as a query.
- Remove equivalent questions: when 2 questions are equivalent, the lower-ranked one leaves the ranking.
- Select the related questions: a predetermined number of the highest-ranked questions, or every question submitted to the search engine more than a predetermined number of times.
The system can also create additional matching questions from templates. The patent's example: if a topic set is a disease name A, the template "What is the cure for [disease]" generates "What is the cure for A?"
How does Google detect equivalent questions?
Google detects equivalent questions with a question graph. 2 questions are equivalent "if they have an equivalent meaning." Each question of the database is a node, and 2 nodes are linked when the questions share at least a threshold number of "common selected resources": pages that users selected after searching each question. The patent's example: at least 1 or 3 common pages among the 3, 5 or 10 most-selected pages of each question. The system can count only the pages that received more than a threshold number of selections.
Instead of, or in addition to, these clicks, the system can submit each question as a query and compare the results. The patent's example: 2 questions are linked when at least 2 of the 10 highest-ranked results of one question identify the same resource as a result among the 10 highest-ranked results of the other.
When a question is linked to a higher-ranked one, the lower question is removed, and its submission count can be added to the remaining question. The system can replace a question with its "best variant": the equivalent question "most frequently submitted to the search engine," or the form given by a template, such as "how old is [X]" for queries of the form "what is the age of [X]." After removing equivalent questions or choosing best variants, the system can re-rank the questions and look for equivalent questions again.
Do answers change which questions appear?
Yes: a question with a good answer can be promoted. When the system has data that associates questions with answers, "questions with no answer can be demoted in the ranking or questions with answers can be promoted." The answers are "derived from the text of a resource," text determined to answer the question, and can be pre-selected answers shown as a formatted presentation. A single question can have several answers.
The patent lists 4 factors for the quality of an answer:
- The quality score of the page the answer comes from.
- The ranking of that page when the question itself is searched.
- The length of the answer, in tokens, terms or characters.
- The agreement with other answers: "the number or proportion of terms in the answer that are repeated in other answers for the question."
This patent does not describe how the answer text is extracted from a page. Google describes the selection of answer passages in another patent, covered in the answer passage scoring patent. The answer criteria appear only in the description: none of the 18 claims mentions answers.
What do the patent's claims protect?
The patent's claims protect the selection of related questions through topic queries. Its 18 claims include 3 independent claims (a method, a system and a computer storage medium) that each require the same steps: receive a query, obtain its results, select for each result page "topic queries" among the past queries that led users to select it, select from the question database past queries in question form that match at least one topic query, and send these questions with the response.
The dependent claims add:
- Ranking the topic queries by submission count or by selections of the page (claims 2, 10 and 16).
- Removing a question when a higher-ranked question is equivalent to it (claims 3, 11 and 17).
- Replacing a question with its best variant, the most frequently submitted equivalent question (claim 4).
- Building the question graph from common selected resources (claims 5, 6, 12 and 13).
- Ranking the questions by the number of times each was submitted as a query (claims 7, 14 and 18).
- Generating questions from a template filled with topic queries (claim 8).
Why does Google show related questions?
Google shows related questions to help users explore a topic and find the right words for it. Among the benefits the patent lists, 3 matter here: users "gain a quick understanding of the major knowledge areas in a newly-explored topic," users with "un-common keywords or terminology" discover the terms "more commonly used to describe their intent," and selecting a question returns "a pre-determined, pre-formatted answer."
What does the related questions patent change for your SEO?
The patent means that the questions shown for a query are the popular questions of the topics that its top results already cover. 5 consequences follow:
- Mine related questions as a content plan. They are past searches, ranked by popularity, attached to the topics of the pages that rank: a direct map of what your audience asks.
- Answer each question under its own heading. A question with an answer can be promoted, and the answer comes from the text of a page. A heading in question form followed by a short answer matches both this patent and the heading context patent.
- Use the most searched wording. Among equivalent questions, the lower-ranked one is removed, and the system can replace a question with its best variant, for example the most frequently submitted one. Use the most searched wording in your heading.
- Earn clicks for the right queries. Your page's topic sets are the top-ranked queries after which users select it, and a question matches a topic set when it contains each of its terms. The queries that bring clicks to your page decide which questions it can bring into the block.
- Write answers that agree with good sources. Answer quality can depend on the share of terms repeated in other answers: state the established facts, then add what is new.
Our SEO content strategy starts from the related questions shown for a topic, then gives each one its own heading and a short answer.
The patent describes what Google's system can do. It does not confirm how People Also Ask selects its questions today.