Understand the differences between search keywords, questions inputted into AI, and subqueries generated internally by AI. Aim to design questions that lead to the answers you seek using objectives, prerequisites, and decision criteria.
Understanding AI Search Queries is Easier When Separating Input and Internal Processing
An AI search query is a natural language question inputted to request information gathering, comparison, or suggestions from AI. It can also refer to search queries generated internally by AI during the answer creation process.
For example, if a user inputs "Compare attendance management services suitable for small businesses," this sentence itself is an AI search query. Meanwhile, AI might expand it into search terms like "attendance management small business comparison," "attendance management implementation cost," or "attendance management law amendment compliance" to find the necessary information for an answer. The latter are subqueries generated internally by AI, not questions directly inputted by the user.
When understanding AI search queries, it's necessary to distinguish between search keywords, questions inputted into AI, and AI's internal subqueries. Treating all three as the same can blur whether you want to investigate user needs, improve questions, or analyze AI's information retrieval.
The flow of AI searching for information and constructing answers is explained in detail in the related article Illustrated AI Search Mechanism: The Process Until Source Selection.
Three Types of Queries Used in AI Search
A "query" means an input requesting search or processing from an information system. However, in the context of AI search, the target changes depending on who generates it and where it is inputted.
Unified Diagram of Three Types of Queries
-
Search Keywords
-
Generator: Searching users or SEO personnel
-
Input Destination: Search windows like Google
-
Purpose: To find related pages or information
-
Example: "attendance management comparison"
-
Questions Inputted into AI
-
Generator: Users utilizing AI
-
Input Destination: Dialogue screens like ChatGPT, Google AI Mode
-
Purpose: To obtain answers, comparisons, suggestions, or summaries according to conditions
-
Example: "Compare attendance management services suitable for a company with about 50 employees"
-
Subqueries Generated Internally by AI
-
Generator: AI search system
-
Input Destination: Search functions, databases, external sources
-
Purpose: To decompose the original question and gather necessary information for answers
-
Example: "attendance management 50 people cost," "attendance management small business case studies"
Search keywords are generally expressed in short words or compound terms. Questions inputted into AI tend to be natural language that includes objectives, situations, and conditions. Subqueries are often invisible to users and are used by AI to retrieve information.
In a broad sense, AI search queries may include both questions inputted by users and subqueries generated internally by AI. In a narrow sense, it only refers to questions inputted into AI by users. When using this term in documents or conversations, clarifying whether it means "input query" or "internally generated query" can avoid confusion.
Characteristics of AI Search Queries and the Reason for Using Conversational Questions
The characteristic of AI search queries is that they can input not only words but also the context contained in the entire question. AI associates the user, purpose, conditions, and criteria within the sentence to interpret the direction of the desired answer.
In traditional searches, inputs like "accounting software comparison" that combine main words are common. In AI searches, you can include the context needed for judgment in a single question, such as "Compare accounting software that is easy to use for first-time individual business owners, based on price and support system."
This question includes the following information:
-
User: Individual business owner
-
Usage stage: First-time introduction
-
Desired result: Comparison of multiple services
-
Comparison criteria: Price and support system
-
Important condition: Ease of use
This information serves as material for AI to determine the scope of the answer. The condition "individual business owner" narrows down the candidates, and "first-time introduction" increases the importance of operability and support system. "Price and support system" specifies the items to be compared in the answer.
In other words, the value of an AI search query is not in its length. It is characterized by the ability to relate multiple conditions in natural language and convey what kind of judgment you want to support.
Differences from Traditional Search Queries and Keywords
A search query is the phrase actually inputted by the user into the search system. Search keywords may refer to the main phrases contained in the search query or the themes targeted in SEO measures.
For example, the search query "I want to compare video production companies for corporations in Tokyo by cost" includes search keywords like "video production company," "corporate," "Tokyo," "cost," and "comparison."
The main difference from AI search queries is the expected processing after input.
In traditional searches, pages with high relevance to the search query are listed as search results. Users open multiple pages and make judgments while selecting the necessary information. In AI searches, information is searched, retrieved, and organized according to the question's intent, and it may be presented as a single answer text.
However, traditional search and AI search are not completely separate mechanisms. In Google's AI Overviews and AI Mode, AI may perform related searches and generate answers while referring to web pages. Google explains that in AI Overviews and AI Mode, query fan-out may be used to perform multiple related searches from a single question.Explanation of AI Features by Google Search Central
The relationship between search keywords and AI search queries is easier to understand when viewed as a hierarchy rather than a confrontation. Search keywords succinctly indicate the theme, while questions inputted into AI supplement the purpose and conditions for that theme. AI's internal subqueries further decompose the question into searchable units.
Types and Examples of AI Search Queries
The form of an AI search query changes depending on what the user wants to do. Deciding on the purpose before creating the question makes it easier to choose the necessary conditions.
Questions to Know Definitions or Meanings
When you want to grasp the outline of a term, specify the target word and the range you want to know.
Vague Example
"Tell me about RAG"
Organized Example
"Explain what RAG is to beginners, following the flow of how generative AI searches for external information and uses it in answers"
In the latter, the target audience and scope of explanation are clear. If internal algorithms are not needed, you can also indicate the excluded range by saying "excluding technical implementation methods."
Questions to Compare Multiple Options
In comparisons, it's important to include not only candidates but also evaluation criteria.
Vague Example
"What's a recommended CRM?"
Organized Example
"Compare CRM for B2B companies with less than 100 employees based on initial cost, operational burden, and sales support functions"
The word "recommended" alone doesn't determine what criteria to use for selecting candidates. Specifying users, scale, purpose, and comparison axes makes it easier to identify candidates that match your company's conditions.
Questions to Know Procedures
When asking about operations or implementation methods, indicate your current location and the state you want to reach.
Vague Example
"How to do access analysis?"
Organized Example
"I want to check visits to my company's site via AI search. Explain step by step from the initial setup of the analysis tool to the items to be checked"
If the usage environment is determined, add the tool name, the authority you have, and the data you can use. If the current location is indicated, it becomes easier to omit explanations of already completed tasks.
Questions to Consider Purchases or Implementations
In purchase considerations, convey not only the budget but also the usage conditions and necessary functions.
Organized Example
"I'm looking for a corporate AI search analysis tool that can be used within a monthly budget of 100,000 yen. Compare based on supported platforms, measurement items, and the presence of implementation support"
Since the budget, target company, and necessary functions are indicated, it's an easy question to narrow down candidates. If you use price or specifications as judgment materials, request to list the information at the time of the answer along with the official page to facilitate the confirmation process.
Questions to Confirm Sources
If the basis of the answer is important, specify the type of source you want.
Organized Example
"Regarding the usage trends of generative AI search, organize it prioritizing primary information where you can confirm the research entity, implementation year, and number of subjects"
The expression "reliable information" alone is vague. By specifying the type of source needed, such as official documents from public institutions, service providers, or academic papers, it becomes easier to judge the acceptance of information.
Questions May Be Expanded into Subqueries Internally by AI
The question inputted by the user is not always used for a single search as it is. AI search systems may divide complex questions into multiple issues and search for related information for each.
Google explains about AI Mode that it breaks down questions into multiple subtopics and performs multiple searches simultaneously, known as query fan-out.Explanation of AI Mode by Google
For example, the question "Compare hotels suitable for families with children in Osaka based on distance from the station, breakfast, and budget" includes the following investigation points:
-
Family-friendly hotels in Osaka
-
Nearest station and walking distance
-
Contents of children's breakfast
-
Accommodation fees and target dates
-
Cancellation conditions
-
User evaluations
AI may generate search queries corresponding to these issues internally and integrate the obtained information to provide an answer. Even if the user inputs one question, AI may split the exploration into necessary facts for comparison.
This information retrieval may utilize a configuration called RAG (Retrieval-Augmented Generation). RAG is a mechanism that does not rely solely on the model's internal knowledge (pre-training) to provide answers but retrieves related information from external web pages or databases and uses that content for answer generation.Explanation of RAG by Google Cloud
RAG involves chunk splitting to divide documents into certain units, embedding to convert the meaning of sentences into numerical expressions, vector search to find similar information, retrieval to obtain candidates, and re-ranking to adjust the order of candidates.
What users should check is not the technical syntax of subqueries but into which issues the original question is divided. If issues like comparison targets, costs, subjects, and implementation conditions are lacking, the range of information AI refers to may also be biased.
The results and details of QFO's investigation are summarized in the related article 〖The Largest Query Fan-Out Survey in Japan〗AI Searches Up to 33 Times Behind One Question.
The Focus of Answers Changes Between Good and Vague Questions
A good AI search query is a question that conveys the necessary information for the purpose without excess or deficiency. It's not about making the question longer, but designing it in the order of purpose, prerequisites, and decision criteria.
Set a Single Purpose
First, decide what you want from AI.
-
Understand the meaning of words
-
Compare options
-
Know the implementation procedure
-
Find candidates that meet conditions
-
Confirm the basis of information
If you include definitions, comparisons, procedures, and future predictions in one question, the focus of the answer will be dispersed. If the purposes differ, separate the questions and ask additional questions based on the initial answer.
Add Prerequisites Within the Necessary Range
Prerequisites are information to narrow down answer candidates. Choose those that affect judgment from users, company size, usage scenes, budget, target regions, implementation timing, etc.
For example, adding conditions like "50 employees," "multiple locations," and "clocking in with smartphones" to the question "What is the recommended attendance management service?" clarifies the necessary functions.
The criterion for increasing conditions is whether the information changes the answer candidates or conclusions. Excluding backgrounds that do not affect the conclusion makes the central request easier to convey.
Specify Decision Criteria
In comparisons or selections, convey not only the candidates to be raised by AI but also the axes for evaluating candidates.
"The cheapest one" alone mixes initial costs, monthly fees, pay-per-use, and additional function costs. Specifying "compare initial costs and monthly fees separately" makes it easier to confirm the cost structure.
If you can't think of decision criteria, you can request "list the main decision criteria first and compare candidates based on those criteria."
Specify an Output Format That Can Be Confirmed
The output format is specified not only to make the answer easier to read but also to find missing judgment materials.
If a comparison of each candidate is necessary, request to align the same items for each candidate. If you want to know the procedure, request the implementation order and the completion conditions for each stage. If you want to verify the basis, request to indicate the reference source for each claim.
Separate Confirmed Information from Unverified Information in the Answer
Information such as prices, laws, service specifications, and coverage areas may change. When such information is included, request to separate "content that can be confirmed with official information" and "content that cannot be judged due to lack of information."
This specification allows you to grasp not only the items AI could answer but also the items that require additional confirmation. In the final judgment, confirm unverified items on the official page or with the providing company.
Investigate AI Search Queries and Apply to Content Design
The purpose of companies investigating AI search queries is not to increase specific words in articles. It is to understand in what situations customers compare and what conditions they check.
Collect Questions for Each Stage of Customer Behavior
Gather the situations where customers actually hesitate from inquiry history, questions during business negotiations, site searches, and interviews with sales representatives.
Questions are divided into stages of behavior such as information gathering, comparison consideration, candidate selection, implementation decision, and post-use confirmation. Even for the same "price" question, the necessary answer differs between the "cost market" in the information gathering stage and the "additional costs not included in the estimate" in the implementation decision stage.
The specific procedure for associating customer behavior stages with questions is summarized in How to Automatically Generate Customer Journey Maps with AI: Visualizing Customer Buying Behavior by Simply Entering a URL.
Record User Input and Internally Generated Queries Separately
Manage the question candidates presented to users and the phrases searched internally by AI as separate items. Mixing the two makes it impossible to distinguish between what the customer directly asked and the issues AI supplemented for information gathering.
Record items are divided into service used, inputted question text, internally generated query, answer content, source, and confirmation date. If the internally generated query could not be obtained, leave it blank and do not substitute with assumed investigation issues.
Evaluate Answers and Sources as a Set
Check not only whether your company name is included in the AI answer but also which page was referenced as the basis and whether the quoted content matches the original page.
Evaluation is divided into the following three to easily identify areas for improvement:
-
Mention: Whether the company name or service name appears in the answer
-
Quotation: Whether your company's page is shown as a source
-
Content Match: Whether your company's features and conditions are correctly explained
Even if there is a mention, if the source is a competitor's site, there is a possibility that your company's information is indirectly understood. If quoted but the content is outdated, it's necessary to review the page update or information placement.
If you want to investigate internally generated queries, you can check the search queries that can be obtained from the inputted question using the Gemini Query Fan-Out Analysis Tool.
Decide the Role of Pages for Each Search Intent
Presenting detailed implementation procedures or case studies to readers who want to know definitions at once makes it difficult to find the desired information. If the central questions differ, such as term definitions, comparisons, implementation procedures, and case studies, separate the roles of pages and connect them with internal links.
If you want to connect contact points via AI search to inquiries, refer to How Inquiry Routes Change with the Spread of AI Search: Strategies to Achieve Results Even with Reduced Inflow.
Query Handling Differs by AI Search Service
There is no common format for AI search queries. Even if the ability to ask questions in natural language is common, the conditions for executing web searches, the data referenced, the method of displaying sources, and the range of conversation history usage differ by service.
In Google AI Mode and AI Overviews, Google's search mechanism and AI-generated answers are combined. Google explains that both functions may use query fan-out to construct answers from multiple related searches.
ChatGPT Search searches the web according to the question content and presents answers that include links to related information sources. Users can also choose the search function.Explanation of ChatGPT Search by OpenAI
When using services separately, check the following items:
-
Whether web search can be used to investigate the latest information
-
Whether links to the basis of the answer are displayed
-
Whether it supports inputs other than text, such as images or files
-
Whether past conversation content is carried over to the next question
-
Whether only internal documents or specified data can be the search target
Even with the same question text, if the search target and available information differ, the answer will change. When comparing results from multiple services, record not only the question text but also the search function, execution date, and the model or settings used.
In internal AI searches, internal documents, contracts, and business data may be the search targets instead of the public web. In this case, the judgment axis is the range of viewing permissions, document update dates, and search targets rather than the expression of the answer.
In tasks dealing with personal information or confidential information, check the usage regulations, contract contents, and conditions for saving input data. Whether you can input questions is judged not by the accuracy of AI's answers but by the internal information management standards.
Common Questions About AI Search Queries
Are AI Search Queries and Prompts the Same?
There are overlapping parts, but they do not always mean the same thing.
A prompt refers to instructions or inputs given to generative AI. It includes requests that do not involve search, such as text creation, translation, summarization, and image generation. An AI search query is an expression that refers to inputs requesting search, comparison, confirmation, or suggestions within that.
Is It Better for AI Search Queries to Be Longer?
What matters is whether the conditions necessary for the answer are included, not the length.
For simple term confirmation, a short question can fulfill the purpose. In comparisons or implementation decisions, add conditions that affect the conclusion, such as users, purposes, budgets, and comparison axes.
Is It Necessary to Include Search Keywords in AI Search Queries?
Include major search keywords within the range where the question's target is conveyed. There is no need to unnaturally line up related words.
By clearly indicating the target as in "Compare attendance management services," and then adding company size or necessary functions, it becomes a natural question.
Can Internally Generated Subqueries by AI Be Confirmed?
The range that can be confirmed varies depending on the service or function. Sometimes related search terms are displayed, and sometimes internal processing is not disclosed.
In investigations, record "queries obtained from screens or tools" and "issues assumed from the question content" separately. This distinction allows analysis without confusing measured data with hypotheses.
Can the Search Volume of AI Search Queries Be Investigated?
It is difficult to confirm the number of searches across all AI services using the same method as traditional search keywords.
This is because AI service providers may not disclose input data, and expressions of questions may vary even for the same purpose. When judging demand, combine not only the number of searches but also customer questions, business negotiation records, mentions in AI answers, and changes in sources.
Summary
An AI search query is a natural language question inputted to request information gathering, comparison, or suggestions from AI. However, depending on the context, it may also refer to subqueries created internally by AI for answer generation.
The first distinction to make is between search keywords, questions inputted by users, and subqueries generated internally by AI. Search keywords indicate the theme, questions convey the purpose and conditions, and subqueries are used by AI to gather information.
When creating questions, first decide on one purpose, then add prerequisites and decision criteria that affect the conclusion. Specify an output format that makes it easy to confirm results, such as comparison items or sources.
If you want to organize the relationship between SEO and AI search response, it is explained in detail in the related article What is SEO? Easy-to-Understand Explanation of Causes of Ineffectiveness and Measures in the AI Search Era.
Related Articles to Connect AI Search Queries to Practical Design
Even if you understand the structure of AI search queries, you may be unsure about what to prepare on your own site. Proceed to content design and measurement/improvement practices from the following articles.
-
Content Design Method for Companies Wanting to Use Both SEO and LLMO
-
Practical Steps for Redesigning Owned Media with Indicators Other Than PV
-
Priority for Small Companies with Few Brand Searches to Start LLMO
-
Practical Strategy Guide to Become Information Selected in Search Results with AI Overview Tools
-
Judgment Axis for Tool Selection to Start Measuring AI Overview Citations
-
Explanation of the Timeline and Measures to Accelerate the Effectiveness of LLMO Measures
-
Period Judgment Axis for Companies Unable to Predict the Timing of LLMO Results