GEO・AI検索 用語集
生成エンジン最適化（GEO）、AIブランドモニタリング、そして大規模言語モデルがどのブランドに言及するかを決める仕組みに関する用語を、わかりやすく解説します。
Agentic Search
Agentic search is research delegated to an AI agent. Instead of answering one question, the agent plans a task (compare vendors, find the best price, shortlist tools), runs many searches, reads the pages itself, and returns a decision. The human sees the conclusion, not the browsing.
AI Assistant
An AI assistant is the product wrapped around a language model: the chat interface, the web search integration, the memory, the app. ChatGPT, Claude, Gemini, Perplexity, and Copilot are assistants. When people say 'AI told me', they mean an assistant's answer.
AIブランドモニタリング
AIブランドモニタリングとは、AIアシスタントが自社ブランドをどう語るかを継続的に確認する取り組みです。言及されているか、正確に説明されているか、AIの回答の中で競合とどう比較されるかを追跡します。
AI引用
AI引用とは、アシスタントが質問に答える際に示す情報源です。モデルが自社ドメインを引用すれば、単に言及されるだけでなく、根拠そのものになります。これはAIの回答で最も強い立ち位置です。
AI Crawler
An AI crawler is a bot that fetches your web pages for an AI system. Some collect content for model training (GPTBot, ClaudeBot, CCBot), others retrieve pages live so an assistant can answer with current information (OAI-SearchBot, PerplexityBot). Whether you allow them in robots.txt shapes whether AI systems can learn about and cite your site.
AI Mode
AI Mode is Google's conversational search tab. Instead of ten blue links, it generates a full answer with Gemini, supports follow-up questions, and pulls sources through a technique called query fan-out, where one question is expanded into many sub-queries searched in parallel.
Google AI Overviews
Google AI Overviewsとは、多くの検索で従来の結果の上に表示されるAI生成の要約です。問いに直接答え、いくつかの情報源を引用するため、注目とクリックの行き先が変わります。
AI Referral Traffic
AI referral traffic is the visits your site gets from AI assistants: someone asks ChatGPT or Perplexity a question, the answer cites your page, and they click through. It shows up in analytics with referrers like chatgpt.com, perplexity.ai, and gemini.google.com.
AI Search
AI search is when a person asks an AI assistant a question and gets a synthesized answer, rather than a list of links to click. Tools like ChatGPT, Perplexity, Gemini, and Google's AI Overviews read across sources and return one response, often naming specific brands and citing some pages.
AIシェア・オブ・ボイス
AIシェア・オブ・ボイスとは、関連するAIの回答のうち、競合と比べて自社ブランドが言及される割合です。マーケターが従来追ってきた可視性指標の、AI時代版といえます。
AI Visibility
AI visibility is how often, how prominently, and how favorably a brand shows up in AI-generated answers. It covers whether an assistant mentions you at all, where in the answer you appear, whether your site gets cited as a source, and what the answer actually says about you.
アンサーエンジン
アンサーエンジンとは、選ばせるためのリンク一覧ではなく、直接の答えを返すシステムです。ChatGPT・Perplexity・Gemini・GoogleのAI Overviewsはいずれもアンサーエンジンのように振る舞います。
アンサーエンジン最適化（AEO）
アンサーエンジン最適化（AEO）とは、アンサーエンジンが明確で直接的な答えをそのまま取り出せるようコンテンツを構成する取り組みです。GEOと大きく重なり、質問と回答の形式をより重視します。
Answer Position
Answer position is where in an AI answer your brand shows up. Being the first name the assistant gives carries most of the value; a mention at the end of a list of seven carries little. It is the closest AI-search equivalent to a ranking position.
ブランド言及
ブランド言及とは、AIアシスタントが回答の中で自社ブランドを名指しする箇所です。GEOでは位置と語り口が重要です。推薦候補として最初に挙げられるのと、脚注扱いされるのとでは大きく異なります。
Chunking
Chunking is the process of splitting a page or document into smaller passages so a retrieval system can index each one and return only the most relevant part to an AI answer. Well-structured content chunks cleanly along clear headings and self-contained paragraphs, which makes the right passage easier to retrieve and quote.
Citation Share
Citation share is the share of AI answers, across a set of relevant buyer questions, that cite your domain as a source. If assistants answer 100 questions about your category and your site is cited in 12 of them, your citation share is 12 percent. It is the evidence-level counterpart to mention rate.
Content Freshness
Content freshness is how current a page is: when it was last updated and whether its facts still hold. Retrieval-backed AI search favors recent sources for anything time-sensitive, so a maintained page with a visible update date and a changelog beats an identical page that looks abandoned.
Context Window
A context window is the maximum amount of text a language model can hold in mind at once, measured in tokens. It includes the prompt, any retrieved sources, and the answer being written. When retrieved pages exceed the window, some content is dropped, so what fits, and where it sits, affects what the model uses.
Conversational Search
Conversational search is research done as a dialogue: you ask an assistant a question, get an answer, and refine with follow-ups, with the assistant remembering the thread. It replaces a series of separate keyword searches with one evolving conversation.
Digital PR
Digital PR is the practice of earning coverage on publications, newsletters, and industry sites: data stories, expert quotes, product coverage. For GEO it is load-bearing, because AI assistants assemble recommendations mostly from third-party sources, and digital PR is how your brand gets onto them.
E-E-A-T
E-E-A-T stands for Experience, Expertise, Authoritativeness, and Trustworthiness. It is the framework Google uses to judge content quality, and the same signals help AI systems decide which sources to trust and cite. Strong E-E-A-T makes a page more likely to be retrieved and quoted in an AI answer.
エンティティ（エンティティSEO）
エンティティとは、ブランド・製品・人物など、検索やAIが識別し推論できる独立した対象です。エンティティが明確なほど、安定して説明され推薦されます。
Fine-Tuning
Fine-tuning takes a trained model and continues training it on a focused dataset to shape its behavior: a support tone, a domain vocabulary, a task format. It is how companies customize models, and one of the ways assistant behavior changes between releases without a new base model.
Generative AI
Generative AI is software that produces new content rather than retrieving existing content: it writes text, generates images, drafts code. Large language models are the text-generating branch, and they power the assistants that now answer buyer questions directly.
Generative Engine
A generative engine is a search system that answers a question by writing the answer, not by listing links. It combines a large language model with retrieval: the engine fetches relevant sources, then generates a synthesized response that may cite a few of them.
生成エンジン最適化（GEO）
生成エンジン最適化（GEO）とは、AIアシスタントが自社ブランドをどう説明し、どう推薦するかを整える取り組みです。狙いはシンプルです。買い手がその分野の質問をAIにしたとき、自社ブランドが登場し、正しく語られ、推薦されることです。
GPTBot
GPTBot is OpenAI's crawler for gathering content that may be used to train its models. It is not the same as OAI-SearchBot, which fetches pages so ChatGPT can answer with live web content, or ChatGPT-User, which fetches a page when a user asks about it directly. Each can be allowed or blocked separately in robots.txt.
グラウンディング
グラウンディングとは、AIの回答を、学習記憶だけに頼らずモデルが取得した実在の情報源に結びつける手法です。根拠ある回答は引用を伴いやすく、まさにそこがブランドの居場所を獲得できる場です。
AIハルシネーション
AIハルシネーションとは、モデルが自信たっぷりに誤った回答を生成することです。ブランドにとっては、古い価格、提供していない機能、信頼を静かに損なう主張として現れ得ます。
Inference
Inference is the moment a trained model runs: it takes the prompt plus any retrieved context and generates the answer, token by token. Training happens rarely and bakes in knowledge; inference happens on every question and assembles the actual answer a buyer reads.
Information Gain
Information gain is what your page adds that the rest of the web has not already said: a new number, a firsthand test, an original dataset, a real example. Pages that only restate existing consensus add zero gain, and both search ranking and AI citation increasingly discount them.
Knowledge Cutoff
A knowledge cutoff is the point in time where a model's training data ends. A model with a January 2026 cutoff knows nothing that happened after that date on its own. Live AI search works around this with retrieval, fetching current web content at answer time.
ナレッジグラフ
ナレッジグラフとは、ブランド・人物・製品などのエンティティと、その間の関係を構造化したネットワークです。検索エンジンやAIは、ページ上の語句だけでなく、自社ブランドが何かを理解するために用います。
大規模言語モデル（LLM）
大規模言語モデル（LLM）とは、膨大なテキストで学習し、言語を予測・生成するAIシステムです。ChatGPT・Claude・Geminiなどを支え、自社ブランドがどう説明されるかを決める存在です。
LLM Optimization (LLMO)
LLM Optimization (LLMO) is the practice of shaping how large language models represent your brand, so that when a model answers a question in your category, it describes you accurately and includes you among its recommendations. It is a near-synonym for generative engine optimization, framed around the model itself.
llms.txt
llms.txtとは、ドメインのルートに置く簡素なファイルで、最も重要で機械可読なコンテンツへAIモデルを誘導します。robots.txtがクローラー向けであるのと同様に、アシスタント向けの親切な地図と考えられます。
Mention Rate
Mention rate is the share of relevant buyer questions where an AI assistant names your brand in its answer. Ask the questions your buyers actually ask, across ChatGPT, Claude, and Gemini, and count how often you appear. That percentage, tracked over time, is your mention rate.
プロンプト
プロンプトとは、人がAIアシスタントに与える質問や指示です。GEOでは、買い手が実際に使うプロンプトこそが勝つべきクエリであり、従来検索におけるキーワードに相当します。
Prompt Engineering
Prompt engineering is the practice of crafting the input you give an AI model to get a more accurate, relevant, or useful output. In brand monitoring, it is how you design the questions used to test what models say about your category, so results are consistent, realistic, and comparable over time.
Query Fan-Out
Query fan-out is how AI search systems research a question. Instead of running your query once, the system generates many related sub-queries, searches them all in parallel, and builds its answer from the combined results. Google describes AI Mode working this way.
検索拡張生成（RAG）
検索拡張生成（RAG）とは、AIが回答時に関連文書を取得し、それを根拠に回答を生成する手法です。最新で構造化されたコンテンツが、アシスタントの語る内容を動かせる理由がこれです。
robots.txt
robots.txt is a plain text file at yoursite.com/robots.txt that tells crawlers which parts of the site they may fetch. It has been a search engine convention for decades, and it now doubles as the control panel for AI: rules for GPTBot, ClaudeBot, PerplexityBot, and Google-Extended decide whether AI systems can learn about your brand and cite your pages.
セマンティック検索
セマンティック検索は、完全一致のキーワードだけでなく、クエリの背後にある意味と意図を一致させます。AIアシスタントが多様な言い回しの質問に答え、買い手の求めを理解できる理由です。
感情分析
感情分析とは、言及に伴うトーンが肯定・中立・否定のいずれかを判定する作業です。AIの回答では、登場の有無以上に、どう枠づけられるかが重要になり得ます。
構造化データ（スキーママークアップ）
構造化データは、多くの場合schema.orgマークアップで追加され、ページの各部分が何を意味するかを機械に伝えます。AIアシスタントや検索エンジンが、推測せずにブランドの事実を読み取れるようにします。
System Prompt
A system prompt is the standing instruction an AI assistant receives before any user input: who it is, how to format answers, when to search the web, how to cite sources. Users never see it, but it shapes every answer, including how brands get recommended and attributed.
Token
A token is the chunk of text a language model actually processes: a word, part of a word, or punctuation. As a rule of thumb, 1,000 tokens is about 750 English words. Model context windows, API pricing, and answer length limits are all counted in tokens.
Topical Authority
Topical authority is being the site that has genuinely covered a subject: the definitions, the how-tos, the comparisons, the data, the edge cases. Search engines use it to decide who ranks; AI systems reflect it in who gets retrieved and cited when questions in that subject come up.
Training Data
Training data is the corpus of text a language model learns from, largely web content gathered by crawlers, plus licensed and curated datasets. A model's default beliefs about your brand are a compression of what that corpus says, which is why what the web wrote about you two years ago can still shape AI answers today.
信頼ギャップ
信頼ギャップとは、言及されることと引用されることの差です。AIが自社を名指ししても、回答の根拠を競合の情報源で固めるなら、モデルは競合を証拠として信頼し、自社を二の次に扱っています。
Vector Embedding
A vector embedding is a list of numbers that represents the meaning of a piece of text, so AI systems can compare how similar two texts are. Retrieval systems embed both the user's question and candidate pages, then pull the pages whose embeddings sit closest in meaning, which is how relevant sources reach an AI answer.
Zero-Click Search
A zero-click search is a query the user gets answered without visiting any website, because the answer appears directly on the search or assistant surface. AI answers and AI Overviews accelerate this: the model resolves the question in place, so brands can be described or recommended without ever receiving a visit.