Gerris AIAI search visibility, AEO, GEO, and SEO

Home › Guides › AI search glossary

The AI search glossary: every acronym and buzzword, defined

AI search has produced more acronyms in three years than SEO produced in twenty. GEO, AEO, AIO, LLMO, AXO, SXO, GPTO, and a dozen more turn up in agency pitches, often for the same work under a different name. This glossary defines each one in plain language, says which describe a real method and which are relabels, and adds the technical terms and metrics that come with them. The definitions are checked against primary sources where one exists, listed at the end.

The short version: almost every acronym here means making a business easy for AI systems to find, understand, verify, and cite. Google's own documentation says there are no additional requirements or special optimizations needed to appear in AI Overviews or AI Mode, and no new machine-readable files or markup. Ask anyone who sells one of these acronyms to name the tasks, the evidence they will collect, and how the result will be measured. The tasks matter; the acronym does not.

The acronyms

Most of these name the same job: getting a business found, described correctly, and cited by AI systems. The differences are usually in who is selling the label.

AAIO (agentic AI optimization) and AXO (agentic experience optimization)
Preparing a website for AI agents that act for a person: comparing options, filling forms, booking, buying. The practical work is old-fashioned: real buttons and links, labeled form fields, prices and stock stated in text, and checkout steps a machine can follow. Accessible sites are already most of the way there.
AEO (answer engine optimization)
Work aimed at being the answer an AI assistant, voice assistant, or answer box gives to a question. Gerris has a full guide: What is answer engine optimization?
AI SEO and AISO (AI search optimization)
Broad labels for SEO adapted to AI search, or for SEO done with AI tools. The phrase covers both meanings in the wild, so it says little until the seller describes the actual tasks.
AIO
Used two ways. Some agencies mean AI optimization, an umbrella for all AI visibility work; others mean AI Overviews optimization, earning citations in Google's AI Overviews specifically. When someone sells AIO, the first question is which one.
ASO (agentic search optimization)
Optimizing to be found, evaluated, and acted on by AI agents that research, shortlist, and transact on a person's behalf, not only to be cited in an answer. The agent is the visitor: it compares options, checks prices and terms, and completes steps, so the site has to hold up to a reader that never sees the design. The label collides with App Store Optimization, which has used the same three letters for years, so expect confusion until one usage wins.
AVO (AI visibility optimization)
Another umbrella label for improving how often and how accurately AI answers mention a brand. Functionally the same work as GEO and AEO.
GAIO (generative AI optimization) and GSO (generative search optimization)
Alternate names for GEO, used by agencies that wanted a label of their own.
GEO (generative engine optimization)
Making a business easy for generative AI systems to find, understand, verify, and cite. The term comes from a research paper of that name by Pranjal Aggarwal and colleagues, first posted in November 2023 and accepted to KDD 2024, which reported that some content changes raised visibility in generative engine responses by up to 40 percent. Gerris explains the practice in What is generative engine optimization?
GPTO (ChatGPT optimization)
GEO aimed at one engine. Optimizing for a single assistant is a weak plan when buyers use several, and the underlying work is the same for all of them.
LLMO (large language model optimization) and LLM SEO
Shaping what large language models say about a brand: consistent facts, a clear entity, and outside sources a model will trust. In practice it overlaps GEO almost completely.
Search everywhere optimization
The idea that visibility now depends on many platforms at once, including YouTube, Reddit, TikTok, Amazon, and the AI assistants, along with Google. The observation is fair: AI answers draw heavily on forums and video. The label adds no new technique.
SEO (search engine optimization)
Making a site crawlable, indexable, and worth ranking in search engines. Every AI label above still depends on it, because AI systems retrieve from search indexes and crawled pages.
SEO 2.0
A marketing name for SEO that accounts for AI answers. It describes a sales pitch more than a method.
SIO (search intent optimization)
Matching pages to what a searcher is trying to do. This has been a core part of SEO for well over a decade under the name search intent.
SXO (search experience optimization)
Combining search visibility with the experience after the click: speed, clarity, and a page that answers the question. With AI answers, part of that experience now happens before any click, in the summary itself.
VEO (voice engine optimization)
Optimizing for spoken answers from voice assistants. The work is the same as AEO: short, direct answers in plain language, attached to a clear entity.

The machinery

These terms describe how AI search actually works. They are worth learning, because they explain what a site can and cannot influence.

Agentic browser
A browser with an AI agent built in that can carry out multi-step tasks for its user. OpenAI's ChatGPT Atlas and Perplexity's Comet are examples. They read pages through the same structure screen readers use, which is why semantic HTML matters to them.
Agentic commerce protocols
The plumbing that lets AI agents transact: presenting products, taking payment, and completing a checkout inside the conversation. Three open protocols divide the work. ACP (Agentic Commerce Protocol), from OpenAI and Stripe, connects an agent to a merchant's existing commerce systems and sits under instant checkout in ChatGPT. AP2 (Agent Payments Protocol), from Google, covers the payment itself, including proof that the buyer authorized the agent to spend. UCP (Universal Commerce Protocol) is an open standard for the whole transaction between agents, storefronts, and payment providers.
Agentic RAG
Retrieval run by an AI agent that decides for itself when to search again, which tools to use, and when it has enough to answer. Deep research features in the major assistants work this way.
AI crawler
A bot that fetches web pages for an AI company, either to train models or to answer questions live. OpenAI, for example, runs separate bots for each job and treats each robots.txt setting independently: blocking GPTBot (training) does not block OAI-SearchBot (ChatGPT search).
AI Overviews and AI Mode
Google's two generative search surfaces. AI Overviews is the summary at the top of some results pages, the successor to the Search Generative Experience test; AI Mode is the separate conversational search that uses query fan-out. Google says there are no extra requirements to appear in either beyond ordinary SEO, and their traffic is counted in Search Console's regular Performance report under the Web search type.
AIPREF
An IETF effort to standardize how publishers declare what AI systems may do with their content, separating training use from search and answer use. The preferences are expressed through robots.txt and related signals, so one statement covers the whole site instead of page-by-page negotiation. This site's robots.txt already carries AIPREF lines.
Answer engine
Any system that replies to a question with a written answer instead of a list of links: ChatGPT, Perplexity, Gemini, Copilot, Claude, and Google's AI Mode among them.
Chunking
Splitting a long document into shorter passages so a retrieval system can index and return each one separately. For a website, the lesson is that each section should make sense on its own, with a heading that says what it covers.
ClaudeBot, Claude-SearchBot, Claude-User, and Claude-Agent
Anthropic's four user agents. ClaudeBot collects content that may be used for training, Claude-SearchBot improves Claude's search results, Claude-User fetches a page when a person's question needs it, and Claude-Agent identifies the browsers Anthropic hosts for Claude to complete tasks for users. Each can be allowed or blocked separately in robots.txt.
Crawler economics
The emerging economics of AI crawling: what a site gives up, and what it gets back, when AI bots fetch its pages. Its parts now have names. Pay-per-crawl arrangements charge AI companies for access that used to be free. The crawl-to-referral ratio counts how much crawling a site absorbs for each visit a crawler sends back. RSL (Really Simple Licensing) is a machine-readable license a publisher attaches to content to state the terms of AI use. Cloudflare's Content Signals Policy is a robots.txt extension for declaring whether content may be used for search, for AI answers, and for training. Web Bot Auth lets an agent sign its requests cryptographically, so a site can verify which agent is actually asking.
Embedding (vector embedding)
A list of numbers that represents the meaning of a passage, so a system can find passages close in meaning to a question even when the words differ. Talk of measuring a brand's semantic distance from a prompt describes this math. No website owner can see or set those numbers directly.
Google-Extended
A robots.txt product token Google uses to let publishers decide whether crawled content may be used for training Gemini models and for grounding. Google states that it does not affect a site's inclusion in Google Search and is not a ranking signal.
GraphRAG
Retrieval over a knowledge graph of entities and relationships instead of, or alongside, similar text passages. Microsoft Research described it in an April 2024 paper by Darren Edge and colleagues: a language model extracts the entities and relationships from a collection, groups closely related entities into communities, summarizes each community, and builds answers from those summaries. It handles questions that span a whole collection or connect several sources better than passage retrieval, and adds little for single-fact lookups. A lighter variant, LazyGraphRAG (November 2024), brought indexing cost down to that of vector retrieval, 0.1 percent of full GraphRAG. It rewards sites whose facts about people, places, and organizations are stated the same way everywhere, because those facts become the nodes.
Grounding
Anchoring an AI answer in sources retrieved at the time of the question rather than in training memory alone. Grounded answers are the ones that carry citations, which makes grounding the moment a website can earn a link.
Grounding queries
The searches an AI system actually ran while assembling an answer, which often differ from what the user typed. Bing Webmaster Tools now shows a sample of them for pages it cited.
Hybrid retrieval
Finding passages two ways at once: keyword matching (the classic method, often BM25) and meaning matching through embeddings, then merging the results. It is why both the exact words buyers use and clear, plain explanations help a page get retrieved.
llms.txt
A proposed Markdown file at a site's root that summarizes the site and links its key pages for language models. It is a proposal, and AI companies have not committed to reading it, but it costs little and this site publishes one: gerris.ai/llms.txt.
MCP (Model Context Protocol)
An open protocol, introduced by Anthropic in November 2024, that lets AI assistants connect to outside tools and data sources in a standard way. For a business, MCP matters when it wants assistants to query its own systems, such as inventory or bookings, directly.
OAI-SearchBot
OpenAI's crawler for ChatGPT's search features. OpenAI says blocking it keeps a site out of ChatGPT search answers, though the site can still appear as a navigational link.
Query fan-out
Google's own term for how AI Mode works: it breaks a question into subtopics and issues many searches at once on the user's behalf, then writes one answer from the results. Google described it in May 2025. A page can be cited for a subquestion the user never typed.
RAG (retrieval-augmented generation)
A design in which a system first retrieves relevant passages from an index, then has a language model write its answer from them, usually with citations. The approach was named in a 2020 research paper by Patrick Lewis and colleagues. AI search assistants are RAG systems built on web indexes. Retrieval is the weak point: when the wrong passages come back, the answer is wrong with full confidence, which is why being the clearest, most retrievable source on a question matters more than being one of many.
Reranker
A second model that rescores the top passages a retriever found and puts the most relevant first, before the answer is written. Passages that answer the question directly tend to survive reranking.

Content and authority

These describe what makes a source worth citing. Most are new names for what good publishers and good PR have always done.

Answer blocks and passage-level optimization
Writing each section so it stands alone: a heading phrased as the question, and a direct answer in the first sentence or two beneath it. AI systems lift passages, and a passage that needs the paragraph before it to make sense gets skipped.
Attribute list
The short set of facts and differentiators a business wants AI answers to repeat about it: what it does, where, for whom, and what sets it apart. Writing one down is useful, because consistency across the web is what lets engines repeat it.
Brand co-occurrence
How often a brand's name appears near the topics, places, and descriptions it wants to be known for, on sites other than its own. It is the AI-era name for being mentioned in the right company, which is earned through press, reviews, and community, and cannot be faked at scale without risk.
Claim layer
Gerris' name for the structured statements a business publishes about itself so machines can read them directly: entity schema, FAQ markup, llms.txt, and one canonical description of what the business is.
Entity
A specific, identifiable thing, such as a person, business, product, or place, that search engines and AI systems track as one item across many sources. Entity-first SEO means making sure every source describes the same entity the same way.
FAQ markup
FAQPage structured data. Google deprecated the FAQ rich result and stopped showing it in Search on May 7, 2026, so the markup no longer earns an expanded listing. It still describes a page's questions and answers in a form any machine can read, and the questions themselves remain useful on the visible page.
Information gain and non-commodity content
What a page adds that other pages on the topic do not: original data, first-hand experience, a tested method, a named expert. Pages with something new to say have a reason to be cited; pages that restate the consensus do not. Google has described the same quality as content that is original and non-commodity.
Knowledge graph
A database of entities and the relationships between them. Google's Knowledge Graph feeds its knowledge panels, and AI systems build similar structures. Knowledge graph alignment means your schema, profiles, and listings all agree.
LLM seeding
Deliberately placing a brand's facts and presence in the sources models already trust and retrieve: reference works, community forums, review sites, and best-of lists. Answers are synthesized from a consensus of sources, so the work is getting the brand into that consensus, accurately and in the right company, instead of only publishing on its own site.
Prompt-to-publish loop
Researching the questions buyers put to AI assistants and publishing pages that answer them. This is keyword research with longer queries, and it works best when each answer lives on one page instead of many near-duplicates.

The metrics

Measurement in AI search is young, and every tool measures a sample. These terms are useful when the method behind them is stated.

AI citation rate
The share of answers, across a fixed set of questions, that link to or footnote your domain. It only means something when the question set, engines, dates, and location are written down and repeated the same way.
AI Performance (Bing Webmaster Tools)
A report Microsoft introduced in public preview on February 10, 2026. It shows how often a site is cited as a source in Microsoft Copilot and in AI-generated summaries in Bing: total citations, the average number of cited pages, per-URL citation counts, and sampled grounding queries. See how to read Bing’s AI Performance report.
Citation gap
The questions, and the sources credited on them, where competitors are cited and your business is absent. Mapped out, the gap is the working plan for closing the distance: which questions are worth answering, and which sources are worth being present in.
Citation share
A site's share of all citations for a given grounding query or set of questions: your citations divided by the citations credited to every site for that query. Bing Webmaster Tools' AI Performance report presents citation share in this form. It measures a share of the evidence an answer was built from, which is related to, but not the same as, ranking.
Cross-model visibility
How consistently a brand appears across different AI systems, such as ChatGPT, Claude, Gemini, Perplexity, and Copilot. Engines draw on different sources, so a brand can be named by one and missing from another.
Ghost citations
Citations to a page that never name the brand in the answer text: the source is credited in the footnotes, and the brand itself stays invisible to the reader. The mirror case is a mention with no citation behind it, where the engine names the brand but credits nothing. Both are counted apart from ordinary citations and mentions, because each is visibility that did not fully land.
Hallucination rate
How often AI answers state something false about a business: wrong prices, closed locations, invented services. Every wrong fact has a source or a gap behind it, and tracing that source is the start of the fix.
Mention rate
How often an AI answer names a brand, with or without a link. Mentions and citations are separate measures: a brand can be named without its site being cited, and a site can be cited for a fact without the brand being recommended.
Prompt monitoring, prompt panels, and prompt volume
Tools that run a fixed set of questions through AI assistants on a schedule and record the answers. A prompt panel is that question set; prompt volume is an estimate of how often people ask something similar. Volume figures are modeled estimates, since no AI company publishes query counts the way search engines publish keyword data.
Recommendation rate and shortlist inclusion
How often an answer engine not only mentions a business but recommends it, or includes it in the shortlist of candidates it puts forward. A mention says the engine knows the business exists; a recommendation spends the engine's credibility on it. The rate is tracked separately from mention rate because the two move separately.
Share of model and share of answer
The AI version of share of voice: how often an assistant names your brand compared with competitors across a set of prompts. Because answers vary from run to run, a single check proves little; repeated, dated runs show the trend.
Zero-click visibility
Presence in answers that users read without visiting any website. Rankings and mentions can grow while clicks fall, so reports should track both.

Which of these names real work?

Three ideas carry the weight. The first is that AI systems retrieve from search indexes and crawled pages, so technical SEO still decides whether a site can be found at all. The second is that answers are assembled from many sources, so the facts on a business's own site, its Google Business Profile, its listings, and the press about it have to agree. The third is that citations go to pages with something worth citing: original evidence, clear answers, and an identifiable author. Every acronym above is a way of packaging some part of those three.

For the measurement side, the guide to how AI answers choose which businesses to name walks through the entity record, the consistency check, and the citability test. For definitions of traditional SEO terms, such as canonical tags, crawl budget, and soft 404s, see the glossaries on christopherabraham.com and gerriscorp.com.

Sources

Ask an AI engine about your business.

Then book the call and tell us what it said. Fixing that answer is the work.

Book a 30-minute call