The vocabulary of AI visibility, without the marketing.

SEOSearch Engine Optimization

SEO is the discipline that makes a site visible in search engine results, through technical, content and authority work.

It covers everything that influences whether and how a site appears on Google, Bing or other engines: technical accessibility, content relevance to search intent and external trust signals. In 2026, Google states explicitly that optimizing for generative results is still SEO, not a separate discipline — a position that contradicts part of the commercial vocabulary in the market.

SEO Consulting →

GEOGenerative Engine Optimization

GEO is the discipline through which a brand gets selected and mentioned in the answers generated by AI engines such as ChatGPT, Perplexity, Google AI Overviews or Copilot.

It works on two distinct channels: retrieval (the engine searches in real time and cites sources) and parametric (what the model already "knows" from its training data). The two call for different interventions — the first is solved through technical access and extractable content, the second through entity presence and mentions in sources the models have learned from.

GEO service →

AEOAnswer Engine Optimization

AEO is the practice of structuring content at the passage level so that an answer engine can extract a self-contained fragment and cite it directly.

The operational difference from GEO: AEO deals with the form of the content (a question followed by a short, complete answer with no dependence on context), while GEO deals with whether you are chosen as a source. In practice they overlap, and many practitioners use them as synonyms — we separate them because they call for different work, not because the industry has an official definition.

AEO service →

AIOAI Optimization / AI Overviews

AIO is used with two different meanings: either as an umbrella term for optimizing a digital presence for AI systems in general, or as shorthand for Google AI Overviews — the AI-generated summary at the top of Google results.

The two meanings call for different answers. As an umbrella: some sources put it above GEO, others exactly the reverse, there is no established definition that usefully distinguishes it from GEO or AEO, and in practice the three terms are used interchangeably. As shorthand for AI Overviews: note that it is not the abbreviation Google uses, since Google says "AI Overviews" — and, more importantly, it names a product, not a discipline. This is the category error in most comparisons: SEO, AEO and GEO are practices, whereas AI Overviews, ChatGPT, Perplexity, Gemini or voice assistants are surfaces that those practices target. We document it because it appears in the market, not because it describes work different from what we do under GEO/AEO.

GEO vs AEO vs AIO vs SXO →

LLMOLarge Language Model Optimization

LLMO is yet another commercial synonym for optimizing visibility in the answers of large language models.

It appears alongside GSO (Generative Search Optimization) and AIO as a variant name for the same practice. Wikipedia explicitly notes that, at the start of 2026, there was no established definition in the literature separating these terms. Choosing between them is a marketing decision, not a technical one.

SXOSearch Experience Optimization

SXO is the optimization of what happens after someone has found you: speed, clarity, how easy it is to complete the desired action.

Unlike GEO and AEO, which deal with discovery, SXO deals with conversion and the quality of the experience. It overlaps substantially with CRO and with Core Web Vitals. The term is more common in English-speaking markets than in Romania.

Content & CRO →

AI Overviews

AI Overviews is the AI-generated summary that Google displays above the classic results, with links to the sources it used.

The main commercial effect is that it answers the question directly on the results page, which reduces clicks to websites even for top positions. Sources cited in the overview gain brand visibility without a guaranteed click — the reason measurement has to shift from "traffic" to "citations".

AI Overviews traffic guide (in Romanian) →

AI Mode

AI Mode is the conversational search mode in Google, in which the user holds a dialogue instead of going through a list of blue links.

Unlike AI Overviews, which complements the classic results, AI Mode replaces them with a conversation interface. The consequence for optimization: follow-up questions matter, not just the initial query — a topic has to be covered in depth, not only at the level of the first question.

RAGRetrieval-Augmented Generation

RAG is the architecture in which an AI model first searches for relevant documents and only then generates the answer, using what it found.

It is the mechanism behind citations: the model does not "remember" the page, it retrieves it at the moment of the question. For optimization it matters that the system works with fragments, not whole pages — which is why passage-level structure decides what can be extracted and cited.

Retrieval channel vs. parametric channel

The retrieval channel is what the AI engine finds by searching in real time, while the parametric channel is what the model already knows from the data it was trained on.

The distinction is practical, not academic: on retrieval you can intervene in days (bot access, structure, schema) and the effect is verifiable immediately; on parametric you intervene over months, through mentions and entity consistency, and the effect only appears at a retraining or model update. A strategy that confuses them promises impossible timelines.

GEO service →

Citation (AI citation)

A citation is the moment an AI engine names your source or links to it in the generated answer.

It is the unit of measurement that replaces ranking position, because in a generative answer there is no "3rd place". A citation brings brand visibility even without a click, which makes reporting harder and easier for vendors to exaggerate.

Crawl budget

Crawl budget is the number of pages a search engine is willing to crawl on your site within a given period.

It becomes a real problem only at scale — tens of thousands of URLs and up — or when parameter URLs multiply uncontrolled. The typical symptom is not an error, but important pages that stay unindexed for months while the bot spends the budget on useless variants.

Technical SEO →

Rendering

Rendering is the process by which a bot executes the code of a page to see the final content, not just the raw HTML received from the server.

It matters because many AI retrieval bots do not execute JavaScript or execute it only partially. A site that loads its text client-side can look perfect in a browser and completely empty to the bot that is supposed to cite it.

Technical SEO →

Schema markup / JSON-LD

Schema markup is a set of structured data added to the page that explicitly describes what the content represents: an article, a product, a question, an organization.

JSON-LD is the recommended format, added as a separate script without affecting the visible text. It guarantees nothing on its own, but it reduces ambiguity: an engine no longer has to infer from the text that you have a frequently-asked-questions section or that a given entity is the author. Google stated in 2026 that there is no special schema for AI — the same standard types are used.

Schema.org, in detail →

Core Web VitalsLCP, INP, CLS

Core Web Vitals are three metrics that measure loading speed (LCP), responsiveness to interaction (INP) and visual stability of the page (CLS).

They are measured in two ways that are easy to confuse: lab data (simulation, e.g. Lighthouse) and field data (real users, e.g. CrUX). Only field data reflects the real experience. For AI citation they matter because a retrieval bot may abandon a page that is too slow and cite a faster source.

Core Web Vitals →

robots.txt

robots.txt is the file in the root of the site that tells bots what they may and may not access.

For AI visibility it is the first thing to check: bots such as GPTBot, ClaudeBot, PerplexityBot or Google-Extended are controlled exactly from here. An inherited block, put in place unintentionally, is the most banal and most frequent reason a site does not appear at all in AI answers.

robots.txt & llms.txt guide (in Romanian) →

llms.txt

llms.txt is a file proposed as a standard, in which a site lists its important pages in Markdown for consumption by language models.

It is not an officially adopted standard, and Google said explicitly, in 2026, that it does not use it. The remaining argument in its favor is that it is cheap to maintain and that consumers other than Google may read it. Anyone selling it as a decisive factor for AI visibility goes beyond what can be demonstrated today.

robots.txt & llms.txt guide (in Romanian) →

Canonical

A canonical is the tag by which you tell the engine which URL variant is the official one, when the same content is accessible at several addresses.

It typically appears with filters, sorting parameters, versions with and without www, or with and without a trailing slash. Without a canonical, signals are split between variants and no version accumulates enough authority.

Search intent

Search intent is the real purpose behind a query: to find out something, to compare options, to get somewhere or to buy.

It is the criterion that decides what type of page should exist, not just what words it contains. The most frequent cause of a page that "does not rank even though it is well written" is that it answers a different intent from the one the engine shows results for.

On-page SEO →

Keyword cannibalization

Cannibalization occurs when two or more pages on the same site target the same search intent and split their signals instead of accumulating them.

The engine does not know which one to show, so it oscillates between them or picks the wrong one. The fix is not "more content", but consolidation: one intent, one page, with the rest redirected or canonicalized to it.

On-page SEO →

Passage-level content

Passage-level content means self-contained fragments — a descriptive heading followed by a complete answer in 2-4 sentences — that can be extracted and understood without the rest of the page.

It is the form a RAG system can cite directly. The practical test: if you cut the fragment out and read it on its own, does it still make sense? If it starts with "this" or "as we said above", it is not a citable passage.

AEO service →

Featured snippet

A featured snippet is the answer block that Google extracts from a page and displays above the classic results.

It is the direct predecessor of AI citation and works on the same principle: the engine looks for a short fragment that answers completely. Optimizing for snippets and optimizing for AI citation overlap almost completely.

E-E-A-TExperience, Expertise, Authoritativeness, Trust

E-E-A-T is the framework by which Google evaluates whether a source deserves trust, along four dimensions: direct experience, expertise, recognized authority and trust.

It is not a score you can see anywhere, but a set of criteria used by human evaluators and reflected indirectly in the algorithms. For AI engines it works as an entry filter: a source without trust signals does not even make it into the candidate set for citation.

E-E-A-T on the AEO page →

Entity

An entity is a distinct, identifiable thing — a company, a person, a product, a place — that engines treat as a unique object, not as a string of characters.

The practical difference from a keyword: an entity has relationships (who founded it, where it is, what it sells) that can be cross-checked between sources. The consistency of these relationships across your site, profiles and external mentions decides whether the engine recognizes you as a coherent entity or as contradictory pieces of information.

Off-page SEO →

Knowledge Graph

The Knowledge Graph is the knowledge base of Google about entities and the relationships between them, used to understand the world beyond the text of pages.

Presence in it cannot be bought and cannot be requested through a form: it is built through consistent reference sources that state the same facts about the entity. It is slow, but it is one of the few assets that also carries over into the parametric channel of AI models.

Backlink

A backlink is a link from another site to yours, interpreted by engines as a vote of relevance and trust.

What matters is the source, the context and whether the link is earned or paid — not the count. Links bought in volume are the most common cause of manual penalties and, unlike a technical problem, recovery takes months.

Off-page SEO →

Internal linking

Internal linking is the network of links between the pages of the same site, which distributes authority and explains to the engine how topics connect.

It is the only authority lever you control completely, without depending on anyone else. The anchor text matters as much as the link itself: "learn more" says nothing, "heading structure for citable passages" says everything.

On-page SEO →

Unlinked mention

An unlinked mention is a reference to your brand in a public text, without a hyperlink to the site.

It brings no traffic and for a long time was considered worthless for SEO. For the parametric channel of models it still matters: the model learns the association between brand and topic from text, not from a link. It is one of the few signals where AI behaves differently from the classic engine.

Share of model voice

Share of model voice is the proportion of AI answers to a set of relevant questions in which your brand appears, compared with competitors.

It is the equivalent of market share for generative visibility. It has a serious methodological trap: model answers vary between runs and between users, so a single measurement means nothing — what matters is the series over time on the same set of questions.

CrUXChrome User Experience Report

CrUX is the public dataset from Google with real performance measurements, collected from Chrome users.

It is the source of "field data" for Core Web Vitals, as opposed to lab data produced by simulation tools. The difference matters: a site can look excellent in a test and poor in CrUX, because real users have weaker devices and connections than the test environment.

Impressions vs. citations

Impressions count how many times a result was displayed, while citations count how many times a source was actually named in an AI-generated answer.

Confusing them is the most convenient form of inflated reporting: impressions exist in Search Console and grow easily, citations require their own querying and verification methodology. A report that talks about "AI visibility" without saying which questions it tested and how many times measures nothing.