AI Search Glossary for Indian Marketers

The category itself

Answer Engine Optimization (AEO)

Answer Engine Optimization is the practice of improving whether and how a brand is cited, described and recommended by AI assistants when they answer user questions. It differs from search engine optimization in that the objective is being named inside a generated answer rather than ranking in a list of links.

Generative Engine Optimization (GEO)

Generative Engine Optimization refers to the content and source work intended to increase the likelihood that generative AI systems cite or recommend a brand. In practice most vendors use GEO and AEO interchangeably, and the distinction between them is not standardised.

LLMO (Large Language Model Optimization)

LLMO is a less common synonym for AEO and GEO, emphasising optimisation for large language models specifically. The proliferation of near-synonyms in this category reflects that no term has yet won.

AI visibility

AI visibility describes how often and how favourably a brand appears in AI-generated answers. It is used both as a general category label and as a specific metric, which is a frequent source of confusion in vendor comparisons.

How the systems work

Citation

A citation is a source an AI system references in support of an answer. A citation indicates that a source informed the response; it is not proof that the source alone caused a particular brand to be named.

Non-determinism

Non-determinism is the property that the same prompt can produce different answers on different runs. It is why a single observation of AI visibility is unreliable and why measurement requires repeated sampling.

Metrics

Visibility

Visibility is the proportion of runs across a prompt set in which a brand is mentioned. A visibility figure is meaningless without its prompt set, sample size and date window.

Share of voice

Share of voice is a brand's mentions as a proportion of all brand mentions across a defined prompt set. It is highly sensitive to which prompts are in the set, so share of voice figures are only comparable when the prompt set is identical.

Position

Position is where a brand appears within a recommendation list in an answer. Being named third in a shortlist differs materially from being mentioned in passing, and metrics that count only mentions treat the two identically.

Citation share

Citation share is the proportion of sources cited in answers about a category that belong to a given domain. It is the most diagnostic metric in AI search measurement, because it identifies which surfaces are actually shaping the answer rather than only whether a brand appears.

Share of model

Share of model is a vendor term for a brand's presence across AI-generated answers in a category, broadly equivalent to share of voice. Definitions vary between providers.

Sentiment

Sentiment in AI visibility measurement describes how a brand is characterised in generated answers, not whether it is mentioned. Sentiment is noisier than mention frequency and is generally better treated as directional than as a precise score.

AI characterisation risk

AI characterisation risk is the exposure created when an AI system describes a brand inaccurately, associates it with negative context, or attributes another entity's issues to it. It differs from traditional reputation risk in that the characterisation is delivered as the answer rather than as one result among several a user scrolls past.

AI product accuracy monitoring

AI product accuracy monitoring is the practice of checking whether the specific factual claims AI systems make about a product, such as prices, rates, fees or eligibility conditions, match what the provider currently publishes. It is distinct from visibility measurement, which tracks whether a brand appears rather than whether what is said about it is correct.

Decision-grade versus directional

Decision-grade and directional are quality classifications from the IAB's 2026 AI visibility measurement framework, distinguishing data reliable enough to act on from data useful only as a signal. Classifying metrics against this distinction is more informative than presenting all figures at equal confidence.

Prompt volume estimate

A prompt volume estimate is a modelled figure for how often a question is asked of AI systems. No AI platform publishes prompt-level query volumes, so every such figure is an estimate derived from third-party data and should be treated as directional.

Measurement practice

Prompt set

A prompt set is the defined list of questions a brand's AI visibility is measured against. It is the single most consequential methodological choice in AI visibility measurement, because a set that misses the questions customers ask produces a confident number about the wrong thing.

Prompt variant

A prompt variant is one version of a base prompt, varied by language, location or phrasing. A base prompt tracked in two languages across four cities generates eight variants, which is the number that determines measurement cost, not the base prompt count.

Intent cluster

An intent cluster is a group of differently phrased prompts expressing the same underlying question. Grouping by intent rather than exact phrasing produces more stable measurement, since users rarely phrase the same need identically.

Prompt tiering

Prompt tiering is the practice of applying different measurement depth to different prompts, with revenue-critical questions sampled across all languages, locations and runs, and long-tail questions monitored at a baseline. It exists because applying full segmentation to every prompt is expensive and rarely necessary.

Cold session

A cold session is a query run without personalisation, conversational history or prior context. Measurement uses cold sessions so results reflect a first-time query rather than one primed by earlier turns.

Collection frequency

Collection frequency is how often a prompt set is run. More frequent collection produces more observations per variant, though how many are needed depends on variance, prompt count and the claim being made. See our measurement methodology.

Brand set overlap

Brand set overlap is the proportion of brands appearing in both of two comparable answers, used to quantify how much results diverge between languages, locations, engines or days. It is the clearest way to express instability in AI answers.

Baseline

A baseline is a dated record of AI visibility taken before any work begins. Without one, later improvement cannot be evidenced, and the absence of a baseline is the most common reason AI visibility programmes cannot demonstrate results.

Indian market terms

Hinglish

Hinglish is the mixing of Hindi and English within a single sentence, typically written in Roman script. It is common in many Indian consumer contexts, including among fluent English speakers, and Hinglish queries can surface different sources and return different brands than their English equivalents.

Transliteration variance

Transliteration variance is the phenomenon of one word in a transliterated language having multiple valid written forms, because no standardised spelling exists. Hindi written in Roman script produces forms such as accha, acha and achha for the same word, meaning a single query intent can appear in many surface forms.

Code-switching

Code-switching is alternating between two languages within a conversation or sentence. It appears frequently in Indian consumer queries, which is why measuring only one language form can give an incomplete picture of a brand's visibility.

City-scoped prompt

A city-scoped prompt is a query framed around or run from a specific location. AI recommendations for the same category can differ between Indian cities, so national-level measurement can average away variation that matters to regionally concentrated brands.

Vernacular content gap

The vernacular content gap describes the underrepresentation of Indian-language content on the web relative to the number of speakers. Where fewer sources exist for a model to draw on, individual sources may carry more weight, which is both a reliability consideration and a potential opening for brands publishing in those languages.

Technical

Crawler definitions below follow OpenAI's published crawler documentation, which distinguishes OAI-SearchBot for ChatGPT Search eligibility, GPTBot for potential training use, and ChatGPT-User for user-triggered fetching. Verify against the current documentation before publishing, as these change.

GPTBot

GPTBot is OpenAI's crawler for gathering data used to train its foundation models. Blocking GPTBot restricts training use and does not remove a site from ChatGPT search results.

OAI-SearchBot

OAI-SearchBot is OpenAI's crawler that determines whether a site's content is eligible to appear in ChatGPT search features. This is the crawler that matters for visibility in ChatGPT answers.

ChatGPT-User

ChatGPT-User is the agent OpenAI uses when a user's request causes ChatGPT to fetch a specific page during a conversation. It does not control search eligibility.

Google-Extended

Google-Extended is a control governing whether content is used for Gemini and Vertex AI grounding and training. It is not the control for eligibility in Google AI Overviews, which depends on ordinary Google crawling and indexing.

llms.txt

llms.txt is a proposed file for giving AI systems guidance about a site's content. Adoption by major AI providers has been minimal and there is little evidence it affects retrieval.

Structured data

Structured data is machine-readable markup describing a page's content, usually in schema.org vocabulary. It helps systems parse content accurately and is required for some search features, though Google has stated it is not required for AI search features and controlled testing has found limited effect on AI citation rates specifically.

Product feed

A product feed is a structured file of product data, including price, availability and attributes, supplied to a platform. Feeds are increasingly important for keeping product data accurate and current, and are required for some emerging commerce integrations.

Agentic Commerce Protocol (ACP)

The Agentic Commerce Protocol is an open specification developed by OpenAI and Stripe for enabling purchases through AI assistants.

Notes on contested terms