AI visibility glossary

Actions works by organising identified issues into practical fixes and an order for addressing them.

AEO means answer engine optimisation: making information clear and accessible so answer engines can use it to answer questions.

AEO, or answer engine optimisation, is work aimed at direct-answer systems, while GEO, or generative engine optimisation, is work aimed at systems that generate responses.

AEO, or answer engine optimisation, is work focused on direct answers, while SEO, or search engine optimisation, is work focused on visibility in search results more broadly.

AI brand monitoring, or artificial intelligence brand monitoring, matters because generated answers can mention, omit or misdescribe a business.

An AI crawler, or artificial intelligence crawler, is automated software that fetches web content for purposes associated with AI systems.

AI recommendation means an artificial intelligence system's suggestion that an option is suitable for a stated need.

AI referral traffic, or artificial intelligence referral traffic, is website traffic attributable to visits from AI services where the source can be identified.

AI search, or artificial intelligence search, works by using AI systems to interpret questions and retrieve, organise or generate relevant information.

AI SEO means artificial intelligence search engine optimisation, a label used for optimisation for AI-led search or for using AI in conventional search work.

AI share of voice, or artificial intelligence share of voice, is a brand's share of a defined measure of presence within sampled AI answers.

AI visibility, or artificial intelligence visibility, is how a brand appears in answers generated by AI systems.

An answer engine is a system that responds to a question with a direct answer rather than only a list of links.

Answer variability works through changes in generation, retrieval and context that can produce different responses to similar or identical questions.

A baseline is a recorded starting pattern used as a reference for later comparison.

A before and after comparison works by examining comparable observations collected before and after a defined change.

Blind spot matters because a gap in relevant answer coverage can be missed when checks focus only on questions that name the brand.

A brand mention is an explicit reference to a brand in an answer or other content.

Brand sentiment in AI, or artificial intelligence, works by interpreting the positive, negative or neutral language surrounding a brand in generated answers.

A branded question is a question that explicitly names the brand being examined.

A branded question is a question that names the brand being examined, while an unbranded question is a question that does not name it.

ChatGPT is a conversational artificial intelligence service; GPT stands for generative pre-trained transformer.

Citation matters because it gives readers a reference they can inspect when assessing an answer's claims.

Claude matters because its generated answers can describe, compare or recommend brands in response to user questions.

ClaudeBot is a crawler user-agent associated with collecting web material for model development.

A clean session is an interaction designed to exclude prior conversational context and personalisation from the question being tested.

A comparison page is content that explains meaningful differences between options using relevant, supportable criteria.

Conversational search works by allowing users to ask follow-up questions and refine a search through dialogue.

Digital PR for AI visibility, or digital public relations for artificial intelligence visibility, matters because credible external coverage can contribute information available to answer systems.

Discovery works by examining whether a brand appears when a user describes a need without naming that brand.

E-E-A-T means experience, expertise, authoritativeness and trustworthiness, concepts used in search quality evaluation.

An entity is a distinguishable person, organisation, place, product or other thing that information refers to.

Gemini works by using generative models to respond to prompts, with available tools and information sources depending on the service and settings.

Generative engine means a system that creates a response, such as a written explanation, rather than simply reproducing a fixed result.

GEO, or generative engine optimisation, matters because generated answers can shape how people discover and understand brands.

Google AI Mode means a search experience that uses artificial intelligence to provide conversational, generated responses and support follow-up questions.

Google AI Overviews is a search feature that uses artificial intelligence to generate summaries for some queries alongside other search results.

Google-Extended matters because it lets site owners express controls over certain uses of their content for generative systems, separately from ordinary search crawling.

GPTBot means a crawler user-agent associated with collecting material that may be used for model training; GPT stands for generative pre-trained transformer.

Grounding matters because it connects an answer's claims to supplied evidence or an external source of information.

Hallucination is generated content presented as factual despite being unsupported, fabricated or inconsistent with the available evidence.

IndexNow works by notifying participating search engines that a website address has been added, changed or removed.

JavaScript rendering matters because some automated readers cannot reliably access content that appears only after browser code runs.

Knowledge cutoff means the stated boundary beyond which a model's training-based information may not reliably include later developments.

A knowledge graph works by representing entities and the relationships between them in a structured form.

A large language model (LLM) works by learning patterns from training data and using context to generate sequences of text.

Live web search works by retrieving web information at the time a question is processed, rather than relying only on learned information.

LLM visibility, or large language model visibility, matters because language-model answers can influence how a brand is described and considered.

LLMO, or large language model optimisation, is work intended to improve how language-model-based systems understand and represent information.

llms.txt is a proposed website file that presents information and links intended to help large language models find useful content.

Mentioned in passing is an incidental reference that names an option without clearly recommending it.

Mentioned is being named in an answer, while recommended is being positively suggested as suitable for a need.

Microsoft Copilot is an artificial intelligence assistant name used across conversational and productivity experiences.

Model version means an identifiable release or configuration of a model used to generate responses.

Named with caution means an option is identified alongside a warning, reservation or limitation relevant to the user's decision.

Not mentioned matters because absence from a relevant answer differs from appearing without a recommendation.

OAI-SearchBot works as a web crawler associated with search retrieval, rather than the training-focused role of GPTBot.

Perplexity works by combining generated responses with retrieved information to answer questions, commonly with source references.

PerplexityBot means a crawler user-agent associated with discovering and indexing web content for search answers.

Personalisation and memory describe how saved preferences and prior interactions can influence which brands appear and how answers are framed.

Prompt means the instruction, question or other input given to a model to guide its response.

Query fan-out works by expanding a user's request into related searches that help a system gather information for an answer.

A question set works by defining the collection of questions used to examine answers consistently.

A questions-and-answers page works by pairing genuine user questions with direct, useful answers.

Receipt means the inspectable evidence supporting an interpretation of an answer.

Recognition is whether a system can identify and describe a brand when the question names it.

Recommended matters because a positive suggestion conveys more than the mere presence of a brand name.

Reddit and forums in AI answers means the use of discussion-site material in answers generated by artificial intelligence systems.

Retrieval-augmented generation (RAG) is a method that supplies retrieved information to a generative model as context for its answer.

Reviews and AI answers is the relationship between published evaluations and responses generated by artificial intelligence systems.

robots.txt works by publishing crawl directives that compliant automated agents read before fetching parts of a website.

Sample size matters because the amount of observed evidence affects how confidently a pattern can be distinguished from isolated answers.

An answer-classification standard separates recommendation strength, incidental mentions, cautions and absence instead of treating every brand reference alike.

Share of model means a defined measure of how often or prominently a brand appears in sampled model outputs relative to other brands.

Structured data means information expressed in a consistent machine-readable format with explicit labels and relationships.

Third-party sources matters because information published outside a brand's own channels can influence how that brand is described.

Top pick means an option presented as a leading choice for the need described in an answer.

Training data is material used to adjust a model during training so it learns patterns that influence its outputs.

Training data is material used to shape a model during training, while live search is information retrieval performed when a question is processed.

Unbranded question means a question that does not name the brand whose visibility is being examined.

Week-to-week variation matters because answers can change over time even when a brand has made no relevant fix.

Weekly range means the lowest to highest value across a stated set of comparable weekly readings.

A wrong fact is a factual claim that conflicts with reliable evidence about the subject.

Zero-click search is a search interaction that ends without a click through to an external website.