Glossary57 terms

Understand MCP, hosts, clients and servers, then compare MCP vs API through an illustrative read-only Search Console and GA4 weekly report.

An AI agent chooses steps and tools toward a goal. Compare it with a chatbot and a fixed workflow through a product-comparison task and completion criteria.

An AI hallucination is plausible-looking but incorrect or unsupported output. Check the original source and the model’s interpretation separately.

AI search combines information retrieval with generated answers. Learn what source links establish, how errors can arise and how it relates to SEO and GEO.

AI Overviews summarize information in some Google results. A linked source is an exposure observation, not a website visit or a conversion.

Anchor text is the clickable wording in a link. Learn why descriptive phrases help readers understand an internal link before opening it.

A CDN distributes content through edge servers. Learn cache hits, browser-cache differences and how bot protections can affect crawler access.

Channel groups classify traffic using rules. Distinguish source, medium and reporting scope, including AI Assistant versus Google AI search traffic.

Learn what circular reporting means, follow a documented case, and use four checks to distinguish repeated claims from independent evidence.

Conversion rate depends on what counts as a conversion and whether the denominator is interactions, sessions or users. Compare examples safely.

Average CPC equals ad spend divided by clicks. Use matching dates and currency, and distinguish a realized average from a maximum bid or customer outcome.

CTR equals clicks divided by impressions. Calculate search and ad examples, combine groups correctly and separate a zero rate from an undefined denominator.

A digital anchor is LS-Labs’ research term for an official information point a brand can maintain, centered on its website. It does not guarantee AI use.

LS-Labs uses digital entity for the recognizable representation of a real business across sources. It is a research distinction, not a platform standard.

An embedding represents information as a numerical vector. Semantic similarity helps retrieve related material but cannot prove a source accurate.

Google Analytics 4 records website and app activity as events. Separate users, sessions and key events, including the limits of AI referral attribution.

Google Business Profile lets eligible businesses manage information in Search and Maps. It is maintained separately from a website and schema markup.

GEO studies content visibility in generated answers. Separate brand mentions, citations and factual accuracy before evaluating changes.

Internal links connect pages on the same website. See a coffee-guide example and how to choose a useful destination and link text.

A GA4 key event is an event marked as important. Learn why a visible button, an event count and the number of people completing an action are different.

A landing page is the first page opened during a session. Pair it with traffic source to distinguish where a visit began from where the visitor came from.

An LLM is a large language model trained on vast text that generates answers by predicting tokens. Definition, tokens and common misconceptions.

LocalBusiness schema describes business information in website markup. Learn how JSON-LD and Google Business Profile serve different roles.

MCAC separates Mention, Citation, Accuracy and Conversion-readiness. Learn the four checks, their units and why they do not form a universal brand score.

noindex tells a search engine not to index a page. The engine must be able to retrieve the instruction, so a robots.txt block can interfere.

Open banking lets specified bank data or services be shared with authorized third parties through controlled interfaces. Read and payment access differ.

RAG retrieves information for a model before it generates an answer. A returns-policy example shows why retrieval still needs source checks.

ROAS is attributed revenue divided by ad spend. A 3× return equals 300%, but profitability also depends on margin, fulfillment costs and attribution scope.

robots.txt gives automated crawlers fetch rules. It is different from noindex and access control, and CDN protections may independently block a request.

A search API lets software query a search service and receive results. Understand titles, URLs, snippets, RAG retrieval and MCP connections.

Search intent is the purpose behind a query. Compare learning, price comparison and buying needs using a practical coffee example.

SEO helps search systems access and understand useful website content. See a venue-booking example and how SEO relates to GEO and AI search.

A SERP is the whole search results page, including links, maps, images and AI summaries. Position alone cannot explain clicks.

A GA4 session groups activity during a visit. Learn engaged-session criteria, engagement and bounce rates, and why session counts differ from users.

A token is a unit a model processes. It is not a fixed word or character count, and model token usage should be separated from search API charges.

User-generated content includes reviews, photos and videos posted by users. Understand examples, sponsorship distinctions and limits in search evidence.

UTM parameters label links for analytics. See source, medium and campaign fields, and why a tag does not reconstruct an earlier AI conversation.

A web crawler automatically retrieves pages. Learn how crawler purpose, robots.txt and CDN rules differ before deciding which access to permit.

Zero-click search means a search produced no website-result click under a stated definition. Learn common situations and why study percentages differ.

TSO (tool selection optimization) is LS-Labs' framework for finding out why an AI agent did or did not pick your tool, API or MCP server, in four stages.

llms.txt is a community proposal by Jeremy Howard: a Markdown guide to a website for AI agents. The v2 format, how it differs from robots.txt, and usage data.

How to open Google AI Mode, view and delete saved searches, and check missing entry points, with illustrated guides based on Google's documentation.

AX (agent experience) is the overall experience AI agents have as users of a product. Netlify's Mathias Biilmann introduced the term in 2025.

Affordance is what an environment lets a particular actor do. A feature existing and the actor finding it are separate questions; the second needs signifiers.

BM25 is a keyword ranking method that scores documents by how often query terms appear, how rare they are in the dataset, and document length.

Candidate retrieval pulls possibly relevant items from a larger pool of content or tools. Learn how Recall@5 differs from a model's first choice.

Function calling lets an AI model use external tools: the model returns the function and arguments to call, and the application runs it and returns the result.

Idempotency means repeating a request has the same intended effect on the server as running it once. It does not guarantee identical responses or success.

Reranking uses a finer model to re-score first-pass search candidates and reorder them by relevance. It usually covers only top results, as it costs more.

TCC (Task–Call Contract) is an LS-Labs method for checking AI tool use: fix the task, tool version and correct behavior first, then verify each decision.

A tool description is the text in a tool definition that says what a tool does and when to use it. It is not a parameter schema and grants no authorization.

Tool metadata is the tool-definition information an AI application receives: name, description, input schema, plus optional output schema and annotations.

AI agents can help people discover and compare products, then carry out supported purchase steps within the buyer’s authorization.

A product feed supplies catalog information in a platform’s required format, including identifiers, names, prices, availability and images.

How attribution assigns conversion credit, how GA4 last click treats direct visits, and why an AI conversation can be absent from the measured path.

Incremental conversions are additional outcomes caused by advertising compared with what would have happened without it.

Understand Simpson’s paradox with a worked example, and see why across-site GEO correlations can disappear when pages are compared within the same site.