AI Commerce Glossary

A/B testing compares two variants against a single metric under random assignment. It answers only the question it was designed around, which is why analysis precedes testing.

Answer Engine Optimization is the practice of structuring content so AI assistants such as ChatGPT, Gemini, and Perplexity extract and recommend your brand in their responses. Where SEO competes for a ranked link, AEO competes for the answer itself.

AI automation is the replacement of recurring manual commerce work with systems that use models under human-defined limits. Its main failure mode is silent failure, not wrong output.

AI crawlers are the agents that fetch web content for AI systems — GPTBot, Google-Extended, PerplexityBot, ClaudeBot, and others. Their names change faster than any other part of a technical stack.

AI Overviews are Google’s AI-generated answers shown above traditional results. They shift traffic from clicks toward on-page answers, which is why AEO measures presence rather than position.

An answer engine is a system that responds to a query with a synthesised answer rather than a list of links. ChatGPT, Perplexity, and Google AI Overviews are answer engines.

Average order value is revenue divided by order count. It is the lever that changes the value of every other conversion improvement.

Author signals are the markup and on-page elements that attribute content to a named person with stated credentials. An Organization listed as author is not an author signal.

A brand entity is a resolvable identity for your brand in a knowledge graph — a record a machine can point at rather than a name it must guess from context. Without one, models cannot reliably distinguish your brand from a similarly named one.

Customer acquisition cost is the total spend required to win one customer. Read alone it says nothing; read against LTV and payback period it says almost everything.

A canonical tag declares which URL is the authoritative version of a page. It consolidates signals across duplicates rather than letting them compete with each other.

Cart abandonment is the share of started carts that never become orders. It is a proportion of carts started, not of orders completed — an inversion that routinely undercounts the opportunity.

Cart recovery is the practice of winning back abandoned carts through timed messaging. Its revenue is a component of the conversion gap, not an addition to it.

Citation building is the practice of earning mentions in sources an AI system already trusts. It is what shifts a model from knowing your brand exists to being willing to recommend it.

Cumulative Layout Shift measures unexpected movement of visible content during load. Reserved dimensions for images, fonts, and dynamic rows are the usual fix.

A content glossary is a structured set of term pages, each marked with DefinedTerm schema. It is one of the few content assets that serves topical authority and AI extraction with the same page.

Conversion rate is the share of sessions that end in a purchase. It is the denominator of most commerce diagnosis, and it is meaningless without the traffic quality behind it.

Core Web Vitals are Google’s user-experience metrics: LCP for loading, INP for responsiveness, and CLS for visual stability. FID was retired in March 2024.

Crawl budget is the number of URLs a crawler will fetch from a site in a period. Large catalogues waste it on faceted and parameterised duplicates more often than on genuine pages.

Conversion Rate Optimization is the practice of increasing the share of visitors who complete a purchase, using instrumentation and controlled experiments rather than opinion.

DefinedTerm is a schema.org type that marks a term and its definition as an entry in a defined set. Answer engines extract these preferentially because the boundary of the answer is unambiguous.

A direct-answer paragraph answers the page’s core question in its first 150 words, before any preamble. It is the single content-structure change with the largest effect on AI extraction.

E-E-A-T is Google’s framework for assessing content quality: demonstrated experience, subject expertise, authoritativeness, and trust. Named, credentialed authors are its most under-implemented component in commerce content.

An embedding is a numerical representation of text that places similar meanings near each other. It lets a system compare passages without matching words.

Entity resolution is the process of deciding which real-world thing a mention refers to. Inconsistent brand naming across sources is the most common reason it fails for commerce brands.

Faceted navigation is filtering that generates a URL per combination. Left uncontrolled it multiplies near-duplicate pages and consumes the crawl budget real pages need.

FAQPage schema is a structured-data type that marks a set of questions and answers as machine-readable. It is the single highest-leverage AEO signal because it hands an answer engine a pre-formed answer.

A featured snippet is an extracted passage shown at the top of search results. The same structural qualities that win one — a direct answer in the opening sentences — also make a page extractable by answer engines.

Gamification is the use of reward and progress mechanics to raise engagement and repeat purchase. Applied carelessly it trains customers to wait for rewards rather than to buy.

A generative engine is an AI system that composes novel responses from learned patterns rather than retrieving stored documents. Its output depends on what it has learned about a brand, not only on what it can currently fetch.

Generative Engine Optimization is the practice of building brand presence inside generative AI models — through entity records, citations, and consistent naming across sources — rather than inside a search results page.

GPTBot is OpenAI’s crawler. Blocking it in robots.txt removes your site from what OpenAI’s models can learn, which directly undermines GEO.

A hallucination is a confidently stated but false model output. In commerce it most often appears as invented product specifications, policies, or availability.

HowTo schema marks a sequence of steps as a machine-readable procedure. It suits process content, which answer engines are asked for constantly and extract from poorly when unmarked.

Interaction to Next Paint measures the latency between a user interaction and the next visual update. It replaced First Input Delay as a Core Web Vital in March 2024.

The inverted pyramid is a structure that puts the conclusion first and supporting detail after. It suits answer engines for the same reason it suits news: the extractable claim is never buried.

JSON-LD is the recommended format for embedding structured data in a page. It sits in a script tag, separate from the visible markup, so content and annotation cannot drift apart.

A knowledge graph is a database of entities and the relationships between them. Search and answer systems use one to decide what a name refers to before deciding what to say about it.

Largest Contentful Paint measures when the largest visible element finishes rendering. It is the closest single proxy for when a page feels loaded.

Lifecycle automation is the set of triggered messages that respond to customer behaviour over time — welcome, abandonment, replenishment, win-back. It is the highest-margin automation in most commerce stacks.

A large language model is a neural network trained on text to predict likely continuations. It is the component that produces the wording of an AI answer, and the reason brand presence in training and retrieval sources matters commercially.

LLM visibility is how often, and how accurately, a brand appears in AI-generated responses for a defined query set. It is measured against a fixed baseline because model updates change results independently of your work.

A loyalty programme is a structured set of benefits that rewards repeat purchase. Tiering by access and recognition preserves premium positioning better than tiering by discount.

Lifetime value is the total margin a customer produces over the relationship. It sets the ceiling on what acquisition can profitably cost.

Payback period is the time taken to recover acquisition cost from a customer’s margin. It governs how fast a business can grow without outside capital.

A product feed is a structured export of catalogue data for external systems. Its completeness increasingly determines whether AI shopping surfaces can represent a product at all.

Retrieval-augmented generation is a technique where a model answers from documents fetched at query time rather than from memory alone. It is what makes an assistant quote your real policy instead of inventing one.

Retention is the share of customers who purchase again within a period. It compounds where acquisition does not, which is why it moves LTV faster than any channel change.

Revenue leak is the quantified gap between what a store currently earns and what its existing traffic could support. Modelled carelessly, overlapping leaks are summed and the total exceeds real revenue.

robots.txt is the file that tells crawlers which paths they may fetch. It is the first thing a GEO audit checks, because a default deny silently cancels every other effort.

Sample size is the number of observations a test needs to detect an effect of a given size. Planned in advance it sets the test duration; ignored, it invites false positives.

Search Engine Optimization is the practice of making a site readable, trustworthy, and relevant to search engines. In an AI-search era it is also the foundation answer engines depend on: a page a crawler cannot reach is a page no model can cite.

Share of model is the proportion of AI responses within a category query set that name a given brand. It is the AI-search analogue of share of voice.

Statistical significance is the probability that an observed difference is not chance. Calling a test before it reaches significance is the most common way conversion programmes produce confident wrong answers.

Structured data is machine-readable markup that states explicitly what a page contains. It removes the need for a crawler or model to infer meaning from layout.

Technical SEO is the work that makes a site reachable, parseable, and indexable — architecture, rendering, status codes, and structured data. It determines whether any other optimisation can register.

Topical authority is the depth and coherence of a site’s coverage of a subject. Both crawlers and models use it to decide whether a domain is a source or merely a participant.

Vector search finds documents by meaning rather than keyword match, comparing numerical representations of text. It is the retrieval half of most RAG systems.

Wikidata is an open, structured knowledge base widely used as an entity source by search and AI systems. A well-formed Wikidata record is one of the cheapest ways to make a brand machine-resolvable.

A zero-click search is one that ends without the user visiting a website, because the answer appeared in the results surface. Being the source of that answer is the objective AEO replaces ranking with.