Reference

Glossary of AI search

The vocabulary of AI search and Generative Engine Optimisation, defined plainly. These are the definitions used throughout Get Recommended by AI, reproduced in full so you can check a term without owning the book.

Agentic Search: The next evolution beyond AI search. Instead of just generating an answer, an AI agent takes multiple autonomous steps: searching across many sources, cross-referencing data, evaluating options, eliminating weak candidates, and presenting a curated shortlist, or taking direct action (booking, purchasing, subscribing). The critical difference from AI search: the user never sees the options that were filtered out.
AI Citation: When an AI system uses your content as a source in its generated answer, naming your brand, linking your URL, or paraphrasing your page. Recent research splits this usefully in two: citation selection (being picked as a source) and citation absorption (your content actually shaping the answer), and the two don't always travel together.
AI Search: Search experiences powered by large language models that return direct answers instead of a list of links: ChatGPT, Gemini, Perplexity, Claude, Copilot, and Google's AI Overviews and AI Mode. Users read a generated answer synthesised from multiple sources instead of scanning links.
AI Visibility: Your presence within AI-generated answers, measured properly as a rate across repeated runs of locked prompts (see Volatility), not as a single answer's contents.
Answer-First Content: Content structured to lead with a direct answer to a specific question, followed by supporting detail. AI extracts self-contained chunks; leading with the answer guarantees at least one exists, and measured citations cluster toward the top of pages.
Author Authority: The perceived credibility of the person behind the content: named authors, real credentials, verifiable backgrounds, and, crucially, Person schema so machines can read the credential (see The Author Paradox).
The Author Paradox: The finding (from SearchScore's accountancy study) that 94% of firms displayed their qualifications but only 8% made them machine-readable. Expertise that humans can see and machines cannot: the legibility gap in one statistic.
Brand Authority: How widely your brand is recognised and referenced across the web: mentions, reviews, press, consistent positioning. The evidence says third-party signals like these decide evaluative AI answers far more than anything on your own site.
Citation Sentiment: Not just whether AI mentions you, but how: positive, hedged, or negative. Chapter 30 is the repair playbook.
Content Hub (Pillar Content): A cluster of interconnected pages around a central topic: a pillar overview plus deep sub-articles, all interlinked. Signals genuine topical coverage, which is what wins fanned-out queries.
Crawler Access: Whether AI bots can reach and read your content, controlled by robots.txt, firewalls, and CDN settings. The key nuance: each company runs separate search bots (OAI-SearchBot, Claude-SearchBot, PerplexityBot), which govern citations, and training bots (GPTBot, ClaudeBot), which govern model memory. Google's AI surfaces are fed by ordinary Googlebot; Google-Extended is a robots-only token for Gemini training/grounding and never appears in logs.
E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness): Google's content-quality framework, evaluated by its quality raters and reflected in the Search systems that also feed AI Overviews and AI Mode. Other engines publish no equivalent; treat E-E-A-T as documented for Google's surfaces and plausible elsewhere.
Entity Clarity: How clearly and consistently your organisation is defined across the web: category, audience, use case, stated identically on your site, your schema, your profiles, and your Wikidata entry.
GEO (Generative Engine Optimisation): The practice of making your brand more likely to be selected and cited in AI-generated answers. The term comes from the field's founding academic paper (Aggarwal et al., KDD 2024), which found quotable statements, statistics, and cited sources lift visibility 30–40% while keyword stuffing backfires.
GEO Score: A 0–100 estimate of how likely a site is to be found, understood, and cited by AI systems, used throughout this book from SearchScore's methodology (see the Appendix for how it's built). Like every score in this field, it measures signals that correlate with citation, a proxy, not a readout of any engine.
IndexNow: A protocol that instantly notifies participating search indexes (Bing, Amazon, Naver, Seznam, Yandex, Yep, not Google) when your content changes. Useful for AI visibility mainly via the Bing → Copilot pipeline.
The Legibility Gap: This book's thesis: most businesses' expertise exists but is not structured in a way machines can recognise. Closing it, making quality machine-legible, is what the three layers do.
LLM (Large Language Model): The AI systems powering AI search. Trained on vast datasets, then augmented at answer time with live retrieval. The two channels, training memory and retrieval, are the Two Games of Chapter 4.
llms.txt: A plain-text file at your domain root that summarises your business and lists your key pages for AI systems. Adoption is roughly one site in ten, and as of 2026 no major answer engine uses it as a citation signal (Google says so on the record; a 137k-site study found 97% of the files are never fetched at all). Worth five minutes as agentic-web insurance, not as a citation lever.
Memory vs Retrieval (The Two Games): The two ways AI recommends brands. Memory: instant, sourceless answers recited from training data; moved by web-wide mentions, on training timescales. Retrieval: live-searched answers with citations; moved by your site's accessibility and extractability, in weeks.
Mention Rate: The fraction of repeated runs of a prompt in which your brand is named. The meaningful unit of AI visibility (see Volatility).
NAP Consistency (Name, Address, Phone): Having your business details identical across every platform. Inconsistency dilutes entity clarity: machines can't confidently merge "Acme Corp" and "Acme Corporation Ltd" into one entity.
Query Fan-Out: The documented mechanism (Google's term) by which an engine splits one question into many hidden sub-queries, retrieves passages for each, and composes the answer. The reason coverage of a decision beats one perfect page.
RAG (Retrieval-Augmented Generation): The technique behind most AI search: retrieve relevant documents from a live index at question time, then generate the answer from them. RAG is why on-page structure matters: the pages that can be retrieved and cleanly extracted are the ones that get cited.
SAVI (State of AI Visibility Index): SearchScore's large-scale benchmark of AI visibility, drawn from a corpus of 1,000,000+ analysed websites and published quarterly, with sector editions (accountancy, dentistry, and others). The source of several figures in this book; methodology in the Appendix.
Schema Markup (Structured Data): Machine-readable JSON-LD embedded in your pages describing what things are (Organization, Person, FAQPage, Article, Product, HowTo). Evidence status: confirmed useful by Microsoft for Bing/Copilot; not required by Google for AI features; no controlled study shows it directly buys citations. Do it as cheap disambiguation hygiene and for entity clarity, not as a ranking trick.
Share of AI Voice: The fraction of brand-recommendation slots in your tracked prompts that go to you versus competitors (a metric name you'll also see used by AI-tracking tools). The competitive read on your tracking data.
Volatility (and Drift): Volatility: the large run-to-run randomness of AI answers, measured in 2026 research at only ~32–43% source overlap between same-day runs of the same prompt, which is why single runs mean nothing and rates are the unit. Drift: the slower, directional movement underneath (model updates, competitor optimisation, query shifts). Scatter is noise; slopes are signal.

Figures quoted in these definitions come from SearchScore's corpus of 1,000,000+ analysed websites, current as of 2026-07-27. Where a definition cites external research, the full reference is listed at searchscore.io/the-book/updates.

These terms are the book's spine

Each one is a chapter or part of one. If a definition here raised a question, the chapter that answers it is in the contents.

See every chapter Buy the book