GEO glossary: AI visibility terms defined
Twenty terms that come up when a firm starts looking at AI visibility, defined in plain language. Written so an owner can read a vendor proposal without a translator.
What do the terms in AI visibility work actually mean?
The vocabulary in this category is young, inconsistent, and partly invented by vendors, which makes proposals harder to compare than they should be. Four labels describe the same trade, several metrics sound interchangeable and are not, and a few terms carry real technical meaning that a buyer needs in order to ask a useful question. The definitions below are the ones used throughout this site. Where an industry term is genuinely contested, that is noted rather than smoothed over.
- Generative engine optimization (GEO)
- The practice of getting a business named, described accurately, and cited inside the answers AI assistants generate. Distinguished from SEO by its target: a sentence inside one synthesized answer rather than a position in a list of links.
- Answer engine optimization (AEO)
- Another name for generative engine optimization, emphasising the goal of being the source an answer engine draws from. Used interchangeably with GEO and LLM SEO by most practitioners.
- LLM SEO
- The same practice as GEO, framed as an extension of search engine optimization into large language models. The label is more common among traditional SEO agencies adding the service.
- Agentic search optimization (ASO)
- A narrower variant of GEO aimed at AI agents that act on an answer — booking, buying, or shortlisting — rather than only presenting it to a person.
- Citation share
- The percentage of a fixed set of tracked prompts where an AI engine names a given firm. The primary metric in GEO, replacing keyword rankings. Only meaningful when the prompt set, the run count, and the scoring rule are held constant between measurements.
- Tracked prompt set
- The fixed list of buyer questions re-run every measurement cycle. Holding it constant is what allows one cycle to be compared with another; a set that changes between cycles makes trends meaningless.
- Clean session
- A signed-out browser session with no account history, memory, or prior context, used so an AI answer is not biased toward businesses the account has already encountered. Standard practice for honest measurement.
- Modal answer
- The answer that repeats when the same prompt is run several times. Logged in place of the single most flattering response, because AI answers vary between runs.
- AI Overview
- Google's AI-generated summary shown above traditional search results. Seer Interactive measured organic click-through falling 61% on queries where one appears, from 1.76% to 0.61%.
- Zero-click search
- A search that is answered on the results page, so the user never visits a website. AI Overviews increased the share of these substantially.
- Retrievability
- Whether an AI engine can actually fetch and parse a page. Governed by robots.txt directives, JavaScript dependence, and server response. The most common single point of failure in AI visibility work.
- GPTBot / OAI-SearchBot
- OpenAI's crawlers. GPTBot gathers training data; OAI-SearchBot supports ChatGPT's search and browsing. Blocking either in robots.txt removes a site from that pathway.
- Google-Extended
- A robots.txt token controlling whether a site's content may be used for Google's generative AI products. It does not affect ordinary Google Search indexing.
- llms.txt
- A proposed plain-text file at a site's root that summarises what the site contains and points to its key pages, intended for large language models. Support is uneven; the cost of publishing one is close to zero.
- JSON-LD
- The structured-data format search and AI engines read to learn facts about a business without inferring them from prose. Common types for a local firm are ProfessionalService, LocalBusiness, Service, and FAQPage.
- Entity strength
- How consistently and unambiguously a business is described across the web. The 5W Public Relations study measured an eight-fold entity-strength multiplier on citation share, the largest single signal effect it recorded.
- Direct-answer structure
- A page pattern where each heading is the question a buyer asks and the paragraph immediately below answers it in roughly 50 to 150 standalone words, giving an engine a passage it can quote without editing.
- Hallucination
- An AI engine stating something false with confidence — an invented service area, a wrong price, two similarly named firms merged into one. Usually corrected by making the true fact easy to retrieve and consistent everywhere, not by contacting the engine.
- Digital PR / citation outreach
- Earning mentions of a business on credible third-party sites. Engines weight corroboration, so a firm mentioned nowhere but its own website is a firm with a single source.
- Aggregator
- A directory such as Yelp, Angi, Houzz, Thumbtack, HomeAdvisor, Porch, or Nextdoor that ranks and lists local providers. Aggregators currently absorb a large share of AI citations in local service categories, with Yelp at an estimated 9.0% and Angi 7.0% in the 5W study of six non-remodeling categories.
- NAP consistency
- Keeping a business’s name, address, and phone number identical across every listing and profile that mentions it. Inconsistency splits an engine’s understanding of the entity, which weakens the corroboration signal that carries the largest measured effect on citation share.
- E-E-A-T
- Experience, expertise, authoritativeness, and trust — the framing Google’s search quality guidelines use for source credibility. Not a direct ranking input, but a useful shorthand for the signals that also make a page quotable: a named author, a visible updated date, and corroboration from sources other than the business itself.
- AI visibility tracking tool
- Software that runs a set of prompts against AI engines on a schedule and reports where a brand is mentioned. Otterly, Profound, and the Semrush AI toolkit are examples. Useful for scale; worth asking any vendor whether a reported number comes from a tool, from manual runs, or from a mix, and whether you can see the underlying answers.
- Cost per lead
- What a firm pays, on average, for one inbound enquiry from a given channel. The comparison that matters for AI visibility work is cost per lead from AI-sourced traffic against cost per lead from shared directory leads, measured over a year rather than a month, since content and citations persist while paid leads stop with the budget.
Which of these terms actually matter when comparing vendors?
Four of the terms defined above decide whether a vendor comparison is meaningful at all. Citation share matters because it is the number a vendor will report and it is meaningless without the next two. Tracked prompt set, because a set that changes between cycles makes every trend unverifiable. Clean session, because a signed-in session can surface a firm the account already visited and quietly inflate a result. And retrievability, because a site the engines cannot fetch will not improve regardless of what anyone is paid. The rest are useful context. These four are where money is won or wasted.
The practical version of this is a short list of questions rather than a vocabulary test. Twelve questions worth asking any vendor puts them in the order that disqualifies fastest.
Where do these numbers come from?
Every figure on this page comes from a named third-party study, listed below. None of the numbers are ours. Full samples, the scope limits each study states, and the four commonly repeated statistics removed from this site rather than published with a vague attribution are all set out on the sources page.
- 5W Public Relations, The Local Services AI Visibility Crisis 2026: national chains, platform aggregators, and the structural exclusion of independent operators, April 2026. View the study ↗
- Seer Interactive, AIO Impact on Google CTR: September 2025 Update, 4 November 2025. View the study ↗
Full sources, samples, scope limits, and the four removed figures →
What else is worth defining?
Is GEO an established term or vendor jargon?
Both, honestly. Generative engine optimization is now the most widely used label and appears across trade press, research, and vendor marketing, so it is established enough to search for. It is also barely two years old, has no governing body, no certification, and no agreed benchmark. Treat the term as real and the surrounding claims of expertise as unregulated.
What is the difference between citation share and share of voice?
Citation share counts the tracked prompts where an engine names your firm, measured against a fixed list. Share of voice usually counts mentions across a broader and less controlled sample, often including social and press. Both can be useful. Only the first can be reproduced by someone who does not work for the vendor reporting it, which is why it is the metric used here.
Do I need to know any of this to hire someone?
No, but knowing four terms changes the conversation. If you can ask which engines are scored, how large and how stable the prompt set is, how many runs per prompt, and what counts as a citation, you will learn more in five minutes than a capabilities deck will tell you in an hour.
Where should you read next?
The glossary is context. These three are the pages that use it.
Austin Brewer founded AI Visibility Partners in 2026, and also founded the interior design studio Astratto Design — a connection disclosed on who we work with because this practice sells to design firms. hello@aivispartners.com