The field is young enough that its words are still settling. These are the definitions we work by, with our own measured numbers where we have them. Each entry is written to survive being quoted on its own.
AEO · Answer engine · GEO · AI Overviews · Answer volatility · Share of voice · Citation · Citation absorption · llms.txt · Query fan-out · Entity · Structured data · Zero-click · Grounding · Noise floor · Reading
AEO (Answer Engine Optimisation)
The work of improving the likelihood that AI assistants name and recommend a business when customers ask buying questions. It is not a checklist: every market's engines retrieve differently, so the work is diagnose, fix, re-measure. Distinct from SEO, which optimises pages for ranked lists rather than claims for generated answers. Not the customs status that shares the acronym.
Answer engine
Any system that answers a question directly instead of returning links: ChatGPT, Perplexity, Google AI Overviews, Copilot. For buying questions they often name only a few businesses directly in the answer, which concentrates attention on the names that appear.
GEO (Generative Engine Optimisation)
A label often used interchangeably with AEO in academic papers and US agencies, though some use it more broadly for all generative-response visibility work. If a vendor treats GEO and SEO as the same service, ask which generated answers they have measured.
AI Overviews
The AI-generated summaries Google shows above normal results. They cite sources and, for commercial questions, name businesses. Google began rolling out dedicated generative-AI performance reports in Search Console in mid-2026; until they reach your site, and for engines beyond Google, measuring the answers directly remains the cleanest read.
Answer volatility
How much an engine's answer changes when the same question is asked again. We measured identical questions changing their verdict 19% of the time on the same day. Volatility is why one-off checks mislead and why measurement must be repeated to mean anything.
Share of voice (SOV)
The share of repeated answers that name a given business, counted once per answer. In one UK storage category we measured (75 questions, three engines, July 2026), a client held 10% while the category leader held 44%. The number we lead with, because it makes visibility comparable across brands and across time.
Citation
A source an answer engine lists or links when composing an answer. Being cited is not the same as being used: engines can cite a page yet take nothing from it, or absorb its facts without prominent attribution.
Citation absorption
Whether an answer actually uses a page's facts and language, rather than merely listing it as a source. The distinction matters because a page can be cited yet contribute nothing, and absorbed claims persuade even when nobody clicks.
llms.txt
An emerging, optional convention: a small markdown file at /llms.txt summarising a site for AI systems, what the company does, key pages, one line each. How widely engines consume it is still settling. It is cheap to do well, and we have seen auto-generated ones ship placeholder text like 'Welcome to WordPress' as a company's machine-facing summary.
Query fan-out
The hidden searches an engine runs before composing one answer: a single question about the best provider can spawn related sub-queries about prices, comparisons and locations. Coverage of those hidden branches, not the visible question alone, influences which sources are retrieved and which names are cited.
Entity
The machine's idea of a specific thing: this company, not one that shares its name. Engines confuse entities constantly, and a business whose name collides with others can be described by a stranger's reputation. Entity establishment is the work of reducing that ambiguity until the machines consistently describe the right business.
Structured data (schema)
Machine-readable facts embedded in a page: this is a product, this is its price, this business serves these places. Engines can use such facts without inference, and 'can' is the honest word, Google's own documentation avoids 'will'. But a price stated in schema is at least readable without guessing; a price drawn by a widget may never be read at all.
Zero-click
A search resolved inside the answer, with no visit to any website. The answer becomes the landing page, which is why content must give engines enough material to describe and qualify a business accurately without the click.
Grounding
When an engine composes its answer from retrieved sources, the live web, documents, or a private corpus, rather than from memory alone. Grounded answers change as retrieval changes, which drives much of the short-term movement we measure.
Noise floor
The measurement error of the instrument itself: re-classifying identical answers should produce near-identical results. Ours re-scores at a 1.6% disagreement rate. That is one component of overall error, sample size and repetition still matter, but it is what separates instrument noise from real movement.
Reading
Our term for one full measurement of a market: real buyer questions asked repeatedly across engines, brands counted once per answer, published with the date and method. A reading is evidence; a dashboard score is an opinion.
Updated July 2026 · related: what is AEO, answer volatility, how we measure