Glossary
AI search terms, defined
The words we use in our research and reports, defined the way we measure them. Where a term is ours, we say so. Where it is in general use, we give the sense we use and a law firm example. Kept current: last reviewed 18 October 2026.
AI fingerprint
The full set of sources an AI model reads before it describes a business: the website, the content written to answer customer questions, the structured data, the data and facts themselves, the statements the models already make about the business, and the earned media that corroborates it. Six elements, three of them owned outright. A SourceNode term. For a law firm it is the practice pages, the directory profiles, the reviews, the state bar record and the videos, read together.
AI search identity management
The practice of managing every source an AI model reads about a business so the model describes it accurately and recommends it when it matters, measured in visits, leads and revenue rather than mentions. A SourceNode term for what we do.
Source diet
The set of domains an engine retrieves for a category of question, ranked by how often each appears. Each engine has its own. For New York car accident questions, ChatGPT reads court and state sites and legal directories, Gemini reads firm websites, Google AI Overviews read YouTube and social profiles. A SourceNode term.
First query
The first, vague question a customer asks an AI model at the start of a decision, before any business is named. The businesses named in the first answer are, with few exceptions, the ones compared and chosen later. A SourceNode term.
Visibility
The share of relevant answers that name the business. If a prompt set produces 100 answers and the business appears in 23 of them, visibility is 23%. We report it by engine and by stage of the customer journey, because the blended number hides where a business is losing.
Share of voice
The business’s share of all business mentions across the relevant answers. A firm can hold 47% share of voice with 23% visibility: named in fewer than a quarter of answers, but usually first and often alone when it is.
Position
The business’s average rank among the businesses an answer names, counted only in answers where it is named. A position near 1 means the engine usually leads with the business.
Retrieval and citation
Retrieval is the engine fetching a page while it builds the answer. Citation is the answer crediting or linking that page. A page can be retrieved and not cited. We count both, because retrieval shows what the engine trusts and citation shows what it tells the customer.
Alignment
The share of claims a model makes about a business that match the business’s verified facts, read claim by claim. The remainder is a correction list with a source for each item.
Fact reconciliation
Making every fact about a business identical across its website, its structured data and every third-party listing that describes it. Practice areas, locations, fee model, consultation terms and published results must read the same everywhere. Where they disagree, the engine hedges or picks another business.
Hedging
What a model does when its sources disagree: it qualifies the claim, names a competitor instead, or declines to recommend. Hedging is the visible symptom of a fact conflict somewhere upstream.
Prompt set
The fixed list of customer questions a business is tracked on, written in the customer’s words and covering every stage from first query to final comparison. It is held constant for the period measured, because changing the questions makes it impossible to tell whether the answer moved or the question did.
Customer journey, in AI search
The stages a customer moves through inside AI answers: first query, shortlist, compare, decide, visit. Each stage has its own questions and its own winners, and we report every metric by stage.
Answer engine optimisation (AEO)
The work of getting a business named, described accurately and cited when someone asks an AI model a question instead of running a search. Generally used for whether a business is cited at all.
Generative engine optimisation (GEO)
The work of shaping how a business is described when an AI model generates an answer about it. Generally used for the accuracy of the description rather than the fact of being cited. We run AEO and GEO as one practice because the same sources decide both.
Earned sources
Third-party sites the engines retrieve and cite in a category: directories, review sites, government and court pages, publications, video and social platforms. In our tracking they produced roughly four times the retrievals of the business’s own site.
Owned sources
The sources a business controls outright: its website, the content it publishes and its structured data. These can be corrected immediately and are where an engagement starts.
Blended score
A single visibility or share of voice number across several engines. It is the number most tools report and it hides the gaps that matter. The same firm held 55% share of voice on ChatGPT and 21% on Google AI Overviews in the same month.
AI referral and AI search attribution
An AI referral is a visit to a business’s site that came from an AI answer. AI search attribution is tracing that visit through to the page it landed on, the lead it produced and the customer it became. Published measurements put the value of an AI referral at several times a traditional organic visit.
Grounding
The step in which an AI model retrieves live sources before writing an answer, rather than answering from training alone. Sometimes called retrieval-augmented generation. Grounding is why the sources can be changed and the answer with them.
Google AI Overviews and AI Mode
AI Overviews are the generated summaries at the top of Google results. AI Mode is Google’s conversational search tab. Both are grounded in retrieved pages, and in our data they read a different source diet from ChatGPT and Gemini, with YouTube first.
Zero-click search
A search that ends without a click on any result, because the answer was read on the results page or in the AI summary. About 60% of searches now end this way, which is why we measure the answer and not only the click.
Structured data
Machine-readable facts added to a page, usually as JSON-LD, so that engines can read names, locations, services and people without guessing. It helps the engines parse the facts. It is not a shortcut to being described well, because consistency across sources does far more.