Glossary v1 · 2026-07-14
Glossary
Terms this site uses when it measures AI visibility, defined once and used consistently. Each
entry opens with a self-contained definition and closes with a plain-language analogy.
Compounding layer
The compounding layer is the accumulated body of retrievable material associated with a brand,
including citations, mentions, reviews, definitions, and original data, that determines whether
AI engines such as ChatGPT, Claude, Perplexity, Gemini, and Google's AI Overviews recommend that
brand. It behaves like capital rather than like campaigns: references beget retrieval, retrieval
begets further references, and the value accrues whether or not anyone is watching. It is
broader than the content a brand publishes; the strongest known predictor of AI brand mentions
is brand search volume, part of the wider footprint. The term describes the layer that now sits
between a company's marketing and its buyers whenever an AI engine composes the answer.
ELI5: Imagine the internet keeps a big scrapbook about your company: every
review, every mention, every time someone links to your data. When someone asks an AI "what
should I buy?", the AI flips through that scrapbook to decide who to recommend. The thicker your
scrapbook, the more often you get named, and every new page makes the next page easier to earn.
That scrapbook is the compounding layer.
Answer layer
The answer layer is the set of responses that AI engines such as ChatGPT, Claude, Perplexity,
Gemini, and Google's AI Overviews give to buying questions, which increasingly replaces the
search results page as the place buyers form shortlists. A brand present in the answer layer
gets named when buyers ask what to buy; a brand absent from it is often never considered, and
the loss is invisible in that brand's analytics because no click ever happens. Presence in the
answer layer varies by engine, by question phrasing, and across repeated runs of the same
question.
ELI5: People used to ask Google and get ten blue links. Now they ask an AI and
get one answer that names a few companies. That answer is the answer layer. If you are in it,
buyers hear about you. If you are not, they never even know you exist, and you never find out
either, because nobody clicked anything you could measure.
Above the funnel
Above the funnel is the stage of a buying decision that takes place before a buyer enters any
company's marketing funnel, and it is increasingly where AI engines such as ChatGPT, Claude,
Perplexity, Gemini, and Google's AI Overviews shape the shortlist. When a buyer asks an AI
engine what to buy, candidates are named and eliminated before any website visit, form fill, or
first touch that a funnel could record, so the outcome is invisible to conventional funnel
analytics. What a brand carries into this stage is its accumulated footprint, not its
funnel-stage tactics: a company can convert well at every stage it measures and still lose above
the funnel, because buyers who never saw it in an AI answer never arrive.
ELI5: Your funnel is like the front door of your shop: you can count everyone
who walks in. Above the funnel is the conversation happening down the street, where an AI tells
the buyer which three shops are worth visiting. If your shop is not on that list, the buyer
never walks down your street, and your door counter shows nothing wrong.
A brand footprint is the full accumulated public evidence of a brand's existence: branded search
demand, mentions across the web, reviews, citations, press coverage, and definitions or data
attributed to the brand. It is the measured driver of AI recommendations, as distinct from any
single input such as published content. Kevin Indig's Growth Memo analysis found brand search
volume to be the single strongest predictor of how often AI engines mention a brand. A brand
footprint takes sustained time to build and is difficult for competitors to replicate quickly.
ELI5: Think of a restaurant. The menu can say "best mango pudding in town," but
that is not why you believe it. You believe it when there is a queue outside, reviews online,
people talking about it, raving social media posts, and other restaurants comparing themselves
to it. A brand footprint is that same public evidence around a company. AI engines are more
likely to trust the restaurant everyone has heard of than the one that only printed a better
menu yesterday.
Citation tracing
Citation tracing is the practice of identifying which sources an AI engine cited or drew on when
producing an answer, and following those citations back to the underlying pages. It turns an
opaque recommendation into an inspectable one: instead of knowing only that an engine named a
brand, you can see which reviews, comparison pages, forum threads, or datasets the answer leaned
on. Citation tracing is what separates a study of AI visibility from a leaderboard, because it
addresses why a brand appears, not just whether it does.
ELI5: When an AI recommends a brand, it is like a student turning in an essay.
Citation tracing is checking the bibliography: which websites did the AI actually read before
writing that answer? Once you can see the sources, the recommendation stops being magic and
becomes something you can study.
Answer variance
Answer variance is the degree to which an AI engine's responses to the same question differ
across repeated runs. AI engines do not return one fixed answer per question: the brands named,
their order, and the cited sources can all change from run to run, and SparkToro's research
documents high inconsistency in AI brand recommendations. Because of answer variance, any single
AI response is a sample, not a ranking, and any measurement of AI visibility that does not
report spread across repeated runs is unreliable.
ELI5: Ask an AI the same question five times and you can get five different
answers, like rolling dice instead of reading a scoreboard. So one answer tells you almost
nothing. If someone shows you a chart of AI recommendations built from asking each question only
once, be suspicious: they rolled the dice one time and called it the truth.