Glossary of AI authority terms
Every term the AI Authority Index™ uses, defined in short sentences, A to Z. Of these, 17 have a page of their own, with what the term tells a buyer or a marketer and how it appears on the Index.
A
- AI engine
- An AI assistant a buyer can ask for advice. Every edition of the Index asks the same five: ChatGPT, Claude, Copilot, Gemini, and Perplexity.
- AI Famous™
- Named in 30 percent of answers or more, and recommended for its benefits more often than cautioned against.
- More on AI Famous
- AI Infamous™
- Named in 30 percent of answers or more, and cautioned against more often than recommended.
- More on AI Infamous
- Answer
- One engine's reply to one buyer question. An edition reads every answer and counts the companies each one names.
B
- Bad Press™
- Named in 15 to 29 percent of answers, and cautioned against more often than recommended.
- More on Bad Press
- Band
- A group of companies the sample cannot tell apart. The Index groups companies by the six cells of the Star Chart in place of bands, and no row carries a rank number.
- Buyer question
- A question in the words a buyer in the market uses, such as which lender to choose for a car loan. No buyer question names a company, so every company in an answer was put forward by the engine.
C
- Cited source
- A page an engine names or links to as the basis for its answer, such as a regulator, a comparison site, or a news outlet. Influence reads the sources the answers cite.
- Composite
- Another name for the CRISP™ score. A company off the Leaderboard is marked No composite, because the market's answers measure only some of its factors.
- Confidence interval
- The range a score could fall in, given the size of its sample. The page draws it as a bracket over a factor's bar, labelled the range the sample supports.
- Consensus
- Whether the engines agree about a company. When one engine names the company on a question, Consensus asks whether the other engines that answered name it too.
- More on Consensus
- Constituent
- A company on the Leaderboard.
- CRISP
- Third Hemisphere's proprietary measure of how AI engines treat a company when buyers ask them for help. It scores five factors, and its name comes from their initials: Consensus, Reputation, Influence, Source, and Presence.
- More on CRISP
- CRISP score
- A company's overall score out of 100, built from its five factor scores at the weights the Method shows. On the Index, only a company with all five factors measured carries one.
- More on CRISP score
- Cult Favourite™
- Named in under 15 percent of answers, and recommended for its benefits in the answers that reach it.
- More on Cult Favourite
- Cut
- The number of answers a company must be named in to join the Leaderboard: 30. Below it, the sample is too small to read Reputation and Influence reliably.
E
- Edition
- One publication of the Index, covering one market or a few related markets, each scored on its own answers. An edition passes the publication gate before it publishes, and it stands until the next edition replaces it.
F
- Floor
- The lowest score a factor can carry: 1. A company whose site crawl reached nothing it could read carries Source at the floor, and its row says so.
- Full read
- The explanation a Leaderboard company's row opens. It says what caused each of the company's scores and what could move them.
I
- Indicative
- An estimate drawn from the evidence at hand, short of a full measurement, which no Index edition prints. When a factor in the free AI Authority Check is indicative, the check says what produced it and calls its total an Initial CRISP score.
- Influence
- Whether trusted third parties stand behind what the engines say about a company, and how many different sources do.
- More on Influence
L
- Leaderboard
- Every company the engines named in 30 or more answers. Each company on it carries all five factors and a CRISP score.
M
- Market
- One category of product that buyers shop for, such as personal loans. The Index scores each market on its own answers and combines no figure across markets.
N
- N (sample size)
- The number of answers a reading is drawn from. The larger the N, the narrower the reading's confidence interval.
O
- Off the Guest List™
- Named in under 15 percent of answers, and cautioned against more often than recommended when it is.
- More on Off the Guest List
P
- Presence
- How often the engines name a company when a buyer describes the need without naming it. Presence is the share of answers that name the company, and it places the company across the Star Chart.
- More on Presence
- Publication gate
- The conditions an edition must pass before it publishes, such as one row per company and all five factors on every company on the Leaderboard. Nothing publishes until every condition passes.
R
- Reputation
- Whether the engines recommend a company more often than they caution against it, across the answers on what to choose and what to avoid. At 50, the line up the Star Chart, recommendation and caution are equal.
- More on Reputation
- Rising Star™
- Named in 15 to 29 percent of answers, and recommended for its benefits more often than cautioned against when it is.
- More on Rising Star
S
- Sample
- The set of answers an edition collected. Every reading describes what the engines said in that sample.
- Site crawl
- A read of a company's own website, page by page, as an AI engine would reach it. It shows how readily an engine can reach, read, and quote the site.
- Source
- Whether a company's own site stands up as the record of what is true about it, and whether an engine can reach, read, and quote it.
- More on Source
- Star Chart
- A chart that places each company by two factors: Presence across and Reputation up. Lines at 15 and 30 percent of answers across, and at 50 up, divide it into six cells.
- More on Star Chart
- Statistically significant
- Said of a difference too large to come from chance in which answers the sample happened to collect. The Index claims an order between companies only where their difference is significant.
U
- Under the Radar
- Every company the engines named in fewer than 30 answers. Each is listed with the number of answers that named it, and none carries a CRISP score.
W
- Wave
- One round of collection: every buyer question in a market put to every engine, with every answer kept. An edition is scored on the answers its wave collected.