The words people use about AI search, defined properly.

A tool that replies in prose instead of returning ten blue links. ChatGPT, Claude, Gemini and Perplexity are the four Prism asks.

Answer engine optimisation, or generative engine optimisation. Two names for one young discipline: getting named and cited inside an AI answer rather than ranked on a results page. The work starts with measuring where you stand today.

How often your brand gets named across a set of answers, measured against every other brand named alongside you. Being in a third of the answers means little until you know whether a rival is in all of them.

AI Visibility → Share of Voice

How early your brand appears once an engine starts naming names. Being mentioned first and being mentioned ninth both count as a mention.

AI Visibility → Share of Voice

A URL an answer engine linked to while replying. Prism logs every one and sorts them into your pages, competitor pages and everything else.

AI Visibility → Citations

A prompt where an engine cited somebody else and not you. The export is an outreach list.

AI Visibility → Citation Gap

The follow-up searches an engine runs before it answers you. Reading them tells you what the model thought your question was really about, which is often not what you assumed.

AI Visibility → Fan-Out

One question you want the engines asked, written the way a customer would type it. Caps are per property: 10 on Starter, 50 on Pro, 100 on Enterprise.

AI Visibility → Prompts

The unit that meters answer-engine calls, and the only thing on the platform that meters anything. A live web-search answer costs 4 and a cached one costs 2. Reports, crawls and GA4 consume none.

Two or more of your own pages splitting the impressions for one query, so neither builds enough authority to win it. Prism finds these by comparing query meaning rather than exact strings.

Search Performance → Cannibalization

Queries ranking just off the positions that get clicked, where a small improvement pays disproportionately.

Search Performance → Opportunities

The queries you gained and lost against the previous period, and where the clicks went. A flat total often hides a large gain and a large loss cancelling each other out, and this is the report that pulls them apart.

Search Performance → Click Gap

A dated handover of the URL Google predominantly returns for one query. Prism counts one only after the incoming page has been the highest-impression page on at least 7 of the next 10 days that carry data, so a one-day flip stays out of the list, and a change too recent to judge is marked as pending. It records that the page changed and when, not why.

Search Performance → Keywords → Analysis

The URL carrying the most impressions for a query on a given day. Impressions rather than clicks, because impressions are what Google chose to show; on a contested query the two URLs can trade places day to day without anything having happened.

Search Performance → Keywords → Analysis

A saved filter over your queries or your URLs, built from plain rules: contains, does not contain, equals, starts with and ends with, matched against any or all of them. Pick one in the toolbar and every Search Performance report stays scoped to it.

Settings → Segments

Google carries on revising its numbers for two to three days after the fact. Prism re-fetches the last five days on every sync rather than trusting the first answer, so yesterday stops being slightly wrong.

The one-off pull of everything Google still holds on the day you connect a property. Sixteen months, one month per pass, resumable, and the only chance you get at it.

The watch for the drop nobody wrote a rule for. Each day's clicks, impressions, click-through rate and position are compared with the last 28 days of history, using the middle value rather than the average so one freak day cannot drag the baseline, and the typical day-to-day wobble is measured the same way. A day that lands about three of those wobbles outside the baseline is raised as an incident. It builds about a week of history before it commits to an opinion, and it skips days with too little volume to judge.

Actions → Action Center

A query turned into numbers that stand for its meaning rather than its spelling, so "cheap running shoes" and "affordable trainers" come out as the same intent. Clustering, cannibalization and the AI ↔ SEO Bridge all run on these.

Crawl findings classified by their effect on traffic: broken while earning, recoverable losses, earning despite issues, housekeeping. A critical error on a page nobody visits is housekeeping.

Site Health → Fix First

An indexing instruction sent in the HTTP response header rather than in the page markup. It overrides nothing you can see in view-source. A page can look indexable and quietly not be.

Site Health → Technical

A score for how much of your site's own link equity reaches a page, computed from your internal links alone. Useful for spotting important pages nothing links to.

Site Health → Internal Linking