AI Visibility Metrics: How Public Relations Success Is Measured in Generative Search

Artificial intelligence is changing how people discover companies, compare products and form opinions about brands. Instead of reviewing a page of traditional search results, users increasingly ask tools such as ChatGPT, Claude and Google Gemini for a direct explanation or recommendation. The sources selected by these systems can influence which organizations appear credible, relevant or authoritative.

Research from Muck Rack helps explain why this shift matters to communication teams. Its “What Is AI Reading?” project has examined millions of links cited in AI-generated answers. The May 2026 edition analyzed more than 25 million links across 17 industries and found that earned media accounted for 84% of citations, while paid and advertorial content represented only 0.3%. Journalism alone supplied 27% of cited sources.

These results require an important clarification. The frequently repeated claim that 96% of AI citations come from earned media does not match the latest report. In the May 2026 data, 96% refers to the proportion of ChatGPT responses that included citations. Earlier Muck Rack research found that 95% of citations came from non-paid media and 89% from earned media. The precise percentages have changed between editions, but the broader conclusion remains consistent: independent, authoritative coverage plays a major role in generative search visibility.

Why AI Visibility Metrics Matter to Public Relations

Traditional public relations measurement often focuses on media placements, potential audience reach, impressions and social engagement. Those indicators can still provide useful information, but they do not show whether an AI platform is using the coverage to describe or recommend a brand. A story may attract limited direct traffic while becoming an influential source for AI-generated answers.

AI visibility metrics address this gap by examining how a brand appears within responses to relevant questions. Measurement may cover the frequency of mentions, the sources supporting those mentions, the position of the brand relative to competitors and the language used to describe it. This connects communication activity with the information customers, investors and other audiences encounter in generative search.

The results must be interpreted carefully. AI platforms differ in how frequently they cite sources and which domains they prefer. According to Muck Rack, ChatGPT included citations in 96% of the responses studied, Gemini in 82% and Claude in 55%. A measurement program should therefore track several platforms, repeat tests over time and document the prompts used instead of treating one isolated answer as a stable result.

Four Core AI Visibility Metrics for Communication Teams

The first metric is AI share of voice. It measures how frequently a brand appears compared with selected competitors in responses to non-branded industry prompts. A software company, for example, might test questions about the best tools for a particular task without mentioning its name. The resulting KPI can express the brand’s percentage of all qualifying mentions across a defined prompt set.

The second metric is positive citation rate. Visibility alone is not necessarily beneficial because an AI system might mention a company in connection with a controversy, limitation or warning. Teams can classify relevant appearances as recommended, favorable, neutral or negative and calculate the percentage that communicates a positive or credible position. The assessment should consider the full context rather than applying sentiment analysis to the brand name in isolation.

The third metric is key message pull-through. This evaluates whether AI-generated answers reproduce the factual differentiators a communication strategy is intended to establish. If a company emphasizes verified environmental performance, for example, the evaluation can determine whether sustainability-related prompts produce that association. Claims should be specific, documented and independently supportable before they are included in the measurement framework.

The fourth metric is source authority. Instead of counting every placement equally, teams examine which publications, government resources, academic materials and specialist outlets actually appear in AI citations. Authority cannot be reduced to a single domain score. Relevance to the topic, editorial standards, expertise, recency and demonstrated citation behavior are all important.

Building a Reliable AI Measurement Framework

A useful framework begins with a controlled prompt library. Teams should group prompts by audience intention, such as discovery, comparison, recommendation, reputation, product features and industry trends. Each group should contain branded and non-branded questions, with wording variations that reflect how real users might ask for information.

Prompts should be tested on a consistent schedule across the selected AI platforms. The record for each test should include the date, platform, model or product version when available, full response, citations, brand mentions, competitors and message classification. Repeating the process matters because responses can change when models, retrieval systems or online sources are updated.

The KPI definitions must also remain stable. Teams should decide in advance what counts as a mention, recommendation, positive citation or successful message match. Human review is valuable for ambiguous cases because automated sentiment tools may misread comparisons, qualifications or industry-specific language. A documented scoring guide makes results more comparable between reporting periods.

How Earned Media Supports Generative Engine Optimization

Generative Engine Optimization, commonly called GEO, is the practice of improving how an organization is represented in AI-generated answers. It complements SEO rather than replacing it. Search optimization helps information become discoverable, while GEO pays particular attention to whether machines can identify, understand and support claims with credible sources.

Earned media contributes to GEO because independent reporting can provide third-party validation. However, the evidence does not establish that AI systems categorically reject all corporate or social content. Muck Rack categorizes citations from journalism, third-party corporate content, first-party corporate content, press releases, academic research, government sources, social platforms and aggregators. The mix varies according to the platform and the type of question.

Recency also matters. Muck Rack reported that more than half of journalism citations in its updated research came from the previous year, with citations peaking soon after publication. This suggests that communication programs need a continuing flow of accurate, newsworthy information rather than occasional bursts of publicity.

From Media Volume to Demonstrable Influence

Communication teams can use these findings to improve content without turning every document into material written only for machines. Press releases, research reports and executive articles should provide clear dates, direct statements, useful headings, transparent methodology and verifiable data. Muck Rack has reported that cited press releases tend to contain more statistics, bullet points and objective language than releases that are not cited.

Placement quality should also be evaluated in context. One respected specialist publication may influence an industry-specific response more than a large number of irrelevant mentions. At the same time, no individual outlet guarantees AI inclusion. The objective is to build a credible body of consistent evidence across sources that audiences and retrieval systems can understand.

The most mature measurement programs will combine AI visibility metrics with established business and communication indicators. Media quality, referral traffic, search visibility, audience trust, leads and reputation research still matter. AI citations add another layer: they reveal how public information is being assembled into answers that may shape future decisions.