Citation, Mention, or Recommendation: A Taxonomy for AI Search Visibility

The industry uses "AI visibility" as a catch-all, but not all appearances in an AI answer carry the same weight. A link in a footnote differs from a brand name dropped mid-sentence, which differs again from an explicit "You should try X." Mixing these three signals into one metric obscures what's actually happening—and what to do about it.

This article proposes a working taxonomy: Citation, Mention, and Recommendation. Each has distinct characteristics, measurement challenges, and optimization implications.

Why Terminology Matters

When Gartner predicts that traditional search volume will drop 25% by 2026 due to AI assistants (Gartner, 2024), teams rush to measure "AI presence." But what are they actually measuring?

Some tools count any brand reference. Others only track clickable links. A few attempt sentiment analysis. Without shared definitions, comparing reports becomes impossible—and internal alignment suffers.

Clear categories let you:

• Set meaningful KPIs for each visibility type
• Allocate resources to the right optimization tactics
• Benchmark against competitors on like-for-like metrics

The Three Visibility Types

Comparison Table

Aspect	Citation	Mention	Recommendation
Click potential	High (link present)	None (name only)	Low (no link, but strong intent)
Verification	Easy (URL visible)	Moderate (text parsing)	Moderate (intent classification)
SEO parallel	Backlink	Brand mention	Featured snippet / answer box
User trust signal	"This is the source"	"This exists"	"This is the answer"
Optimization lever	Authoritative content, structured data	Brand awareness, entity salience	Direct answers, product-market fit

Measurement Implications

Citations Are the Easiest to Track

Perplexity displays numbered footnotes. Claude and ChatGPT surface links when web search is active. These are machine-parseable. The challenge: citation formats vary by engine and query type.

Mentions Require NLP

Detecting "Mentio" in flowing text seems simple until you account for partial matches, abbreviations, and context. Is "mentio" the Latin word or the brand? Entity disambiguation matters.

Recommendations Need Intent Classification

A sentence can contain your brand name and still not recommend you: "Mentio is one option, but it may not suit enterprise needs." Distinguishing endorsement from neutral mention—or even criticism—requires more than keyword matching.

According to research from the University of Waterloo, large language models exhibit detectable patterns when expressing preferences versus stating facts (University of Waterloo, 2023). This distinction is measurable, but not trivial.

Optimization Differs by Type

To earn citations: Publish original research, maintain accurate structured data, ensure your pages are crawlable and load fast. Traditional SEO hygiene still applies.

To earn mentions: Build brand salience. Get covered in sources the models train on. Ensure consistent NAP (name, address, phone) across the web.

To earn recommendations: Answer buyer questions directly. Demonstrate clear product-market fit. Models recommend what they "believe" solves the user's problem—shaped by the corpus they've seen.

The tactics overlap, but the emphasis shifts. A citation strategy focuses on content quality and linking. A recommendation strategy focuses on positioning and direct-answer content.

The Funnel View

Think of it as a funnel. Many queries generate mentions. Fewer generate citations. Fewer still generate explicit recommendations. Each layer represents a different user moment—and a different business value.

What This Means for Dashboards

A single "AI visibility score" hides more than it reveals. Better to track:

1. Citation count per engine, per query cluster
2. Mention share-of-voice vs. competitors
3. Recommendation rate for high-intent queries

Separating these lets you diagnose problems. Dropping citations but steady mentions? Check your content freshness. Strong mentions but no recommendations? Your positioning may be unclear.