---
title: "Presence Rate, Mention Rate, Citation Rate: The Metrics of AI Visibility | Kitbase Blog"
description: "The metrics of AI visibility explained: presence rate, mention rate, and citation rate, how each is defined, when they diverge, why per-engine breakdowns matter, and how to compare them honestly."
canonical: https://kitbase.dev/blog/ai-visibility-metrics/
---

The three core **AI visibility metrics** are **presence rate**, **mention rate**, and **citation rate**. Presence rate is the share of AI answers that either name your brand or cite your domain; mention rate is the share that *name* you in the answer text; citation rate is the share that *link* your domain as a source. Presence is the union of the other two.
An answer counts as present if either is true. Everything useful in AI visibility measurement is built on getting these three definitions exactly right and applying them identically to every brand, engine, and day.

That last part is where most people go wrong. It's easy to eyeball an answer and say "we showed up." It's much harder to say *what showing up means* in a way that's consistent enough to trend over months and compare against competitors. This post defines each metric precisely, shows when they diverge and what each divergence tells you, and explains why definitional consistency is the whole game.

## The three metrics, defined precisely:

For a given prompt run against a given engine, the engine returns an answer: a block of text plus (usually) a list of cited source URLs. Every metric below is a rate — a count of qualifying answers divided by the total number of answers in the window.

- **Mention rate:** the share of answers whose *text* names your brand or one of its aliases. "Kitbase is a good option for…" is a mention. Matching is done against the full alias list you configure, because engines use short forms, product names, and domains interchangeably.
- **Citation rate:** the share of answers that *cite your primary domain* as a source. This is measured against the answer's citation list, not its prose: if `kitbase.dev` appears among the linked sources, that answer counts as a citation, whether or not the text names you.
- **Presence rate:** the share of answers that mention you **or** cite you. It's the union: an answer with a mention, a citation, or both counts once. Presence is the closest thing AI visibility has to a "ranking", the single headline number you trend over time.

**How one AI answer resolves into mention, citation, and presence**

```mermaid
flowchart TD
  A["AI answer to one prompt"] --> B{"Names your brand<br/>or an alias?"}
  A --> C{"Cites your<br/>primary domain?"}
  B -->|"Yes"| D["Counts as MENTIONED"]
  C -->|"Yes"| E["Counts as CITED"]
  D --> F["PRESENT<br/>mentioned OR cited"]
  E --> F
  B -->|"No"| G["Not mentioned"]
  C -->|"No"| H["Not cited"]
```

The reason to separate mention from citation rather than collapsing both into one "we appeared" flag is that they are *different achievements earned by different work*, and they fail for different reasons.

## When the metrics diverge and what each gap means:

If mention rate and citation rate always moved together, you'd only need one number. They don't. The gaps between them are the most actionable signal in the whole dataset.

**Mentioned but not cited.** The AI talks about you from what it already knows meaning you're established enough that it names you without needing to look you up. Good for awareness, but you don't control how you're described, and buyers verifying the claim click someone else's link. High mention rate + low citation rate = the engine knows you exist, but your pages aren't winning the retrieval step.

**Cited but not named.** The engine links your domain as a source, it pulled a stat, a definition, or a comparison from your page but the prose recommends someone else, or names you only as one of several references. Your content is quotable enough to retrieve, but not persuasive enough to make you the answer. This gap often shows up when you publish good reference content but weak positioning: you're feeding the answer that sells a competitor.

**Both, consistently.** The engine names you *and* cites you is the strongest position, because you're both the recommendation and the verifiable source behind it.

The tactical read is direct. If you want to lift mention rate, the work lives in the training and third-party layer: consistent entity naming, coverage on the sites engines already trust, being discussed in your category. If you want to lift citation rate, the work lives in the retrieval layer: crawlable, quotable, extraction-friendly pages that engines pull in at answer time. The [GEO pillar](/blog/what-is-generative-engine-optimization/) walks through both sets of tactics; the point here is that your metric gaps tell you *which* set to prioritize.

| Pattern | What it means | Where to work |
|---|---|---|
| High mention, low citation | Engines know you, your pages don't get retrieved | Retrieval-layer pages: quotable, crawlable, extractable |
| Low mention, high citation | Your content is a source, but you're not the pick | Positioning, comparison content, entity strength |
| High both | You're the recommendation and the source | Defend and expand to more prompts |
| Low both | Effectively invisible for this prompt | Start with prompt fit, then content |

## Per-engine breakdown: one number hides four different games.

A blended presence rate across all engines is a convenient headline and a misleading one. Perplexity, Gemini, Claude, and ChatGPT build answers through *different retrieval systems*, so your presence rate genuinely differs between them (often dramatically). Perplexity grounds and cites almost every answer, so citation rate tends to run higher there; an engine leaning more on model knowledge may name you often while citing rarely. Averaging those together produces a number that describes no engine in particular.

Track all three metrics **per engine, over time**. The per-engine view is what tells you *where* you're strong, where you're invisible and where a content change actually moved something, because a tactic that lifts retrieval-based engines may do nothing for a knowledge-based one. This is also why the same prompt can name you on one engine and omit you on another on the same day: they aren't running the same machine. (That per-answer variability is its own topic, see [why you can't trust a single AI answer](/blog/ai-answers-non-determinism/).)

[Kitbase AI Visibility](https://docs.kitbase.dev/ai-visibility) reports presence, mention, and citation rates both blended and split per engine, and plots a line per engine over your run history so you can see which engines are picking you up and which are trailing.

## Beyond the big three: framing metrics.

Presence, mention, and citation tell you *whether* you appear. A second tier of metrics describes *how* you appear when you do:

- **Recommended rate:** the share of mentions where the answer actively recommends you, versus merely listing you. Being named in a "here are some options" sentence is not the same as being the pick.
- **Sentiment:** whether each mention is positive, neutral, or negative. Presence with negative sentiment is a problem you'd never see from a presence rate alone.
- **Average answer rank:** when an answer presents a ranked list, your typical position in it. Moving from fifth-mentioned to first-mentioned is progress a binary "present" flag can't capture.
- **Share of voice:** your presence rate normalized against the total presence of every tracked brand in the same prompts. Being mentioned in 40% of answers means one thing if your top competitor sits at 10%, another entirely if they're at 90%. Share of voice is the competitive version of presence; we cover it in depth in [AI share of voice](/blog/ai-share-of-voice/).

These are derived during the same answer-analysis pass that extracts mentions, so they come for free once you're measuring presence but they only make sense layered on top of a solid presence number, not instead of it.

## Why definitional consistency is the whole game:

Here is the failure mode that quietly ruins AI visibility measurement: **comparing numbers that were defined differently.**

If you compute your own "visibility" as presence (mention OR citation) but glance at a competitor and count only whether they were *named*, you're comparing a union against one of its halves and you'll conclude you're winning when you may not be. If you count a mention loosely one week (any string match, including a competitor whose name contains yours) and strictly the next, your "trend" is measuring your own inconsistency. If one engine's answers include citations and another's don't, and you fold both into a single citation rate, the blended number moves whenever the engine mix moves, not when your visibility does.

Consistency has to hold along three axes at once:

1. **Across brands:** you and every competitor are scored by the identical rule (mention = alias match; citation = primary-domain match), so share of voice is a fair comparison.
2. **Across engines:** the same rule applies to every engine, and per-engine rates are reported separately so a citation-heavy engine doesn't distort a citation-light one.
3. **Across time:** the definition doesn't drift, so a trend line reflects reality changing, not your methodology changing.

This is exactly why AI visibility tooling matters more than manual spot-checks. Kitbase applies one definition everywhere: an answer is *present* for a brand when it names any of the brand's aliases or cites its primary domain, scored identically for your brand and every competitor, on every engine, on every run. Because matching happens in plain code against your stored answers, **editing a brand's aliases re-scores your entire history retroactively**, so fixing a missing alias doesn't just help going forward, it corrects every past run under the new, consistent definition, with no re-querying and no extra AI cost.

One consistency detail worth knowing: metrics that were added after a run was recorded (sentiment, recommended, rank) report as *not analyzed* — null — for older runs, never as zero. That distinction keeps a mixed window of old and new runs from silently understating your rates by treating "we didn't measure this yet" as "this was zero."

## Putting it together.

If you track only one number, track **presence rate per engine, over time** it's your AI-visibility ranking. But the moment you want to *act*, split it: mention rate and citation rate tell you whether your problem is entity strength or retrievable content, and the framing metrics tell you whether appearing is even helping. Just make sure whoever computes these numbers computes them the same way every time, for every brand, on every engine. A metric that shifts definition mid-stream isn't a metric it's noise with a label.

## FAQ

**What's the difference between mention rate and citation rate?**
Mention rate counts answers that *name* your brand in the text; citation rate counts answers that *link* your domain as a source. An engine can do either without the other, talk about you without linking, or cite your page while recommending a competitor. Presence rate is the union of both.

**What is a good presence rate?**
There's no universal benchmark, because it depends entirely on your category, the prompts you track, and how established your brand is. The useful comparison isn't against an absolute number, it's against your own trend over time and against your competitors' share of voice in the same prompts.

**Why do my metrics differ by AI engine?**
Because Perplexity, Gemini, Claude, and ChatGPT retrieve and synthesize answers through different systems. A citation-heavy engine like Perplexity will tend to show higher citation rates than one that leans on model knowledge. Always read per-engine rates rather than a single blended number.

**Can I be cited without being mentioned?**
Yes, and it's common. The engine links your domain as a source by pulling a definition or statistic from your page, while the answer's prose recommends someone else. It means your content is quotable but your positioning isn't winning the recommendation.

**Why does definitional consistency matter so much?**
Because AI visibility is measured as rates trended over time and compared across brands. If the definition of "present" drifts between brands, engines, or weeks, your comparisons and trends measure your methodology instead of reality. One rule, applied everywhere, is what makes the numbers mean anything.

---

*Want presence, mention, and citation rates tracked per engine and per competitor automatically? [Start your free trial](https://app.kitbase.dev/signup/) — 7 days, no credit card required — and see your AI visibility metrics in minutes.*