One AVI Score Isn't Enough: Mentions, Sources, Citations

A single composite score like AVI (AI Visibility Index) isn't enough because brands show up in AI search in different ways. For the same prompt, your page may be picked up as an evidence source while your brand name never appears in the answer; or, the other way around, your brand may be mentioned but with weak evidence links. If you can't see this difference, your team will keep getting stuck on "We're visible, so why don't we feel it?"

Why isn't one composite score enough?

A summary score like AVI is still useful for reporting current visibility to executives in one line. Day-to-day operations are different, though. Look only at one score and the three situations below get lumped together.

1. Your brand name appears directly in the answer
2. You're picked up in the answer's source or citation list, but your brand name doesn't show
3. You're not picked up in sources or citations at all

These three call for completely different responses. Case 1 means brand mentions are happening, so you can grow the comparison and recommendation context further. Case 2 means you're being used only as anonymous evidence for AI, so you need to rework content format and brand signals. Case 3 means you need to revisit accessibility, structure, topical fit, and external signals themselves.

In real reports, it's common for these three situations to be mixed behind the same AVI score. Say AVI comes out at 65, exactly the same as last month. If a few prompts that were in case 2 moved to case 1, and a few that were in case 1 slid to case 3, the score looks unchanged on the surface, but the makeup of your content's performance has actually shifted a lot. Teams that only watch the score miss this change entirely.

The three metrics each answer a different question

Mention rate, source rate, and citation rate each answer a different question.

Don't treat source rate and citation rate as the same thing either. Source rate looks at whether your domain made it into the candidate pool used to build the answer, and citation rate looks at whether, among those candidates, it actually surfaced as a linked piece of evidence. In other words, a source is closer to entering the candidate pool, and a citation is closer to being adopted as evidence.

Put all three metrics side by side and you can see at which stage your content gets stuck. If source rate is high but citation rate is low, AI often considers your page as a candidate but finds something lacking when it comes to using it as actual evidence. The information on the page is likely scattered, or there's no paragraph that gives a clear answer. Conversely, if citation rate is fine but mention rate is low, treat it as a brand signal problem rather than a content quality problem.

What exactly is a ghost citation?

This is the core concept Kevin Indig raised in Growth Memo. A ghost citation is when AI uses your content as evidence but doesn't say your brand name in the answer text. The link is there and you're picked up as evidence, yet the user never sees your brand name.

The study, run jointly with the Semrush AI Visibility Toolkit, analyzed 3,981 domains, 115 prompts, 14 countries, and 4 AI engines. It found that 74.9% were cited, but only 38.3% had the brand name mentioned along with the citation, and 61.7% of the citations were ghost citations^[1]. Being used as evidence by AI doesn't mean your brand stays in the answer.

What this means for practitioners is clear. Don't relax just because citation rate says "our content is being used well in AI answers." More than half of those citations may be cases where only the information gets carried off without your brand name. From the user's side, they got useful information but won't remember which brand it came from. Your content is doing the work, but your brand isn't getting the credit.

Also track the conversion rate from citation to mention

This metric reads ghost citations in reverse. The definition is simple.

> citation→mention conversion rate = the share of cited prompts where the brand name also appears in the answer text

As formulas:

• citation→mention conversion rate = (number of prompts with both citation and mention) / number of cited prompts
• ghost citation rate = (number of prompts cited but not mentioned) / number of cited prompts

You need both because they're reported to different audiences. Product and operations teams need to know where things leak, so the ghost citation rate is useful to them. Sales and executives want to know how much citation actually turns into brand mentions, so the conversion rate may be more intuitive.

Plotting both metrics together over time is also a good way to check whether content improvements worked. For example, if you added brand entity sentences to comparison pages, you can verify whether that work actually had an effect by checking in the following week's report whether the ghost citation rate fell and the conversion rate rose. If both metrics stay flat, the problem lies somewhere other than brand signals, and you need to look for the cause again.

Splitting into four states makes interpretation easier

In practice, sorting every prompt into the four states below makes interpretation easiest.

Even the same 10-point increase can be a completely different situation once you look inside. If AVI rose and source rate rose but mention rate stayed flat, it may look good at a glance, but you may actually just have gone further into AI's anonymous reference library. Conversely, if citation rate dipped a bit but mention rate and conversion rose, that may be the better change from a brand mention standpoint.

Color-coding these four states by prompt makes conversations within the team much easier. For example, if half of 20 core prompts are No show, this quarter's goal can be concrete, like "move half of the No show prompts to Mention only or Source/Citation only," rather than "raise AVI by some number of points." Conversely, if most are already clustered in Citation + Mention, the next goal should shift to keeping that state while widening the set of question clusters you cover.

How do these metrics change your content strategy?

Strong informational guides can increase citations. Definition posts, concept explainers, and FAQ roundups are easy for AI to use as answer evidence, so they're good at generating citations and sources^[2].

Strong comparison and recommendation posts can increase mention conversion. Posts like A vs B, situational recommendations, and selection checklists are more likely to get AI to say your brand name directly in the answer text.

Weak brand entity signals can enlarge ghost citations. Even if your domain gets in as a source, if your brand identity doesn't stand out clearly within the text structure, AI may absorb the content as if it were anonymous information.

In the end, split your content portfolio by question cluster. Trying to win every question with one type of page makes operations unstable. It's better to separate roles: informational pages for citations, and comparison, recommendation, and adoption pages for mention conversion. This kind of structuring ultimately aligns with the basic principle of creating content that is useful and verifiable for users^[3].

When putting this principle into practice, you don't need to rebuild your content calendar from scratch. Adding a single column next to upcoming content, "Is this post's goal citations or mention conversion?", is enough. Without that distinction, a post's purpose easily gets blurry, with comparison phrases forced into informational posts or comparison posts padded with excessive definitions.

Here's how to fit this into a working routine

The point of AI visibility isn't to make the score look nice. It's to understand which evidence AI chooses and to align your assets and sources so your brand can be part of that choice. Using TRAIL Search to see which state (No show, Source only, Mention only, Citation+Mention) your prompt set is concentrated in right now can be the starting point.

For related reading, see Why it's risky to view AI visibility as a single score: ChatGPT and Perplexity are different, What is the AVI score? A guide to brand visibility metrics in the AI search era, and A framework for measuring AI search performance: GSC, GA4, and prompt tracking as weekly and monthly routines.