AI Visibility Metrics: Mentions, Citations, Impressions, and Share of Voice

Quick Answer

AI visibility metrics are only useful when each metric answers a different operating question. Mentions show whether a brand appears in an AI answer. Citations show which pages or domains the AI system exposes as sources. Impressions estimate the demand behind prompts where the brand appears. Share of Voice compares the brand's answer-space presence against competitors. Rank, sentiment, and raw response review show whether the answer is prominent, positive, and accurate.

The mistake is blending all of these into one number too early. A brand can be mentioned without being cited. A page can be cited without the answer recommending the brand. A Share of Voice chart can move because the competitor set changed. A high-impression prompt can be low intent. AI visibility teams need the metrics together, but they should diagnose them separately.

For ReachLLM teams, the workflow is: track a fixed prompt set, review mentions and Share of Voice, inspect citations and found sources, read the raw answers for sentiment and accuracy, then ship focused fixes through content, schema, llms.txt, page rewrites, GEO audits, PR outreach, and follow-up measurement.

Why AI Visibility Metrics Need Clear Definitions

Ahrefs' AI Visibility Metrics documentation is a useful market signal because it separates mentions, citations, impressions, and AI Share of Voice. That distinction matters. These metrics often appear in the same dashboard, but they do not measure the same thing.

Use this framing before reporting AI visibility to leadership:

Metric	The plain-English question
Mentions	Did the AI answer name the brand?
Citations	Which pages or domains did the answer cite as sources?
Found in	Which pages were retrieved or used in the background, even if not cited?
Impressions	How much estimated search demand sits behind prompts where the brand appears?
AI Share of Voice	What share of the answer opportunity belongs to us versus competitors?
Average Rank	When we appear, how early are we mentioned?
Sentiment	Does the answer describe us positively, neutrally, or negatively?
Raw response accuracy	Is the model saying the right thing?

If the team cannot explain which question a metric answers, the metric will create confusion. The goal is not to collect every possible number. The goal is to know what needs to change next.

Mentions: The Visibility Starting Line

Ahrefs defines a mention as a brand appearing at least once in an AI-generated response. If the same answer says the brand three times, that still counts as one mention for that brand in that response.

ReachLLM's Visibility Score follows the same practical idea at the prompt-run level: for each platform, it measures the percentage of tracked prompts where the platform's answer mentions the brand. The overall score averages across platforms that ran at least one prompt.

Mentions matter because buyers increasingly ask AI systems for recommendations before they visit websites. If the answer names three competitors and skips your brand, the buyer may never know to search for you.

But mentions are only the starting line.

Mention pattern	What to inspect next
Brand never appears	Prompt coverage, entity clarity, category language, and source presence.
Brand appears only in branded prompts	Unbranded discovery visibility is weak.
Brand appears after competitors	Average Rank and competitor source evidence.
Brand appears with vague wording	Product positioning and source accuracy.
Brand appears but answer is negative	Sentiment drivers and third-party proof.

Do not stop at "we were mentioned." Read the answer. A weak or inaccurate mention can be worse than absence because it shapes the buyer's first impression.

Citations: Evidence, Not Just Visibility

Ahrefs separates cited sources from pages found in the background. Its documentation notes that AI systems may retrieve pages while generating an answer but cite only some of them. It also counts one citation for a domain when at least one page from that domain is cited, even if several pages from the same domain appear.

This distinction is critical for GEO work.

Signal	What it usually means
Mentioned and own domain cited	The answer sees the brand and exposes owned evidence.
Mentioned but own domain not cited	The brand may be known, but evidence is coming from elsewhere.
Own domain cited but brand not recommended	The page may explain the topic without making the product role clear.
Found but not cited	The page may be retrievable but not strong enough to expose as evidence.
Competitor cited repeatedly	Their owned or earned sources may be shaping the answer.

ReachLLM's Sources view is built around this operational split. It shows which domains AI platforms cite, exact URL-level source inventories, content types, source detail, and whether the user's own domain was cited.

That matters because the fix depends on the source pattern. If your page is found but not cited, improve structure, extractability, schema, headings, and answer-first sections. If a third-party source is cited for competitors, the next action may be PR outreach, directory correction, partner content, or review coverage rather than another blog post.

Found In: The Hidden Source Queue

"Found in" is useful because it shows pages that may have influenced an answer without becoming visible citations. In practice, this is a middle signal: stronger than no retrieval, weaker than an exposed citation.

Use it carefully:

Found-in pattern	Likely action
Your page is found but not cited	Tighten the answer section, add clearer claims, improve schema, and link to supporting pages.
Competitor page is found and cited	Compare the page format, proof, freshness, and specificity.
Third-party page is found but not cited	Track whether it becomes a citation over time before overreacting.
Many weak pages are found	Consolidate or improve the strongest page instead of publishing more thin content.

Found-in data should not become a vanity metric. It is a diagnostic queue. The question is whether the page can become a better answer source.

Impressions: Demand, Not Proof Of Influence

Ahrefs calculates impressions by summing search volumes, based on Google, for prompts where the brand appears as an AI answer. The documentation describes using the highest-volume keyword whose SERP results show the prompt as a People Also Ask question.

This is useful because not all prompts have the same demand. A brand appearing in a high-demand prompt has a different opportunity than a brand appearing in a rare internal phrase.

But impression-style metrics need caution:

Risk	How to handle it
Search volume is not AI usage	Treat impressions as demand context, not exact AI audience size.
High-volume prompts may be low intent	Segment by funnel stage and buyer relevance.
One broad prompt can dominate the chart	Review prompt-level contribution before celebrating.
Branded prompts can inflate comfort	Separate branded and unbranded visibility.
Search-backed prompt sources can miss sales-language prompts	Add custom prompts from sales calls, support questions, and customer language.

ReachLLM teams should pair demand context with prompt intent. A lower-volume prompt like "which AI visibility tool tracks citations and sentiment" can be more commercially useful than a broad educational prompt if it maps to a buyer decision.

AI Share Of Voice: Competitive Context

Ahrefs defines AI Share of Voice as a brand's percentage share of impressions compared with other tracked brands, and notes that it can change a lot depending on how entities and competitors are configured.

ReachLLM's Share of Voice is presence-based across analyzed answers: it counts brand appearances in the tracked comparison set and reports the percentage that belongs to the user's brand. Other Companies stay visible for review but do not affect the metric until promoted.

The exact formula matters less than the reporting discipline: Share of Voice is comparative. It is only meaningful when the tracked entity and competitor set are clean.

Before reporting Share of Voice, check:

Are brand aliases grouped correctly?
Are direct, indirect, SERP, and "other company" competitors classified correctly?
Are misspellings or unrelated entities removed?
Are branded and unbranded prompts separated?
Did the prompt set change during the reporting period?
Is one high-demand prompt skewing the result?

If the answer to any of these is unclear, report Share of Voice as directional until the entity setup is fixed.

Rank And Sentiment: Quality Of The Mention

Mentions and Share of Voice show whether the brand appears. Rank and sentiment show how useful that appearance is.

ReachLLM's Average Rank records where the brand appears among tracked competitors when it is mentioned. A brand listed first in a three-vendor answer has a different buyer impact than a brand buried after caveats.

Sentiment classifies the tone of brand mentions as positive, neutral, or negative. This matters because visibility can spread stale or unfavorable claims.

Use this diagnostic table:

Pattern	Interpretation
High visibility, strong rank, positive sentiment	The brand is becoming a strong answer candidate.
High visibility, weak rank	AI systems know the brand but prefer competitors.
High visibility, neutral sentiment	The brand may need clearer proof, positioning, or differentiation.
High visibility, negative sentiment	Fix source material before chasing more mentions.
Low visibility, strong citations	The owned content may be useful but not clearly tied to the brand.

This is why raw response review is still required. Sentiment labels help triage, but a human should read the answer before deciding what to ship.

The Metric Pairings That Actually Diagnose Work

No single AI visibility metric tells the whole story. Pair metrics to find the likely next action.

Metric pairing	What it diagnoses
Mentions + Share of Voice	Whether the brand appears and whether competitors own the same answer space.
Mentions + Average Rank	Whether the brand is present but secondary.
Mentions + Sentiment	Whether visibility is helpful, neutral, or harmful.
Mentions + citations	Whether the brand appears with owned or third-party evidence.
Citations + source type	Whether the next fix is owned content, docs, PR, directory, community, or review coverage.
Impressions + funnel stage	Whether a visible prompt is commercially important.
Raw answers + shipped fixes	Whether the work changed the actual recommendation.

For example, a low citation rate with decent mentions suggests a source-trust problem. A low Share of Voice with strong own-domain citations may point to competitor authority or missing comparison content. Strong impressions with neutral sentiment may mean the team has attention but not enough proof.

A Practical Metric Review Workflow

Run this weekly if the team is actively shipping GEO work:

Review Visibility Score, Share of Voice, Average Rank, sentiment, citation rate, and raw answer changes.
Split branded prompts from unbranded discovery prompts.
Sort by high-intent prompts where competitors appear and your brand does not.
Open the raw responses behind the biggest changes.
Inspect cited sources and found-in pages.
Tag each issue as missing mention, weak rank, weak owned citation, third-party source gap, sentiment issue, or factual inaccuracy.
Pick one owned-page fix and one source action.
Ship the fix through content, schema, llms.txt, page rewrite, docs update, PR outreach, or brand fact correction.
Re-run affected prompts or wait for the next scheduled run.
Record whether the answer, citation, rank, or sentiment changed.

This keeps the team from staring at dashboards. Metrics should become a work queue.

Where ReachLLM Fits

ReachLLM is built for teams that need AI visibility metrics connected to execution. It runs tracked prompts across enabled AI platforms, analyzes brand and competitor mentions, calculates Visibility Score, Share of Voice, Average Rank, sentiment, citation rate, and source data, and keeps raw responses available for review.

The platform then connects those findings to the work that can move the metrics: GEO audits, content generation, page rewrites, structured data, llms.txt, PR outreach, integrations, and the ReachLLM agent.

That distinction matters. A metrics dashboard can show that citations are weak. An operating workflow should show which prompt, which answer, which source, which page, and which fix should happen next.

FAQ

What are AI visibility metrics?

AI visibility metrics measure whether AI systems mention, cite, rank, and accurately describe a brand across tracked prompts. Common metrics include mentions, citations, found-in pages, impressions, Share of Voice, Average Rank, sentiment, and raw answer accuracy.

What is the difference between AI mentions and citations?

A mention means the AI answer names the brand. A citation means the answer exposes a page or domain as a source. A brand can be mentioned without its own website being cited, so teams should measure both separately.

Is AI Share of Voice the same in every tool?

No. Tools can calculate Share of Voice differently. Some weight by impressions, while others count brand appearances across tracked answers. Always check the formula, competitor set, aliases, and prompt set before comparing numbers.

Should teams optimize for impressions or high-intent prompts?

Use impressions as demand context, but prioritize high-intent prompts that map to real buyer decisions. A lower-volume vendor-evaluation prompt may be more valuable than a broad high-volume educational prompt.

How does ReachLLM help with AI visibility metrics?

ReachLLM tracks prompts across AI platforms, measures Visibility Score, Share of Voice, Average Rank, sentiment, citation rate, cited sources, and raw responses, then helps teams turn metric gaps into GEO audits, content updates, schema, llms.txt, PR outreach, integrations, and agent workflows.