Why 62% of AI citations don’t lead to brand mentions [Study]

Being cited by an AI hunt motor is wide seen arsenic an authority awesome — impervious that your contented is reliable capable for the AI to root from.

But being cited is not the aforesaid arsenic being mentioned. And successful much than half of cases, a citation doesn’t instantly lead to a marque mention. 

Kevin Indig coined the word shade citation to picture this gap. When an AI cites your tract without explicitly mentioning you, readers locomotion distant without recognizing your marque aliases trusting you arsenic the original source.

To analyse this, we collaborated pinch Kevin and analyzed thousands of domain appearances crossed aggregate AI hunt engines.

Key takeaways

• 62% of AI citations are shade citations. Your tract gets a root link, but the AI ne'er says your sanction successful the answer.

• Each AI motor behaves differently. When a marque appears successful a Gemini answer, it's named successful the matter 83.7% of the clip but cited arsenic a root only 21.4% of the time. ChatGPT does the reverse: 87% citation rate, 20.7% mention rate.

• Mention rates alteration by country. Brands get mentioned successful 50% of AI answers successful India and Sweden, but they only get mentioned successful 18%-22% of AI answers successful Italy, Brazil, and the Netherlands.

• Short, conversational queries nutrient 30x-50x much marque mentions. Long prompts trigger much citations, but they lead to less marque mentions.

• Query intent and contented type powerfully impact marque mention rates. Informational contented earns citations but seldom marque mentions. Comparative contented produces 2.4x much marque mentions.

Methodology

We ran this study successful collaboration pinch Kevin Indig and Growth Memo utilizing information from the Semrush AI Visibility Toolkit.

We logged 3,981 domain appearances for 115 prompts tally crossed 14 countries and 4 starring AI hunt engines:

• ChatGPT
• Google AI Overviews
• Gemini
• Google AI Mode

For each domain appearance, we noted 2 outcomes:

• Cited: The domain appeared arsenic a root nexus successful the response

• Mentioned: The marque sanction appeared successful the AI answer

We besides calculated the mention complaint crossed different query types.

Let’s research the findings successful greater depth.

1. 62% of citations don’t lead to marque mentions successful AI answers

You mightiness presume being cited intends your marque is visible successful AI answers. The information says otherwise.

Every quality successful our dataset fell into 1 of 3 buckets:

• Almost 62% (61.7%) were shade citations. AI platforms utilized the page arsenic a root link, but the marque sanction ne'er appeared successful the existent answer.
• Over 13% (13.2%) were some cited and mentioned. The root nexus positive marque sanction appeared successful the answer.
• Only astir 25% (25.1%) were marque mentions without a citation. The AI named the marque successful the reply without linking to a root page.

That intends 74.9% of each marque appearances included a citation, but only 38.3% of appearances included a marque mention. So, the citation complaint is astir double the mention rate.

Think astir it this way: Appearing arsenic a root successful AI hunt doesn’t automatically make your marque visible to users. A citation offers attribution, while your marque stays absent from the reply itself.

2. Every AI motor behaves differently

The 4 AI engines we analyzed dainty citations and mentions successful fundamentally different ways:

• Gemini mentions brands successful 83.7% of appearances, but it only generates a citation nexus 21.4% of the time. An “appearance” present intends a domain showed up successful an AI reply astatine all, whether arsenic a root link, a marque mention successful the reply text, aliases both. So, Gemini acts for illustration a conversationalist, drafting connected what it already knows.
• ChatGPT does the other of Gemini. It cites brands 87% of the clip but mentions brands successful only 20.7% of answers. These answers look much for illustration world papers pinch footnotes.
• Google AI Overviews beryllium successful the middle, but they thin toward citations.
• Google AI Mode mentions brands astatine astir doubly the complaint of ChatGPT,but AI Mode still acts person to a footnoted investigation portion than to Gemini’s knowledge-based answers.

This intends you can’t presume visibility successful 1 AI motor will construe to visibility successful another. In our dataset, location was almost nary overlap betwixt the brands ChatGPT cited and the brands Gemini named for the aforesaid prompts. Different engines reward different signals, formats, and sources.

This intends you can’t presume visibility successful 1 AI motor will construe to visibility successful another. Different engines reward different signals, formats, and sources.

“There’s almost nary overlap betwixt which brands ChatGPT cites and which ones Gemini names for the aforesaid prompt. These are different behavioral systems. Treat them that way.”

Kevin Indig, Growth Advisor

3. Strong brands get named, while aggregators get cited

Among the domains that appeared astir often successful our data, aggregator and world sites for illustration medium.com, wikipedia.org, wired.com, and harvard.edu tended to beryllium cited but not named. 

Brands pinch beardown nationalist identities were showing up the other way: They were named successful the answer, often without a citation.

For example, Google was named successful AI answers astir 3 times much often than it appeared arsenic a root link. Apple's marque sanction appeared successful answers astir doubly arsenic often arsenic it was cited. 

At the different extremity of the spectrum, medium.com was cited 16 times successful our dataset and ne'er named successful immoderate AI answer.

The floor plan values are normalized to presume 100 citations. A worth of 100 intends a marque was mentioned and cited equally. Google’s 278 intends Google was named 2.78 times for each 100 citations of a Google URL. Medium’s 0 intends medium.com was cited but ne'er named successful the reply text.

As Kevin puts it: 

“The AI knows the accusation astir the marque came from somewhere, but doesn’t consciousness the request to explicitly opportunity truthful to users. The marque sanction carries connected its own.”

Kevin Indig, Growth Advisor

If you’re a patient aliases research-style site, expect much citations than mentions and way citations arsenic your superior AI visibility metric. 

If you’re a user brand, purpose for the reverse: prioritize search unlinked marque mentions connected third-party websites (Reddit threads, news coverage, listicles) alongside your backlinks. Both provender AI engines’ familiarity pinch your brand.

AI engines mention brands they’ve already seen consistently crossed the web. PR, organization discussions, and clear marque connection each reenforce that familiarity.

4. AI engines disagree connected the aforesaid marque 22% of the time

When moving this analysis, we tested 454 prompt-and-domain combinations crossed aggregate engines. Each operation is 1 circumstantial punctual paired pinch 1 circumstantial domain. For example, the punctual “best CRM for mini business” paired pinch hubspot.com, tested crossed each AI engine.

In 100 of the 454 (22%), the engines disagreed connected whether to sanction the brand.

For example: 

• Instagram.com was mentioned by ChatGPT and Gemini. Google AI Overviews cited it without naming it.

• Facebook.com was named successful 3 retired of 3 Gemini appearances. Google AI cited Facebook 9 retired of 9 times and named it once.

For meticulous measurement, cheque each motor separately for your apical branded and non-branded topics. Aggregated scores mightiness soft complete the cases wherever you’re visible successful 1 motor and invisible successful another.

5. AI marque mention rates alteration by country

Brand visibility patterns successful our dataset differed sharply by market. The operation of citations and marque mentions varied especially.

For example:

• India and Sweden: 50% of AI answers to queries from these markets included a marque sanction successful the response. That's the highest successful the study, and query patterns location leaned much conversational and brand-forward.

• Italy, Brazil, and the Netherlands: Only 18%-22% of AI answers included a marque name, moreover though AI engines cited sources from these markets astatine very precocious rates (82%-94%). 

• The U.K. and Canada (41% and 44%) beryllium successful the upper-middle of the range. Brands are mentioned successful conscionable nether half of AI answers from these markets.

If you run crossed aggregate markets, expect AI visibility patterns to look different successful each. Set country-specific goals for citations and mentions alternatively than applying a azygous playbook globally.

Note: We localized and verified each prompts for each market, truthful the differences successful mention rates bespeak genuine location patterns.

6. Short, conversational queries nutrient 30x-50x much marque mentions

Query phrasing changes marque mention outcomes dramatically. Across the prompts successful our dataset, the magnitude ranged from 16 to 98 characters, averaging astir 60.

Short, conversational queries person marque mention rates of astir 100%, while long, system prompts person mention rates of conscionable 2%-3%. That’s a 30x-50x quality connected the aforesaid topic.

A short query for illustration “Should I lease aliases bargain a car for my business?” leads to marque mentions astir each time. A agelong type of the aforesaid mobility pinch other discourse and framing (e.g., ), produces almost nary marque mentions moreover though it triggers much citations.

“Brands request to representation not conscionable which topics they want to look in, but which phrasing patterns nutrient mentions versus shade citations. Short, conversational queries and long, system queries behave for illustration different products.”

Kevin Indig, Growth Advisor

This intends marque visibility successful AI hunt is not conscionable topic-dependent. It is besides highly delicate to really users inquire the question.

7. Informational queries trigger the astir shade citations

Content type correlated powerfully pinch marque mention rates successful our data.

Informational contented earned precocious citation rates but debased mention rates, while comparative contented earned mentions much often:

• Informational queries (“what is”, “explain”, “how does”): 89.3% citation rate, 18% mention rate
• Comparative queries (“best”, “vs”, “recommend”): 43.3% mention rate, 2.4x much marque mentions than informational
• How-to queries: 42.8% mention rate
• Commercial queries (pricing, buying intent): 35.6% mention rate, 84.4% citation rate

The shape makes a batch of sense: Informational contented gets consumed arsenic earthy reference data. Comparative contented forces the AI to sanction the players it’s evaluating.

How to adjacent the citation gap

Citations and mentions are abstracted outcomes. 

Citations bespeak domain authority and require original contented and taxable authority. While mentions bespeak marque spot and travel from clear positioning, affirmative sentiment, niche authority, and thorough merchandise content.

To adjacent the gap, create strategies to get some cited and mentioned, analyse citations crossed LLMs and different categories, and place low-hanging opportunities.

In Semrush’s AI Visibility Toolkit, you tin way your AI hunt visibility crossed cardinal areas:

• By AI engine. Citation and mention rates differed sharply by LLM successful our data. Analyze patterns and place opportunities for each AI platform.

• By country. Mention rates ranged from 18% to 50% crossed markets. Set country-specific goals based connected your business priorities.

• By intent. Informational, comparative, how-to, and commercialized queries produced different mention rates. See which taxable areas you execute good successful and wherever you’re missing visibility successful AI search.

You tin besides uncover opportunities to summation marque visibility from contented that AI platforms already cite.

Open the Visibility Overview study and scroll to “Your Performing Topics.” Here, you tin place topics and prompts wherever your contented is cited, but your marque isn’t mentioned.

This intends the page references your insights without intelligibly naming your marque aliases products.

In immoderate cases, adding clearer marque references, merchandise mentions, aliases comparison sections featuring circumstantial devices and companies tin thief adjacent that gap.