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

Being cited by an AI search engine is broadly seen as an authority sign — proof that your content material is credible sufficient for the AI to supply from.

However being cited isn’t the identical as being talked about. And in additional than half of circumstances, a quotation doesn’t immediately result in a model point out. 

Kevin Indig coined the time period ghost quotation to explain this hole. When an AI cites your web site with out explicitly mentioning you, readers stroll away with out recognizing your model or trusting you as the unique supply.

To analyze this, we partnered with Kevin and analyzed hundreds of area appearances throughout a number of AI engines like google.

Key takeaways

• 62% of AI citations are ghost citations. Your web site will get a supply hyperlink, however the AI by no means says your identify within the reply.

• Every AI engine behaves in another way. When a model seems in a Gemini reply, it is named within the textual content 83.7% of the time however cited as a supply solely 21.4% of the time. ChatGPT does the reverse: 87% quotation price, 20.7% point out price.

• Point out charges differ by nation. Manufacturers get talked about in 50% of AI solutions in India and Sweden, however they solely get talked about in 18%-22% of AI solutions in Italy, Brazil, and the Netherlands.

• Quick, conversational queries produce 30x-50x extra model mentions. Lengthy prompts set off extra citations, however they result in fewer model mentions.

• Question intent and content material sort strongly have an effect on model point out charges. Informational content material earns citations however not often model mentions. Comparative content material produces 2.4x extra model mentions.

Methodology

We ran this examine in collaboration with Kevin Indig and Growth Memo utilizing information from the Semrush AI Visibility Toolkit.

We logged 3,981 area appearances for 115 prompts run throughout 14 international locations and 4 main AI engines like google:

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

For each area look, we famous two outcomes:

• Cited: The area appeared as a supply hyperlink within the response

• Talked about: The model identify appeared within the AI reply

We additionally calculated the point out price throughout completely different question varieties.

Let’s discover the findings in higher depth.

1. 62% of citations don’t result in model mentions in AI solutions

You may assume being cited means your model is seen in AI solutions. The information says in any other case.

Each look in our dataset fell into certainly one of three buckets:

• Nearly 62% (61.7%) had been ghost citations. AI platforms used the web page as a supply hyperlink, however the model identify by no means appeared within the precise reply.
• Over 13% (13.2%) had been each cited and talked about. The supply hyperlink plus model identify appeared within the reply.
• Solely about 25% (25.1%) had been model mentions with no quotation. The AI named the model within the reply with out linking to a supply web page.

Meaning 74.9% of all model appearances included a quotation, however solely 38.3% of appearances included a model point out. So, the quotation price is almost double the point out price.

Give it some thought this manner: Showing as a supply in AI search doesn’t routinely make your model seen to customers. A quotation gives attribution, whereas your model stays absent from the reply itself.

2. Each AI engine behaves in another way

The 4 AI engines we analyzed deal with citations and mentions in essentially alternative ways:

• Gemini mentions manufacturers in 83.7% of appearances, nevertheless it solely generates a quotation hyperlink 21.4% of the time. An “look” right here means a site confirmed up in an AI reply in any respect, whether or not as a supply hyperlink, a model point out within the reply textual content, or each. So, Gemini acts like a conversationalist, drawing on what it already is aware of.
• ChatGPT does the other of Gemini. It cites manufacturers 87% of the time however mentions manufacturers in solely 20.7% of solutions. These solutions look extra like tutorial papers with footnotes.
• Google AI Overviews sit within the center, however they lean towards citations.
• Google AI Mode mentions manufacturers at practically twice the speed of ChatGPT,however AI Mode nonetheless acts nearer to a footnoted analysis piece than to Gemini’s knowledge-based solutions.

This implies you may’t assume visibility in a single AI engine will translate to visibility in one other. In our dataset, there was nearly no overlap between the manufacturers ChatGPT cited and the manufacturers Gemini named for a similar prompts. Completely different engines reward completely different alerts, codecs, and sources.

This implies you may’t assume visibility in a single AI engine will translate to visibility in one other. Completely different engines reward completely different alerts, codecs, and sources.

“There’s nearly no overlap between which manufacturers ChatGPT cites and which of them Gemini names for a similar immediate. These are completely different behavioral techniques. Deal with them that method.”

Kevin Indig, Development Advisor

3. Robust manufacturers get named, whereas aggregators get cited

Among the many domains that appeared most frequently in our information, aggregator and tutorial websites like medium.com, wikipedia.org, wired.com, and harvard.edu tended to be cited however not named. 

Manufacturers with robust public identities had been displaying up the other method: They had been named within the reply, usually with no quotation.

For instance, Google was named in AI solutions practically 3 times extra usually than it appeared as a supply hyperlink. Apple’s model identify appeared in solutions practically twice as usually because it was cited. 

On the different finish of the spectrum, medium.com was cited 16 instances in our dataset and by no means named in any AI reply.

The chart values are normalized to imagine 100 citations. A worth of 100 means a model was talked about and cited equally. Google’s 278 means Google was named 2.78 instances for each 100 citations of a Google URL. Medium’s zero means medium.com was cited however by no means named within the reply textual content.

As Kevin places it: 

“The AI is aware of the details about the model got here from someplace, however doesn’t really feel the necessity to explicitly say so to customers. The model identify carries by itself.”

Kevin Indig, Development Advisor

In the event you’re a writer or research-style web site, count on extra citations than mentions and observe citations as your main AI visibility metric. 

In the event you’re a client model, intention for the reverse: prioritize monitoring unlinked model mentions on third-party web sites (Reddit threads, information protection, listicles) alongside your backlinks. Each feed AI engines’ familiarity together with your model.

AI engines point out manufacturers they’ve already seen constantly throughout the online. PR, neighborhood discussions, and clear model language all reinforce that familiarity.

4. AI engines disagree on the identical model 22% of the time

When operating this evaluation, we examined 454 prompt-and-domain combos throughout a number of engines. Every mixture is one particular immediate paired with one particular area. For instance, the immediate “finest CRM for small enterprise” paired with hubspot.com, examined throughout every AI engine.

In 100 of the 454 (22%), the engines disagreed on whether or not to call the model.

For instance: 

• Instagram.com was talked about by ChatGPT and Gemini. Google AI Overviews cited it with out naming it.

• Fb.com was named in three out of three Gemini appearances. Google AI cited Fb 9 out of 9 instances and named it as soon as.

For correct measurement, examine every engine individually on your high branded and non-branded subjects. Aggregated scores may easy over the circumstances the place you’re seen in a single engine and invisible in one other.

5. AI model point out charges differ by nation

Model visibility patterns in our dataset differed sharply by market. The mixture of citations and model mentions diversified particularly.

For instance:

• India and Sweden: 50% of AI solutions to queries from these markets included a model identify within the response. That is the best within the examine, and question patterns there leaned extra conversational and brand-forward.

• Italy, Brazil, and the Netherlands: Solely 18%-22% of AI solutions included a model identify, regardless that AI engines cited sources from these markets at very excessive charges (82%-94%). 

• The U.Okay. and Canada (41% and 44%) sit within the upper-middle of the vary. Manufacturers are talked about in slightly below half of AI solutions from these markets.

In the event you function throughout a number of markets, count on AI visibility patterns to look completely different in every. Set country-specific objectives for citations and mentions slightly than making use of a single playbook globally.

Word: We localized and verified all prompts for every market, so the variations in point out charges replicate real regional patterns.

6. Quick, conversational queries produce 30x-50x extra model mentions

Question phrasing adjustments model point out outcomes dramatically. Throughout the prompts in our dataset, the size ranged from 16 to 98 characters, averaging about 60.

Quick, conversational queries have model point out charges of practically 100%, whereas lengthy, structured prompts have point out charges of simply 2%-3%. That’s a 30x-50x distinction on the identical matter.

A brief question like “Ought to I lease or purchase a automobile for my enterprise?” results in model mentions practically each time. A protracted model of the identical query with additional context and framing (e.g., ), produces nearly no model mentions regardless that it triggers extra citations.

“Manufacturers must map not simply which subjects they need to seem in, however which phrasing patterns produce mentions versus ghost citations. Quick, conversational queries and lengthy, structured queries behave like completely different merchandise.”

Kevin Indig, Development Advisor

This implies model visibility in AI search is not only topic-dependent. It’s also extremely delicate to how customers ask the query.

7. Informational queries set off essentially the most ghost citations

Content material sort correlated strongly with model point out charges in our information.

Informational content material earned excessive quotation charges however low point out charges, whereas comparative content material earned mentions extra usually:

• Informational queries (“what’s”, “clarify”, “how does”): 89.3% quotation price, 18% point out price
• Comparative queries (“finest”, “vs”, “suggest”): 43.3% point out price, 2.4x extra model mentions than informational
• How-to queries: 42.8% point out price
• Industrial queries (pricing, shopping for intent): 35.6% point out price, 84.4% quotation price

The sample makes loads of sense: Informational content material will get consumed as uncooked reference information. Comparative content material forces the AI to call the gamers it’s evaluating.

How you can shut the quotation hole

Citations and mentions are separate outcomes. 

Citations replicate area authority and require unique content material and matter authority. Whereas mentions replicate model belief and are available from clear positioning, optimistic sentiment, area of interest authority, and thorough product content material.

To shut the hole, develop methods to get each cited and talked about, analyze citations throughout LLMs and completely different classes, and determine low-hanging alternatives.

In Semrush’s AI Visibility Toolkit, you may observe your AI search visibility throughout key areas:

• By AI engine. Quotation and point out charges differed sharply by LLM in our information. Analyze patterns and determine alternatives for every AI platform.

• By nation. Point out charges ranged from 18% to 50% throughout markets. Set country-specific objectives primarily based on your corporation priorities.

• By intent. Informational, comparative, how-to, and business queries produced completely different point out charges. See which matter areas you carry out effectively in and the place you’re lacking visibility in AI search.

You can even uncover alternatives to extend model visibility from content material that AI platforms already cite.

Open the Visibility Overview report and scroll to “Your Performing Matters.” Right here, you may determine subjects and prompts the place your content material is cited, however your model isn’t talked about.

This implies the web page references your insights with out clearly naming your model or merchandise.

In some circumstances, including clearer model references, product mentions, or comparability sections that includes particular instruments and firms might help shut that hole.