A cited domain is not always a named brand, so measure both

Ask ChatGPT a buyer question in your category. If you are not in the shortlist, that is the gap the report shows.

A cited domain is not always a named brand, so measure both

Evaluating AI citations vs brand mentions helps explain how a company appears in generative search. An answer engine may use a website as a source without naming the company. A citation shows where information came from; a brand mention means that the company, product, or service appears in the generated answer.

These outcomes are connected but not interchangeable. Tracking both can show whether a company is being used as evidence, presented as an option, or discussed without a link. Separating them is part of build an evidence map for AI search.

Defining ghost citations in generative engines

A generative system can link to a source page while summarizing its information without naming the company behind it. This is a “ghost citation”: the domain is visible, but the associated brand is not.

A Semrush study conducted with Kevin Indig logged 3,981 domain appearances from 115 prompts across four AI systems and 14 countries. It classified 61.7% of those appearances as ghost citations. In the study’s ChatGPT results, domains were cited in 87% of appearances, while brands were named in only 20.7%.

A brand mention may be unlinked. An AI citation, by contrast, links to a specific source page. A response can contain either outcome, both, or neither. For each tested answer, record:

whether the brand was named;
whether the company’s domain was cited;
whether the cited page was owned by the company or hosted by a third party;
whether the mention was favorable, neutral, or unfavorable; and
whether the answer recommended the brand or used its content only as evidence.

This separation prevents a high citation count from being mistaken for strong brand recognition.

The source of brand mentions in AI answers

Owned content is only one source of generative-search visibility. An eMarketer study reported that 85% of brand mentions in AI answers originated from third-party pages rather than owned domains. Review platforms, forums, editorial comparisons, and other external sources can therefore influence how an answer describes a company.

That finding should not be merged with research about the category of cited sources. Separate reported findings said that brands represented 52.5% of citations and 59.8% of Google AI Overview citations. These figures address a different question from eMarketer’s: they describe the type of cited source, while the 85% figure describes where a brand mention originated. Neither establishes how often a brand was named, whether it was recommended, or which source influenced the decision. The figures are not one shared benchmark.

Evidence that AI answers affect commercial research also comes from different methods. In an April 2026 G2 survey of 1,076 B2B software buyers, 51% said they started research in an AI chatbot more often than on Google, and 71% used AI chatbots at some point during research. The survey found that 69% changed their intended vendor choice because of AI chatbot recommendations, while 33% bought from a vendor they had not previously known.

Third-party validation was important in the same research. Forty-five percent of respondents called citations from software review sites the single most confidence-building element in an AI response, and 83% said they felt more confident in their purchase after using AI chatbots.

These results do not show which external source type is most influential for every category. Review profiles, analyst coverage, customer stories, partner directories, community discussions, and editorial comparisons may serve different roles. Monitor the sources that appear in the company’s own prompt set rather than assuming that one channel consistently drives recommendations.

How to measure AI citations versus brand mentions

There is no settled measurement standard for generative-engine visibility. A 2026 review described GEO terminology, metrics, and evidence standards as heterogeneous. Separate measurement research argues that visibility is a distribution across repeated prompt, platform, and time observations rather than a one-time ranking.

Results can change between runs. A brand may be named in one response, cited without a mention in another, and absent from a third. A useful monitoring process should capture that variation.

For manual monitoring, Semrush recommends selecting 5 to 10 question-based category prompts and rerunning them weekly across major AI platforms. Keep the prompts consistent and record the platform, model or search mode, location, date, and result. This makes it easier to distinguish a meaningful change from normal answer variation.

A dashboard can track:

Citation rate: the proportion of tested answers that link to the company’s domain.
Brand-mention rate: the proportion that name the company, product, or service.
Recommendation rate: the proportion that present the brand as a suitable option.
Third-party mention rate: the proportion of brand mentions supported by external pages.
Context and sentiment: whether the mention is positive, neutral, negative, or qualified.
Business outcomes: visits, assisted conversions, opportunities, and revenue where attribution is available.

Analytics tools add a referral layer to prompt testing. Google Analytics introduced an “AI Assistant” default channel group on May 13, 2026, to group referral traffic from platforms including ChatGPT, Gemini, and Claude. Microsoft Clarity also made its Citations dashboard generally available for Copilot grounding-query and citation analysis.

Referral analytics are incomplete. Last-click reporting may capture a later AI referral while missing earlier exposure, branded search, or direct visits. Treat referral data as one measure of influence, not a complete record of it.

Conversion rates from AI search referrals

AI referral traffic is small on average but may carry strong intent. Conductor’s 2026 benchmark reported that AI referrals represented 1.08% of total website traffic, grew about 1% month over month, and came primarily from ChatGPT, which generated 87.4% of that traffic in the benchmark.

Several studies reported higher conversion rates for AI-referred visitors. Ahrefs found that AI-search visitors produced 12.1% of signups while accounting for 0.5% of total traffic, a reported 23x conversion-rate premium over standard organic traffic. Similarweb reported an 11.4% conversion rate for AI referral traffic versus 5.3% for organic search. Adobe Digital Insights reported that AI-referred traffic converted 42% better than non-AI traffic.

These figures are directional, not interchangeable. The studies use different datasets and definitions, and high conversion may reflect users who were already predisposed to buy. The more useful test is whether AI visibility creates incremental qualified pipeline rather than whether a small referral segment converts well.

Measure citations and mentions alongside assisted conversions, direct traffic, branded search, and opportunity creation. A citation may generate a click, while a brand mention may influence a later visit without producing an identifiable referral.

Structured data and generative visibility

Structured data can help describe visible entities and content, but current evidence does not support treating special schema as a reliable way to force AI recommendations.

Google’s AI-search documentation, updated July 10, 2026, says that no special schema.org markup is required for eligibility in AI Overviews or AI Mode. Google also states that llms.txt, content chunking, AI-specific rewriting, and special schema are unnecessary for AI feature visibility. Markup that does not match visible page content can violate Google’s structured-data policies, so entity markup should describe verifiable on-page facts.

Research on schema is mixed. Some studies report associations between structured data and AI citations. However, Ahrefs’ matched study of 1,885 pages found no causal lift from adding JSON-LD. Google AI Overview citations fell 4.6% relative to controls, while changes in AI Mode and ChatGPT citations were statistically indistinguishable from zero. A searchVIU experiment cited by Ahrefs found that several AI systems extracted visible HTML during direct retrieval while ignoring JSON-LD, hidden Microdata, and hidden RDFa.

The measured conclusion is limited: valid structured data may support clear entity representation, but it should not replace visible, authoritative content or external reputation.

Research also disagrees about the relationship between traditional rankings and AI citations. Ahrefs reported that 76.1% of cited URLs ranked in Google’s top 10. BrightEdge reported that about 17% of citations came from the organic top 10, while AirOps reported that roughly 60% came from URLs outside the top 20. Classic SEO appears relevant, but the size and form of that relationship remain unsettled.

A durable strategy makes the company’s identity, products, category, and evidence clear in visible content while building credible third-party coverage. Then measure both outcomes: whether AI systems cite the domain and whether they name the brand as a relevant recommendation.

Sources

Published through BrandKarma

Christoph Menge built the product. This article was researched and published through it, and delivered over the same public content API your developers would call. About · How it works · Pricing.

See what AI says about you

If this article named a gap you already feel, request the report. We run your top categories across ChatGPT, Perplexity and Gemini and send it within 48 hours.

See what AI says about you