Being Cited by AI Is Not the Same as Being Seen

Every SEO tool in 2026 will happily show you a number that says your site is being "cited" by ChatGPT, Gemini, or Google's AI Overviews. Agencies put that number in monthly reports. Founders screenshot it for the board deck. And for most of those citations, it means almost nothing to the person actually reading the AI's answer.

A June 2026 study run by Semrush in partnership with SEO strategist Kevin Indig and Growth Memo put a number on the gap between "cited" and "seen." They logged 3,981 domain appearances across 115 prompts tested in 14 countries, spread across four AI search engines — ChatGPT, Google AI Overviews, Gemini, and Google AI Mode — and classified every single appearance into one of two outcomes: was the domain cited (a source link appeared in the response) and was the brand mentioned (the brand's actual name appeared somewhere in the answer text)? (Semrush, The Ghost Citations Study, June 2026)

The result: 61.7% of all citations were what Indig calls ghost citations — the AI links to your page as a source, but never says your brand's name in the words a person actually reads. The source link sits in a footnote or a small citation chip most users never click. The reader walks away having consumed your content, credited it to no one, and remembering nothing about who wrote it.

The three shapes of an AI appearance, from 3,981 logged instances: 61.7% were ghost citations (linked, never named). 13.2% were both cited and named — the rarest outcome, and the only one that reliably builds brand recognition. 25.1% were named with no citation link at all. Add it up and 74.9% of appearances included a citation, but only 38.3% included a brand mention — meaning citation rate runs almost double the mention rate. (Semrush/Kevin Indig, 2026)

If your AI visibility dashboard only tracks citations, you are, by this data, overstating your actual brand exposure by roughly 2x. And the gap isn't evenly distributed across platforms — the study's engine-by-engine breakdown found ChatGPT cites sources in the large majority of responses but names the underlying brand in only about one in five, functioning, as Indig put it, "like an academic paper with footnotes: lots of references at the bottom, names rarely said in the text" (Cite Solutions, citing the Growth Memo/Semrush data, 2026).

Why This Matters More Than Your Citation Count

Citations without mentions still send some traffic — a curious reader clicks through, sees your brand for the first time on your own site. But most people reading an AI answer don't click through at all. Multiple 2026 studies on AI search behavior put no-click rates on AI-generated answers well above 90% (a pattern this site has covered separately in our reporting on AI Mode's 93% no-click session rate). When someone doesn't click, the only brand signal they get from your appearance is whatever the AI actually said out loud. If your name isn't in that sentence, you were the invisible source behind someone else's answer.

That distinction — citation vs. mention — maps onto two different business outcomes. A citation without a mention gets you, at best, some referral traffic from the small share of users who click through and check where the answer came from. A mention gets you brand recall whether anyone clicks or not, which compounds the same way branded search volume compounds: the more times a buyer sees your name attached to a topic, the more likely they are to type your name directly next time, skip the comparison shopping, and convert faster.

The reframe: stop treating "AI visibility" as one metric. It's two separate metrics that move independently — citation presence and mention presence — and most AI visibility tools collapse them into a single number that hides exactly the gap that matters. Measure them side by side, and if you're a consumer brand or service business (as opposed to a pure publisher), weight brand mentions higher than citation counts when deciding where to invest content effort.

What Actually Predicts Whether You Get Named, Not Just Linked

The most useful part of Indig's research wasn't the headline ratio — it was what correlated with landing in the rare "cited and mentioned" bucket instead of the common "ghost citation" bucket. Comparison and evaluation content — "best X for Y," "X vs. Z," buyer's guides that explicitly weigh named options against each other — generated roughly 30x more brand-name mentions than purely informational, how-to-style content covering the same topics (Cite Solutions, 2026). That tracks with how these AI systems actually work: an AI answering "what is X" pulls from whatever source explains the concept best, and doesn't need to name a brand to convey the definition. An AI answering "which is better, X or Y" has to attach a name to each side of the comparison to make the answer coherent — so it names brands almost by structural necessity.

A follow-up Semrush study published two months later, "Topical Gravity," dug into a related question: once a brand starts getting mentioned consistently in one topic area, does that authority transfer anywhere else? The dataset was large — 1,094 US categories tracked monthly in ChatGPT from January through June 2026, covering 283,215 domain-category citation observations and 76,493 brand-mention observations across more than 1,458 mapped brand entities (Semrush, How Topical Authority Spreads (and Where It Doesn't) in ChatGPT, August 2026).

The finding: citations travel easily across categories if your content simply meets ChatGPT's source criteria — relevance, authority, freshness. Brand mentions do not travel the same way. Getting named by ChatGPT happened almost entirely within categories close to a brand's established core expertise, and depth in a category (showing up consistently across many related prompts) mattered far more than breadth (showing up once across twenty unrelated categories). A brand present in one prompt out of twenty different topics built essentially no mention share anywhere. A brand present in four-of-five prompts within one topic built durable ownership of that topic.

Topic ownership is rare, and once won, it's sticky: a companion Semrush analysis of the same 1,094-category dataset found only 15.2% had a brand that qualified as a clear category "owner" (highest mention share, present in at least 4 of 5 related prompts, leading the runner-up by 5+ percentage points). Another 31.2% had an emerging leader; 53.7% remained completely unsettled with no dominant brand. But once a brand did establish a clear lead, it held that lead into the following month 90.4% of the time. (Search Engine Land, reporting on Semrush data, July 2026)

Two more findings from that same dataset should change how you think about "authority" in the AI-search era. First, the categories with the highest search demand were less likely to have a clear owner, not more — only 11.3% of high-volume topics had an established leader, versus 19% of lower-volume topics, even though the high-volume half accounted for 98% of all AI search demand in the sample. Translation: most of the AI search volume that matters is still up for grabs. Second, traditional SEO strength barely predicted who won a topic. When Semrush compared category owners against their closest competitors, owners had higher branded search volume in only 55.7% of head-to-head comparisons, higher organic traffic in 48.4% (worse than a coin flip), and a higher Authority Score in just 52.5% — statistically indistinguishable from chance. Only branded search volume reached statistical significance as a predictor (Search Engine Land, 2026).

Maybe the most counterintuitive number in the whole dataset: being the domain ChatGPT cites most often in a category and being the brand ChatGPT names most often in that same category overlapped in only 21% of cases. Four out of five times, the site earning the most footnotes wasn't the brand actually getting remembered (SEO Kreativ, summarizing the Semrush/Growth Memo dataset, 2026).

What this means in practice: domain authority, backlink count, and even your organic ranking position — the metrics most SEO strategy has optimized for over the last two decades — do not reliably decide who an AI names in an answer. What decides it is topical depth (showing up across most of the related questions in one topic, not scattered coverage across many), content that names and compares options rather than just explaining concepts, and consistent brand naming that lets AI systems tie a name to a subject with confidence.

How to Actually Build Toward Mentions, Not Just Citations

None of this means citations are worthless — for research-heavy or reference content, citations remain the baseline signal that your content is even in the pool AI systems draw from. But if the goal is brand recognition and demand generation rather than pure referral traffic, the strategic shift is specific and actionable:

  1. Pick two or three topics to fully own before trying to cover twenty. The data is explicit on this: showing up once across many categories built no measurable mention share, while depth within a narrow set of related topics built durable ownership that held for months. Map the 4-5 prompt variants a real buyer would ask around your core topic (the definition question, the comparison question, the "best X for Y" question, the problem-solution question, the recommendation question) and make sure you have strong, citable content answering each variant — not just the one you'd naturally think to write first.
  2. Build comparison and evaluation content deliberately, not as an afterthought. Given the roughly 30x gap in brand-mention generation between comparison content and purely informational content, a "Best [category] for [use case]" or "[Your category] buyer's guide" page that explicitly names and weighs options is doing more mention-building work than another how-to post on the same topic.
  3. Keep brand naming and positioning identical everywhere. AI systems tie citations to entities more reliably when a brand name, description, and category association stay consistent across your own site, review platforms, and third-party mentions. Inconsistent naming (abbreviations, rebrands, inconsistent legal-vs-marketing names) makes it harder for an AI system to confidently attach a name to your content, even when it's using that content as a source.
  4. Earn mentions in the sources AI answers already cite for your topic. Review platforms, industry publications, and the forums/communities where your actual buyers compare options feed AI answers directly. A brand that's discussed by name on Reddit threads, review sites, and trade press is more likely to get named by an AI system than one whose only footprint is its own website, even a well-optimized one.
  5. Track citations and mentions as two separate KPIs, starting now. If your current AI-visibility reporting is a single blended number, split it. Most AI-visibility tools (including Semrush's own AI Visibility Toolkit) can show citation presence and brand-mention presence separately — use both, and treat a rising citation count with a flat mention count as a signal that your content is being used, not remembered.

The good news buried in the data: 85% of the 1,094 topic categories Semrush tracked in ChatGPT had no established owner at all as of mid-2026. Topic ownership in AI search is still mostly unclaimed — and because traditional SEO authority metrics barely predict who wins it, a smaller brand with genuine topical depth has a real, data-backed opening that didn't exist in traditional Google rankings, where domain authority has compounded for incumbents for two decades. (Semrush, Topical Gravity, August 2026)

Where This Fits Alongside Your Existing SEO Work

This isn't a replacement for the fundamentals — it's an added layer on top of them. Sites still need the technical foundation, on-page structure, and E-E-A-T signals that earn citations in the first place (see our on-page SEO checklist and entity SEO and E-E-A-T playbook for that groundwork). What's different in 2026 is that earning the citation is no longer the finish line. A page can rank well, get crawled, and get linked as a source in an AI answer, and still fail to build any brand recognition if it never gets named — which is exactly the ghost-citation pattern this data quantifies for the first time.

It also connects directly to the measurement gaps we covered in our reporting on the August 2026 AI traffic-and-conversion shift: GA4's own "AI Assistant" channel and Search Console's generative AI report can tell you an AI surface sent a click, but neither tool tells you whether your brand was actually named in the answer that produced it. Citation-vs-mention tracking has to happen with a dedicated AI-visibility tool, because your existing analytics stack wasn't built to see the gap at all.

Frequently Asked Questions

What is a "ghost citation" in AI search?

A ghost citation is when an AI system like ChatGPT, Gemini, or Google's AI Overviews links to your page as a source in its answer, but never mentions your brand's name anywhere in the actual response text. The reader gets your content's information without ever learning who produced it. Kevin Indig coined the term, and a 2026 Semrush study found 61.7% of all AI citations fall into this category.

How common are ghost citations across different AI platforms?

Across the four platforms tested in Semrush's 2026 study — ChatGPT, Google AI Overviews, Gemini, and Google AI Mode — ghost citations made up 61.7% of the 3,981 domain appearances logged. ChatGPT specifically cited sources in the large majority of its responses but named the underlying brand in only about one in five, the widest citation-to-mention gap among the platforms tested. Rates vary by platform and by whether the query is informational or comparison-based.

Does getting cited by AI still matter if my brand isn't mentioned?

Yes, but less than most people assume. A citation without a mention can still generate some click-through traffic from users who check the source, and it confirms your content is at least in the pool of material AI systems consider credible. But since most AI-answer sessions never result in a click, a citation alone builds little to no brand recognition. If your goal is demand generation rather than pure referral traffic, brand mentions matter more.

What kind of content is most likely to get my brand actually named by AI?

Comparison and evaluation content — buyer's guides, "best X for Y" roundups, and head-to-head comparisons that explicitly name and weigh options — generated roughly 30 times more brand-name mentions than purely informational, how-to-style content in the same topic, according to Kevin Indig's research with Semrush. Content that names options directly gives an AI system something concrete to attach your brand name to when it forms an answer.

Does my site's domain authority or backlink profile predict whether AI names my brand?

Barely. A Semrush analysis of 1,094 topic categories tracked in ChatGPT found that the brand with the highest Authority Score won topic ownership in only 52.5% of head-to-head comparisons against the runner-up — statistically close to a coin flip. Branded search volume was the only traditional metric that reached statistical significance as a predictor. Topical depth (consistent presence across most of a topic's related questions) mattered far more than raw domain authority.

How do I start tracking citations and brand mentions separately?

Most dedicated AI-visibility platforms, including Semrush's AI Visibility Toolkit, report citation presence and brand-mention presence as separate metrics rather than blending them into one score. Start by pulling both numbers for your top 10-20 target topics. If citations are climbing while mentions stay flat, that's a direct signal your content is being used as source material without building any recall — and a cue to shift toward comparison-style content and more consistent brand naming across the platforms AI systems already cite for your industry.