Skip to main content
Back to List
geo·Author: RanketAI Editorial Team·Updated: 2026-06-13

Ghost Citations: Why 62% of AI Citations Never Mention Your Brand

Ghost citations — source links that never name your brand — make up 61.7% of AI citations, per a Semrush and Kevin Indig study of 3,981 domain appearances. Why citations and mentions are separate axes, how ChatGPT and Gemini differ, and what to measure.

AI-assisted draft · Editorially reviewed

This blog content may use AI tools for drafting and structuring, and is published after editorial review by the RanketAI Editorial Team.

Key takeaway: A ghost citation is an AI answer that links your page as a source while never saying your brand's name in the answer text. In a Semrush and Kevin Indig study of 3,981 domain appearances across four AI engines, 61.7% of all citations were ghost citations (study). Being cited does not mean being seen — citations and mentions need to be measured as separate axes (as of 2026-06-13).


TL;DR

  • Citations and mentions are different events. Appearing as a source link and being named in the answer text happen independently — both occurred together in only 13.2% of appearances.
  • Engines behave in opposite ways. ChatGPT is citation-heavy (87% cited, 20.7% mentioned); Gemini is mention-heavy (83.7% mentioned, 21.4% cited). Visibility in one engine cannot be assumed to carry over to another.
  • Your site type decides the primary metric. Publishers and research-style sites should track citations first; consumer brands should track mentions first. The full picture needs both axes.

What a ghost citation is — cited, but never named

A ghost citation is an appearance where the AI links your page as a source but never writes your brand name into the answer itself. Kevin Indig coined the term, and the joint study with Semrush (June 2026) is the first to quantify it.

The method is straightforward. The study ran 115 prompts across 14 countries on four engines — ChatGPT, Google AI Overviews, Gemini, and Google AI Mode — collecting 3,981 domain appearances, each classified on two outcomes: cited (appeared as a source link) and mentioned (brand name appeared in the answer text). Every ratio in this post comes from that single source, so read the numbers as structure and trend rather than absolutes.

The core finding: a citation is widely treated as an authority signal, but whether users actually see your brand is a separate question. In the researchers' words:

"The AI knows the information about the brand came from somewhere, but doesn't feel the need to explicitly say so to users. The brand name carries on its own." — Kevin Indig, Ghost Citations Study

The AI knows where the information came from and even links to it — it just does not bother telling the user the source's name.

The three shapes of an appearance — 61.7% · 25.1% · 13.2%

The 3,981 appearances split into ghost citations at 61.7%, mentions without a citation at 25.1%, and both together at just 13.2%. The rarest outcome is the one most teams assume is the default.

Appearance type Share What it means
Ghost citation (cited, not mentioned) 61.7% A source link exists, but users never see the brand name
Mention only (mentioned, not cited) 25.1% The brand is named, with no source link
Cited + mentioned 13.2% Link and name both appear — the rarest shape

Adding these up, 74.9% of all appearances included a citation, but only 38.3% included a brand mention — the citation rate is roughly double the mention rate. That gap sits between the report "our site is being cited by AI" and the reality "users see our brand in AI answers."

Every engine behaves differently — ChatGPT footnotes, Gemini converses

ChatGPT is citation-driven and Gemini is mention-driven — and for the same prompts, the brands the two engines surfaced barely overlapped. The per-engine numbers:

  • ChatGPT — cites in 87% of appearances, mentions in only 20.7%. Its answers read like papers with footnotes.
  • Gemini — mentions in 83.7% of appearances, cites in only 21.4%. It answers like a conversation drawing on what it already knows.
  • Google AI Overviews — sits between the two, leaning toward citations.
  • Google AI Mode — mentions brands at nearly twice ChatGPT's rate, but still behaves closer to a footnoted research piece.

The researchers' summary:

"There's almost no overlap between which brands ChatGPT cites and which ones Gemini names for the same prompt. These are different behavioral systems. Treat them that way." — Kevin Indig, Ghost Citations Study

The practical implication is direct: strong citation performance in ChatGPT says nothing about your visibility in Gemini, and vice versa — each engine needs its own check. The study covered Google's engines and ChatGPT; engines with yet another behavioral pattern exist too, such as Perplexity with its citation-forward answer format, so a wider measurement scope is the safer default.

Strong brands get named; informational sites get linked

The stronger the brand identity, the more it gets named without a citation — and the more informational or aggregator-like the site, the more it gets linked without a name. The study's extremes make the pattern visible.

  • Google was named in answers about 2.78 times more often than a Google URL was cited as a source; Apple's name appeared roughly twice as often as its citations.
  • At the other end, medium.com was cited 16 times in the dataset and never once named in an answer. Wikipedia and university domains followed the same pattern — consistent with Ahrefs' most-cited domains in Perplexity, where informational sites dominate the citation rankings.

Query and content shape matter too. Short, conversational queries produced 30x–50x more brand mentions than long prompts, and comparative content produced 2.4x more brand mentions than informational content. Informational content earns citations; comparative content earns mentions. "A vs B" formats are what pull a brand name into the answer text itself.

What to measure — two axes, tracked separately, viewed together

Collapsing citations and mentions into one "AI visibility" number makes ghost citations invisible. Measure the two axes separately and view them side by side. The primary metric depends on your site type:

  • Publishers and research-style sites — citations come first. Source links naturally outnumber name mentions for this content type, so the citation trend is your baseline.
  • Consumer brands and SaaS — mentions come first. Users build recognition from seeing the name in the answer, so priority goes to what raises mention rates: comparative content and clear entity signals.

RanketAI's brand visibility analysis measures, repeatedly and per engine, whether your brand is mentioned in the answer text and whether it is cited as a source across ChatGPT, Perplexity, and Gemini, and competitor comparison shows how rival brands surface for the same questions. Tools such as Semrush AI Visibility and Ahrefs Brand Radar also track citations and mentions — the key is picking a tool that fits your market and language, and watching both axes together.

Frequently asked questions

Is a ghost citation a bad signal?

It depends on your site type. For publishers and informational sites, ghost citations still deliver traffic and authority, so they are close to the natural state. If brand recognition is the goal, a ghost citation means doing the work without getting the name out — check your mention rate separately and work on raising it.

Should I grow citations or mentions first?

If your revenue model is the content itself (subscriptions, advertising), citations come first. If you sell a product or service, mentions come first. In the study, comparative content produced 2.4x more mentions, so "X vs Y" and "which one for which situation" formats are an efficient starting point when mentions are the goal.

Should I optimize for ChatGPT or Gemini?

Both. One of the study's core findings is that the brands the two engines surfaced for the same prompts barely overlapped. Optimizing against one engine leaves the other engine's gaps invisible. Measure citations and mentions per engine, then fill the empty engine-by-question combinations first.

Do these patterns hold for non-English answers?

The study spans 14 countries, and mention rates varied widely — from 18% to 50% by country — so there is no guarantee the exact figures transfer to your language market. Treat the structure (citations ≠ mentions, engines differ) as portable, but verify your own brand's actual citation and mention rates in your language with direct measurement.

  • Google Search Console's new Generative AI reports — the free way to see your Google-side AI impressions. Seeing ghost citations requires the next layer of measurement.
  • How to build an answer block ChatGPT quotes verbatim — the content structure that earns citations, pairing with the comparative formats that earn mentions.

Execution Summary

ItemPractical guideline
Core topicGhost Citations: Why 62% of AI Citations Never Mention Your Brand
Best fitPrioritize for geo workflows
Primary actionStandardize an input contract (objective, audience, sources, output format)
Risk checkValidate unsupported claims, policy violations, and format compliance
Next stepStore failures as reusable patterns to reduce repeat issues

Frequently Asked Questions

What problem does "Ghost Citations: Why 62% of AI Citations Never…" address, and why does it matter right now?

Start with an input contract that requires objective, audience, source material, and output format for every request.

What level of expertise is needed to implement Ghost Citation effectively?

Teams with repetitive workflows and high quality variance, such as geo, usually see faster gains.

How does Ghost Citation differ from conventional geo approaches?

Before rewriting prompts again, verify that context layering and post-generation validation loops are actually enforced.

Data Basis

  • Core data: Semrush × Kevin Indig (Growth Memo) joint study (2026-06-09) — 115 prompts run across 14 countries on four engines (ChatGPT, Google AI Overviews, Gemini, Google AI Mode), yielding 3,981 domain appearances classified as cited (source link) versus mentioned (brand name in the answer). All ratio figures in this post come from this single source.
  • Supporting data: Ahrefs' most-cited domains in Perplexity (2026-06) — used as a cross-reference for the pattern that informational and aggregator sites dominate citation rankings.

Key Claims and Sources

This section maps key claims to their supporting sources one by one for fast verification. Review each claim together with its original reference link below.

External References

The links below are original sources directly used for the claims and numbers in this post. Checking source context reduces interpretation gaps and speeds up re-validation.

Is your site visible in AI search?

See for free how ChatGPT, Perplexity, and Gemini describe your brand.

Start Free Diagnosis →

Related Posts

These related posts are selected to help validate the same decision criteria in different contexts. Read them in order below to broaden comparison perspectives.

RanketAI Guide #10: Building a ChatGPT Answer Block It Will Quote

ChatGPT lifts a citation-ready 'block' instead of reading your whole page. Learn the four-element answer block (direct answer, statistic, inline source, date) and a template to raise your ChatGPT citation odds, from RanketAI's measurement view.

2026-06-07

Google Search Console's New Generative AI Reports: What They Show and What's Missing

Google launched Generative AI performance reports in Search Console on June 3, 2026 — the first official view of AI Overviews and AI Mode impressions. We cover the five dimensions, missing click data, the blocking toggle, and measuring beyond Google.

2026-06-12

RanketAI Guide #11: Put Your Brand Inside ChatGPT with a Custom GPT

ChatGPT visibility isn't only about being cited on the web. Build a brand Custom GPT — instructions, knowledge, actions — list it in the GPT Store, and occupy an internal surface inside ChatGPT itself, with RanketAI's own example.

2026-06-08

AI Search Asks in 23 Words, Not 4: Designing Content for Conversational Queries (2026)

AI search queries average 23 words and sessions run about 6 minutes — nothing like traditional search (4 words, seconds). We cover how to win the "first question" and design content for follow-ups in the long-tail era, with 2026 data.

2026-06-05

Why AI Search Traffic Converts Higher: Measuring GEO ROI and Attribution (2026)

AI referral traffic is only about 1% of the web, yet it reportedly converts 4x or more higher than organic search. We cover why it converts, why the numbers vary so much by source, and how to split and measure AI traffic in GA4 to justify GEO ROI.

2026-06-04