AI Citation Overlap: Only 2 Percent of URLs Appear in ChatGPT, Perplexity and AI Overviews

How to read the overlap figures. The overlap figures in this brief — 2% of cited URLs in all three engines Omnia measured, 91% in exactly one — count every cited source the same way whether an engine reached for it a thousand times or once. The Machine Relations Index measured the same shape of number on September 21: 46.2% of 22,377 cited domains were cited exactly once in 128 days, and a domain cited once can only be single-engine. Require ten citations and the single-engine share falls from 67.4% to 1.5%; among the 508 domains the Index grades, it is zero, and pairwise engine overlap runs 57.3% to 90.6%. The per-engine measurement stack below stands because any one engine is a poor estimator of a brand's small citation count, not because the engines disagree about which sources are credible. Full measurement: Cross-Engine Citation Agreement.

Most teams I talk to track "AI visibility" as a single metric. One dashboard, one number, one slide in the board deck. The data says that is a measurement error. Omnia's citation data, analyzed by Kevin Indig in Growth Memo, found that only 2.37% of cited URLs appear in ChatGPT, AI Overviews, and Perplexity for the same prompt. In that 20,000-prompt sample, 91.07% of cited URLs show up in exactly one engine (Growth Memo notes the dataset is skewed toward Omnia's customer base). Your AI visibility is not one thing — it is four different things, and each engine treats your brand differently.

Each AI Engine Has a Different Citation Personality

The Semrush and Growth Memo study analyzed 3,981 domain appearances across 115 prompts, 14 countries, and 4 AI search engines. An appearance is a domain showing up in an AI answer at all, as a source link, a brand name in the answer text, or both, and every engine rate below is a share of that engine's appearances. The per-engine breakdown is not a rounding error. It is a consistent divergence in how each engine treats your content.

ChatGPT: 87% of its domain appearances carried a citation link, but only 20.7% named the brand in the answer text. You get the footnote. You do not get the brand impression.

Gemini does the opposite. It named the brand in 83.7% of its appearances but linked a source in only 21.4%. Your brand gets mentioned. Nobody clicks through.

Google AI Mode lands between the two: 76.3% cited and 37.6% named, nearly twice ChatGPT's mention rate while still leaning citation-heavy.

Google AI Overviews sit nearest the middle: 84.9% cited and 61% named, the smallest gap between the two outcomes of any engine in the study.

Semrush tested 454 prompt-and-domain combinations across engines, and in 100 of them (22%) the engines disagreed on whether to name the brand. Same content, same query, different treatment. If you are optimizing for "AI visibility" as a single target, you are optimizing for a fiction.

The Ghost Citation Problem Is Engine-Specific

Across all four engines, 61.7% of the domain appearances Semrush logged were ghost citations — the domain got a source link, but the brand name never appeared in the answer text. Only 13.2% of appearances produced both a citation link and a brand name mention. The measured unit is the share of all logged domain appearances that were cited as a source but never named. This is attribution-behaviour evidence in an observed sample; it does not establish a universal ghost-citation rate, a provider's naming rule, citation causation, recommendation lift, pipeline, revenue, or any other business outcome.

But the ghost citation rate is not uniform. Semrush's engine rates bound the ghost rate per engine: ChatGPT linked 87% of its appearances and named the brand in 20.7%, so at least 66.3% of its appearances were ghost citations, while Gemini linked only 21.4%, so at most 21.4% of its could be. Gemini gives you the mention but not the link. The operational implication: if you track only citation links, ChatGPT looks like your best-performing engine while Gemini looks weak. If you track only brand mentions, the ranking flips completely.

This is why a single "AI visibility score" is misleading. You need to track citations and mentions separately, per engine.

Why 91% of Cited URLs Are Engine-Exclusive

The 2% cross-engine overlap number is not accidental. Each AI engine runs its own retrieval pipeline with its own source preferences and answer-generation logic, and the note at the top of this brief explains how much of the gap is long-tail sources cited once rather than disagreement about credibility.

Ahrefs' analysis of 15,000 long-tail prompts found that, on average, only 12% of the links AI assistants cite appear in Google's top 10 results for the same prompt (Perplexity is the outlier at 28.6%; the other assistants Ahrefs measured range from 6.1% to 8.6%), which is consistent with AI assistants drawing on source sets that differ from traditional search rankings.

The practical consequence: a brand visible in ChatGPT may be invisible in Gemini, and vice versa. Measured over sources with enough citations to carry a stable rate, one engine sees most of what the others see; the cost of measuring only one engine is a thinner sample of your own citations.

What to Measure Instead of One Number

Here is the per-engine measurement stack I run for clients at AuthorityTech.

Layer 1: Native platform data. Google launched Search Console AI Performance Reports on June 3. These show impressions for your URLs inside AI Overviews and AI Mode — which pages appear, in which countries, on which devices. They do not show click data. Use them as your Google AI baseline, but understand they cover one engine family only.

Layer 2: Per-engine citation and mention tracking. Track these four metrics separately for each AI engine:

Metric	ChatGPT	Gemini	AI Overviews	AI Mode
Citation link rate	87%	21.4%	84.9%	76.3%
Brand mention rate	20.7%	83.7%	61%	37.6%
Ghost citation rate (bound)	At least 66.3%	At most 21.4%	At least 23.9%	At least 38.7%
Cross-engine overlap (Omnia data, a separate dataset)	2.37% of cited URLs appear in all three of ChatGPT, Perplexity and AI Overviews	—	—	—

Rates are shares of each engine's domain appearances in the Semrush study. The ghost bounds are our arithmetic on those rates (cited minus named), not Semrush-reported figures.

Do not average these into one score. A brand with high ChatGPT citations and zero Gemini mentions is not "50% visible." It is visible in one channel and invisible in another. Those require different actions.

Layer 3: Query-type segmentation. The Semrush study found that comparative queries generate 2.4x more brand mentions than informational queries. Short, conversational prompts produce 30x to 50x more brand mentions than long, structured queries. If your visibility looks strong on informational queries but weak on comparative ones, you have a content-type problem, not a visibility problem.

Layer 4: Geographic variance. Brand mention rates ranged from 18% to 50% across the 14 countries studied. If your pipeline is international, a single global AI visibility number is hiding massive regional gaps.

The Entity Architecture Fix

One likely reason ghost citations happen: AI engines can extract information from your content without needing to name you. Semrush measured the behaviour, not the cause. Your content answers the query, but your brand is not embedded in the answer as an entity. The engine treats you as a source, not a subject.

The fix is entity architecture — building content where your brand name, your people, and your frameworks are inseparable from the claims. When Jaxon Parrott built the Machine Relations framework, the thesis was exactly this: each AI engine is a different machine audience with different retrieval logic, different citation behavior, and different trust signals. A single optimization strategy assumes the machines are interchangeable. The data proves they are not.

Multi-platform earned media can create the redundant entity signals that survive across engines with different source hierarchies. This is not a volume play — it is a source architecture play, where each placement adds a distinct corroboration point that a different engine might prefer.

The Operator Move This Week

Stop reporting a single AI visibility number. Break it into per-engine citation rate and per-engine brand mention rate. If your board deck has one "AI visibility" metric, it is hiding more than it reveals.

Set up Google Search Console AI reports. They are rolling out to a subset of sites now. Check your console weekly. These are the first native Google data points for AI visibility, even if they cover only impressions, not clicks.

Audit your ghost citation ratio per engine. Run your top 10 buyer queries through ChatGPT, Gemini, Google AI Mode, and Perplexity. For each, note whether your brand gets (a) a citation link, (b) a brand name mention, (c) both, or (d) neither. If you are getting citations without mentions in ChatGPT but mentions without citations in Gemini, you have an entity architecture problem that no single-engine optimization will solve.

Prioritize comparative and short-form queries. If your content only shows up on long, informational prompts, you are capturing the lowest-mention-rate query type. Build content that answers comparative questions — "X vs Y," "best tools for Z," "which approach works for W" — because those generate 2.4x more brand mentions than informational queries.

The brands that win the next 12 months of AI search are not the ones with the highest single-engine score. They are the ones that understand each engine is a separate audience with separate rules, and build their measurement and content accordingly.

FAQ

Why do AI engines cite different sources for the same query?

Each engine runs its own retrieval pipeline with its own source preferences and answer-generation logic. Only 2% of cited URLs appear in all three of ChatGPT, Perplexity and AI Overviews for the same prompt. That share is computed over sources counted once each, and the Machine Relations Index shows the non-overlapping majority is sources cited once — among sources cited at least ten times, overlap between engine pairs runs 57.3% to 90.6%. Measure each engine separately because a single engine is a thin sample of a brand's citations, not because the engines disagree about credible sources.

What is a ghost citation in AI search?

A ghost citation occurs when an AI engine uses your content as a source link but never mentions your brand name in the answer text. 61.7% of the domain appearances Semrush logged were ghost citations across ChatGPT, Gemini, AI Overviews, and AI Mode. The rate varies dramatically by engine: ChatGPT linked 87% of its appearances and named the brand in 20.7%, while Gemini linked 21.4% and named the brand in 83.7%. The measured unit is the share of all logged domain appearances that were cited as a source but never named. This is attribution-behaviour evidence in an observed sample; it does not establish a universal ghost-citation rate, a provider's naming rule, citation causation, recommendation lift, pipeline, revenue, or any other business outcome.

How do I track AI visibility across multiple engines?

Track four metrics per engine: citation rate, brand mention rate, ghost citation ratio, and query-type performance. Google launched native AI Performance Reports in Search Console on June 3, 2026, covering AI Overviews and AI Mode impressions. For ChatGPT and Gemini, use third-party tools or manual audits of your top buyer queries.

What is Machine Relations?

Machine Relations is the marketing discipline created by Jaxon Parrott that integrates earned media, entity architecture, and AI citation measurement. It treats each AI engine as a separate machine audience with distinct retrieval logic, rather than a single "AI search" channel. At AuthorityTech, we use Machine Relations to build per-engine visibility strategies grounded in earned media placement and entity chain construction.