Invisible to AI: 85% of AI citations go to sites brands don't own

Traditional SEO reports show green arrows: Google rankings are rising, organic traffic is stable. Yet the sales team reports falling conversions, and prospects keep asking, "Why hasn't anyone heard of you?" The gap is not a reporting error. It is a shift in where buyers now start.

B2B research now starts inside an AI, not a search box

According to G2's "The Answer Economy" report (April 2026, a March 2026 survey of 1,076 B2B software buyers), 51% of B2B software buyers now begin their research with an AI chatbot more often than with Google, up from 29% a year earlier. The starting point of the buying journey has moved, and most measurement has not moved with it.

The problem is simple to state. Traditional SEO measures clicks. In AI search, citations are what count, and here most brands are simply invisible. Below are the hard numbers that define the new landscape, drawn from recent academic and industry research, with sources listed at the end.

The new search landscape, in four numbers

1. Your own website is not AI's main source

A June 2026 arXiv study by Dmitrij Zatuchin, "How Large Language Models Source Brand Reputation Across Languages and Markets", analysed 167,551 URL-grounded citations and found a stark split: 85.7% of the citations behind AI answers about brands point to sites the brand does not own. Only 14.3% link to a company's own domain. AI systems lean on external verification more than on a brand's own pages.

2. Earned media dominates AI citations

Muck Rack's "What Is AI Reading?" study (May 2026 edition, more than 25 million links cited by ChatGPT, Claude and Gemini) confirms the pattern: 84% of AI citations come from earned media, and the share has stayed between 82% and 89% across the three editions since July 2025. Paid and advertorial content accounts for 0.3%.

3. The third-party discovery effect

AirOps's analysis of 21,311 brand mentions across ChatGPT, Claude and Perplexity, "The Influence of Offsite Signals in AI Search", found that for top-of-funnel commercial queries brands are 6.5 times more likely to be mentioned through third-party content than through their own site; 85% of mentions came from external domains. If you are absent from the comparison pages, industry publications and review profiles AI already reads, you effectively do not exist for AI.

4. Old SEO tricks now work against you

The paper "GEO: Generative Engine Optimization" (Aggarwal et al., Princeton University and IIT Delhi, KDD 2024) tested strategies across 10,000 search queries. Adding quotations, statistics and an authoritative tone raised visibility by up to 40%. Keyword stuffing did the opposite, cutting visibility by roughly 10% against the base page. AI models penalize content that reads as if it were written for robots.

Why this quietly kills your sales funnel

Picture a prospect asking an assistant: "Recommend a trusted provider of software X in Poland." The AI does not return ten blue links. It returns one synthetic answer. If that answer cites a forum thread, a competitor's article or a stale review, and your company is not named, you lose the lead.

Worse, that loss never shows up in Google Analytics as a referral, because the user never clicked. This is the dark funnel: demand that forms and resolves inside an AI answer, invisible to your analytics. You cannot optimize what you do not measure, and measuring it takes a different approach than an SEO audit. For the mechanics behind this, see how AI assistants decide which businesses to recommend.

The four layers that decide whether AI cites you

At CitePulse we do not guess how AI "sees" a page from its HTML. We measure it directly, asking real buyer questions to ChatGPT, Perplexity, Gemini and Claude. Across thousands of audits, four layers consistently build the AI Reputation Score:

• Readable. Can AI crawlers (GPTBot, PerplexityBot and the rest) reach your content at all, or are they blocked by a bad robots.txt or content that only renders after heavy JavaScript?
• Understood. Does the AI know who you are? This needs consistent entity signals: structured data (schema.org), an llms.txt file, and matching information across external sources.
• Cited. Does the AI actually name your brand or domain when answering buying questions in your category? This is your Citation Share.
• Quotable. Is your content formatted so the AI can lift it straight into an answer, with direct answers, tables and lists rather than a wall of text? See how to write content AI can extract.

What to do about it

Moving from "winning clicks" to "becoming a source" means changing the metric you treat as success. Before you commission new content, you need to know what AI says about your brand today. Guessing is expensive, and the dark funnel gives you no feedback.

We built a tool that measures this in about 30 seconds, with no registration and no credit card, so you can check your own domain right now. You will see your score on a 0 to 100 scale, learn which competitors AI recommends instead of you, and get ready-made fixes (schema, llms.txt) you can apply immediately.

See also: weekly AI visibility monitoring · Fact Integrity, wrong prices and facts in AI answers.