New research from Semrush introduces a category most marketers haven't measured: the ghost citation. In 61.7% of cases where AI platforms sourced a page, the brand received a citation link but no mention of its name in the actual answer. Only 13.2% of citations include both a source link and a brand mention — the outcome most teams are actually optimizing for.

The platform breakdown reveals why this happens: ChatGPT cites sources in 87% of responses but mentions brands in only 20.7% — a 4x gap between being sourced and being recognized. Gemini flips the pattern: it mentions brands in 83.7% of cases but cites sources in just 21.4%. The type of query matters as much as the platform: short conversational queries generate 30–50x more brand mentions than long structured prompts, and comparative queries produce 2.4x more mentions than informational ones.

llms.txt is a structured file that tells AI crawlers which content is authoritative, how your site is organized, and what should be cited. OpenAI, Anthropic, Stripe, Cloudflare, Vercel, and Mastercard have all implemented it. With 41% of users now relying on AI tools for information discovery, the file is becoming a baseline crawlability requirement — not an optional extra.

"The ghost citation finding reframes the whole measurement problem. Most GEO teams track whether their domain appears in AI citations — but appearing as a source link is completely different from having your brand named in the answer. You can be sourced by ChatGPT in 87% of relevant responses and still have no brand recognition benefit if only 20% include a name mention. The implication: short conversational queries and comparison-framed content are the formats that actually convert citations into brand mentions. That's where optimization time should go."