Writesonic, an AI content and visibility-tracking platform, analyzed roughly 16 million brand appearances across seven AI engines and found that about 40 percent of citations never named the source brand in the generated answer. The rate reached 52 percent on Perplexity, the highest of any engine Writesonic measured, according to data the company published in a Search Engine Land analysis on Wednesday.

For any team treating citation counts as a proxy for AI visibility, the finding reframes what that number actually measures. A citation confirms an engine pulled from a page as a source. It does not confirm a reader ever saw the brand behind it.

One disclosure needs to be stated plainly, not buried at the end: one of the two co-authors, Samanyou Garg, runs Writesonic as its founder and chief executive, and Writesonic is the company that produced the underlying data. A disclosed financial interest does not make a finding wrong. Credit is also due here, since the disclosure runs inside the article’s body rather than in a footnote or bio line. Still, readers weighing the 40 percent figure should know it originates from a vendor whose own executive co-wrote the article presenting it, not from an independent third party.

The core term is what Writesonic calls a ghost citation: an AI engine links to a brand’s page as a source but never names the brand in the answer text itself. Writesonic treats a citation, a source link the engine includes, and a mention, the brand’s name appearing in the generated answer, as two separate outcomes. The data show those outcomes diverge often enough that collapsing them into one metric hides real variation.

Ghost citation rates ranged from 19 percent to 52 percent depending on the engine, per Writesonic:

Writesonic groups the engines into two rough behaviors. Gemini and Microsoft Copilot act as namers: they include the brand in the answer text more consistently but link out to source pages less often. Perplexity and Google’s answer engines act as citers: links out are plentiful, but whose page it is often goes unsaid. ChatGPT and Grok sit somewhere in the middle, and Writesonic notes that neither pattern is the ideal outcome on its own.

The practical instruction is to measure mention rate and citation rate as two separate metrics, then segment the results by engine rather than reporting one blended score. An aggregate citation number can hide a much worse mention rate on a single platform. A brand cited at a similar volume across engines could still be named in the answer only half as often on Perplexity as on Copilot, and a single combined figure would never surface that gap. Essentially every AI visibility dashboard on the market currently reports citation count without this split, per Writesonic’s framing of the problem.

Writesonic’s most actionable recommendation is to build named, proprietary research: a study, index, or framework that carries the brand’s name in its title rather than an unbranded statistic. Because the finding references the brand by definition, that kind of asset is harder for an engine to ghost than a generic tip whose author is identified only somewhere further down the page.

Search teams reporting AI citation counts as a visibility win should ask whether those counts come with a mention rate broken out by engine. Without that split, a strong citation number on Perplexity or Google AI Mode could be masking a brand that AI answers rarely say out loud.

Search Engine Land published this analysis, co-authored by Nikki Lam and Samanyou Garg, on July 29, 2026.