A page being cited by an AI search product does not mean that the answer names the brand behind it. A new Search Engine Land article based on Writesonic data says roughly 40% of cited pages in its sample did not have their brand named in the answer text. The authors call this a “ghost citation.”
The figure is notable, but it should not be treated as a universal industry rate. The dataset comes from a technology vendor’s recent 30-day window, and one author is Writesonic’s founder. An independent Semrush study found a higher 61.7% rate with a different sample. That difference supports the measurement problem while also showing that there is no single stable percentage. The practical response is to separate source visibility, named mentions, and business outcomes before claiming success in AI search.
What does the study report?
The Search Engine Land article describes a dataset of about 16 million brand appearances across Perplexity, Google AI Mode, AI Overviews, ChatGPT, Gemini, Grok, and Copilot. Around 40% of cited pages did not have the associated brand in the answer text, with reported platform rates ranging from roughly 19% to 52%.
This does not mean the AI system found the page unimportant. A model can use a statistic, definition, or explanation and include the source link while omitting the publisher’s name from its narrative. The page earns a citation, but the brand may not gain the same degree of recall.
What is established and what remains uncertain?
The established point is that a citation and a named mention are different events. Ahrefs’ AI visibility documentation likewise treats citations, mentions, impressions, and AI share of voice as separate metrics. A link in a source panel is not the same signal as a brand appearing in the main answer.
The uncertain part is how ghost-citation rates vary by sector, country, language, query type, and product. The Writesonic analysis does not establish cause and effect. It does not prove that adding more brand names to a page will increase mentions. Semrush’s study of 3,981 domain appearances, 115 prompts, 14 countries, and four tools reported 61.7%; that result is not a universal internet benchmark either.
Research on repeated sampling adds another caution. A 2026 preprint argues that AI visibility measurements can change when the same prompt is repeated and that single-run scores may look more precise than they are. A screenshot or one-day crawl should therefore not become the foundation of a strategy.
Three separate signals: source, brand, and outcome
A sound model keeps three lanes apart. The first is citation: does your domain or page appear among the sources? The second is named mention: does the answer explicitly name the brand, product, or a recognised variant? The third is outcome: does that visibility connect to qualified traffic, branded demand, leads, or revenue?
No lane can substitute for another. A citation can support authority and discovery while delivering little direct brand recall. A mention can occur without a link and may add little useful information. Traffic and conversion may not map neatly to either signal because referral and user-level attribution are often incomplete.
Who is affected?
SEO and content teams face the risk of using citation counts as the full success report. Source visibility may improve while named mentions stay flat. Brand teams should not claim every mention either: an AI answer may name the brand critically, comparatively, or out of context.
Performance teams meet the final question: does visibility produce demand? AI products do not always expose complete referral, click, or identity data. Branded search trends, direct traffic, assisted conversions, and qualitative lead research may need to be interpreted together.
What to do now: a repeatable audit
1. Define a stable prompt universe. Keep brand, product, category, problem, comparison, and local intent in separate groups. For Türkiye, record Turkish, English, and common spelling variants.
2. Repeat the same prompts. Sample them on several dates and at different times. Do not report one answer as a persistent rank.
3. Store the product and surface. ChatGPT, Gemini, Perplexity, AI Mode, and AI Overviews can differ in sourcing and presentation. Review each before combining them.
4. Label citations and mentions separately. Record the source URL, brand string, product name, answer context, and visible position in distinct fields.
5. Classify source ownership. Separate owned sites and social accounts from partners, media coverage, and independent third parties. Where the citation lands changes the next action.
6. Connect business signals. Where possible, compare AI referrals, branded demand, landing-page behaviour, lead quality, and source statements in sales conversations. When attribution is incomplete, do not present correlation as causality.
7. Log content interventions. Date changes to company descriptions, expert authorship, product naming, structured data, and source transparency. Compare later movement with the prior baseline.
This gives SEO services, content production, and digital marketing a shared visibility vocabulary.
What not to do and where to wait
Adding the brand name to every paragraph is not the answer. The studies do not prove that repetition causes named mentions, and forced repetition can weaken readability and trust. Start with clear human-facing relationships between the organisation, product, author, and evidence. The page still needs independently useful information worth citing.
Do not interpret one vendor’s score as market share. Tool coverage, prompt set, country, and sampling frequency all change the result. It is sensible to wait for several weeks of repeated data before a major rewrite or budget decision.
There is no need to wait before fixing the measurement dictionary. If citations and mentions currently share one column, the historical dataset will become harder to separate. Start the next reporting period with the three-lane model.
Fark Studio perspective
The ghost-citation discussion shows why AI visibility cannot be compressed into one number. Being selected as a source speaks to content authority, being named speaks to brand memory, and a business result speaks to marketing value. A useful strategy does not confuse those goals.
Fark Studio’s priority is not another proprietary score but a repeatable measurement system. If your brand needs to assess AI search visibility at the evidence, context, and outcome levels, you can plan an AI visibility audit with Fark Studio.
Sources
Search Engine Land, Ghost citations: Why AI search cites your content, not your brand, July 29, 2026.
Semrush, The Ghost Citations Study, June 2026.
Ahrefs Help Center, AI Visibility metrics, updated June 26, 2026.
arXiv, Quantifying Uncertainty in AI Visibility: An Empirical Analysis of Measurement Variance in Generative Search, March 9, 2026.
