Around 40% of AI citations do not name the brand being cited, according to an analysis of roughly 16 million brand appearances published on 29 July 2026 by Search Engine Land. The authors, Nikki Lam and Samanyou Garg, call this a ghost citation: the engine links your page as a source but never says your brand name in the answer a reader sees. Rates varied widely by engine, from 19% on Microsoft Copilot to 52% on Perplexity.
The data comes from Writesonic, and one author runs the company
The article carries an explicit disclosure that co-author Samanyou Garg is the founder and CEO of Writesonic, the AI platform whose data the piece reports and which sells AI-visibility tracking. The dataset is a recent 30-day window of AI answer data across thousands of brands, and the ghost-citation analysis covered approximately 16 million brand appearances. Nobody outside Writesonic can check the sample, and the conclusion the piece reaches happens to be that brands need the kind of measurement Writesonic sells. The finding is still worth knowing, and both facts belong in the same sentence when you repeat it.
A ghost citation is a link without a name
Writesonic defines a ghost citation as an AI engine linking to your page as a source while never mentioning your brand name in the answer itself. The distinction the authors draw is between contribution and visibility: a citation confirms that your content fed the answer, but it does not mean a reader came away knowing who you are. That splits one number people currently treat as a single score. A brand can be feeding answers all day and still be invisible in them.
Ghost-citation rates ran from 19% to 52% depending on the engine
The engine-level spread is the most useful part of the study: the same content can be credited by name on one platform and anonymised on another. These are the rates Writesonic reported for each engine, against an overall rate of around 40%.
| AI engine | Citations that did not name the source brand |
|---|---|
| Perplexity | 52% (highest in the study) |
| Google AI Mode | 49% |
| Google AI Overviews | 41% |
| ChatGPT | 37% |
| Gemini | 25% |
| Grok | 22% |
| Microsoft Copilot | 19% (lowest in the study) |
The two highest rates sit on Perplexity and Google AI Mode, which are also the two surfaces most often held up as the future of discovery. If you are being cited there, the reported odds are close to even that your name is not in the sentence.
On cause, the authors say the field does not really know
The piece says ghost citations are likely a by-product of how different engines synthesise sources rather than a deliberate decision to omit brand names, and it states outright that nobody really knows why the rates differ. That hedge is worth keeping when you repeat the finding. Nothing in the study identifies a rule, a policy or a ranking factor behind the gap, so treating a high ghost rate as a penalty against your site would be an invention. It is an observed pattern with an unproven cause.
The proposed fix is to write your name into the claim
Writesonic's recommendation is to build the brand name into the claim itself, so a summariser cannot strip it out without breaking the sentence. Their worked example uses their own study: write "Writesonic's Ghost Citation Study found that around 40% of AI citations didn't name the source brand" rather than the nameless version that begins "Our analysis found...". The nameless version is trivially easy for an engine to anonymise, because the finding does not carry a name. The first version cannot be quoted without the name coming along.
The rest of their advice follows from that. Build named proprietary research, datasets and frameworks so the brand and the fact are the same object. Place the company name directly next to key data points rather than in a byline or a footer. This is a writing decision more than a technical one, which is why it sits at the awkward border between content work and generative engine optimisation.
Track mentions and citations as two separate metrics
The study's measurement recommendation is to stop reporting AI visibility as one number and track mentions and citations separately. Citations tell you your content is being used. Mentions tell you readers are learning your name. A brand with strong citations and weak mentions has a writing problem, not a crawling problem, and only the split view shows it. The authors also advise reading the individual answers behind a low mention rate, because the answer text is where you can see which phrasing got kept and which got dissolved.
What this means for marketers in Thailand
The study reports no country or language split, so the engine-level percentages cannot be read as Thai-market figures. What does transfer is the diagnostic: if an AI-visibility dashboard shows your site as a cited source while nobody arriving on a sales call has heard of you, ghost citations are a plausible explanation rather than a mystery.
For Thai brands the naming tactic is cheap to test. Name your own data, so a statistic reads as your company's finding rather than as a floating number, and keep the brand name in the same sentence as the figure in both your Thai and English pages. Whether Thai-language answers show the same 19% to 52% spread is speculation, and should be treated as speculation until someone measures it. Teams already tracking ChatGPT visibility or working on AI SEO can add the mention-versus-citation split to reporting without new tooling, by reading the answers themselves.
Frequently asked questions
What is a ghost citation?
A ghost citation is when an AI engine links to your page as a source but never mentions your brand name in the answer. The term comes from Writesonic's analysis published on Search Engine Land on 29 July 2026, which found around 40% of citations worked this way.
Which AI engine hides brand names most often?
Perplexity, at 52% of citations not naming the source brand, was the highest in the study. Google AI Mode was second at 49%, and Microsoft Copilot was lowest at 19%.
Is this independent research?
No. The data is Writesonic's own, and the article discloses that co-author Samanyou Garg is the founder and CEO of Writesonic, which sells AI-visibility tracking. The sample cannot be audited from outside the company.
Why do engines leave the brand name out?
The authors say it is likely a by-product of how each engine synthesises its sources, and they state that the field does not really know. The study identifies no policy or rule behind it, so any confident explanation goes beyond the evidence.
Does the study cover Thailand or Thai-language answers?
No. The piece reports a 30-day window across thousands of brands and roughly 16 million brand appearances without breaking results out by country or language, so there is no Thai figure in it.
If your reporting shows AI citations but no lift in people knowing your name, the gap is measurable. Our GEO work separates the two and looks at the answer text itself.
