Back to Blog
AI Search15 min read15 October 2026

Ghost Citations: Why AI Cites Your Page but Doesn't Name You

AI answers often link to your page without naming your business. What the studies found, how to spot ghost citations, and how to turn them into mentions.

By AI Schema Gen Team

You check what ChatGPT says about a question in your field. Your page is right there in the list of sources. You read the answer itself, and your business isn't in it. The AI used your work, gave the reader your facts, and never said who they came from.

That's a ghost citation: an AI answer that links to your page as a source but doesn't name your business in the text people actually read. The link is there for anyone who opens the sources panel. Most people don't.

It's a new name for something that has been happening since AI answers started showing sources, and in the last few months it has started being measured. This guide covers what the numbers say (they disagree, and the disagreement is the useful part), how to find out whether it's happening to you, why an AI drops a name it clearly had access to, and what you can change on your own pages and beyond.

What the studies found, and why they don't agree

Three studies published in 2026 looked at this directly. Their headline numbers are far apart.

Semrush, with growth advisor Kevin Indig (June 2026). Across 3,981 appearances of websites in AI answers from ChatGPT, Gemini, Google AI Overviews and Google AI Mode, the Semrush study found 61.7% were ghost citations: cited as a source, not named in the answer. Only 13.2% were both cited and named.

Writesonic, reported in Search Engine Land (July 2026). Across roughly 16 million brand appearances, about 40% of citations didn't name the source brand. The rate ran from 52% on Perplexity down to 19% on Microsoft Copilot, according to Search Engine Land's write-up, which notes that one of its co-authors runs Writesonic.

Orbit Media (September 2026). Tracking 72 prompts a week for three business websites across ChatGPT, Claude, Perplexity and Gemini, this study found almost the opposite. Its write-up on LLM citation sources reports that out of 1,052 query runs, "only 17 cited a brand's page without also mentioning the brand."

So is it 62%, 40%, or under 2%? All three can be right, because they measured different things.

Semrush counted every website that appeared in an answer, including reference sites and publishing platforms. Its own examples show the pattern: medium.com was cited 16 times and never named once, while Google was named less than three times for every hundred times its pages were cited. Sites like these are used as raw material. Nobody asking a question needs to know which blogging platform hosted the answer.

Orbit tracked questions chosen for three specific businesses, in their own field, where those businesses were a natural part of the answer. In that setting, an AI that found the page useful almost always named the company too.

The question type matters as much as the site. In the Semrush data, answers to informational questions ("what is", "how does") named the cited site only 18% of the time. Comparison questions named it 43.3% of the time, and how-to questions 42.8%. When the answer is about choosing between options, the AI has to say what the options are. When the answer is a general explanation, it rewrites the facts in its own words and the source becomes a footnote.

That gives you the first practical point. Ghost citations aren't spread evenly across your site. They cluster on your general information pages and on questions where your business isn't the subject. Which pages and which questions is something you can find out.

Does it actually cost you anything?

Not always. A ghost citation still means an AI engine read your page, trusted it enough to use it, and linked to it. That's better than not being used at all, and some readers do open the sources.

Whether it matters depends on what the page was for.

It costs little when the page answers a broad question that has nothing to do with choosing a provider. A plumbing company's guide to finding a stopcock is useful, and being the source behind an AI's answer about stopcocks is fine. Nobody reading that answer was about to pick a plumber.

It costs a lot when the page carries something only you have. Original research, your own method, a price comparison you put together, a local fact about your area that you documented first. These are the pages built to make your name known. If the AI passes on the finding and drops the name, the work earns credit for nobody, and the next business to repeat your numbers gets the same treatment you do.

It costs the most when the question is about choosing, and you're cited but someone else is named. Your page supplies the information, and a competitor's name appears in the recommendation. If you see this pattern, it's the one to fix first.

How to find your own ghost citations

You need two things written down for every answer you check: whether your page was in the sources, and whether your name was in the text. Many dashboards lead with one blended visibility score, which hides the problem completely. Our guide to AI visibility metrics explains why mentions and citations are two separate events and how each tool counts them.

A simple way to check by hand:

  1. Pick 15 to 20 questions and mix three kinds: general questions in your field, questions about choosing a provider like you, and questions about something specific you published, such as a study or guide.
  2. Ask them in two or three AI engines, not just one. The Writesonic data shows engines behave very differently: Perplexity and Google's AI answers link often but name less, while Gemini and Copilot name more and link less.
  3. Sort each answer into one of four groups: cited and named, cited but not named (the ghost citations), named but not cited, or neither.
  4. Note which of your pages was cited. This is the step most people skip, and it's the one that tells you what to fix.

Because AI answers change from one ask to the next, one run isn't enough. Our guide to tracking ChatGPT mentions covers how many asks it takes before a pattern is real.

After a round of this you'll usually see a pattern: the same handful of pages ghosted again and again, often your most useful guides or your research. That's your list.

Why an AI drops a name it had

Nobody outside the AI companies can see exactly how an answer gets written, and Search Engine Land's write-up says plainly that the field doesn't have a complete explanation yet. But the way these systems work points to a few causes you can actually do something about.

The common thread: before an AI can name the source, it has to know whose page this is, and the answer has to be worth including. A ghost citation usually means one of those two broke.

The answer is rewritten, and the name wasn't attached to the fact

AI engines don't quote pages. They read several sources and write a new answer in their own words. Facts survive that rewrite. A name only survives if it's tied to the fact closely enough to travel with it.

Compare two sentences on a page:

  • "Our survey of 400 homeowners found that 62% had never tested their smoke alarms."
  • "Maple and Finch's 2026 Home Safety Survey of 400 homeowners found that 62% had never tested their smoke alarms."

The first gives the AI a fact and a pronoun. When the answer is rewritten, "our" means nothing and gets dropped. The second makes the name part of the fact. To repeat the finding accurately, the AI has to repeat who found it. Search Engine Land's write-up makes the same suggestion: put the name in the sentence with the claim, not a few paragraphs away.

The page doesn't make clear who owns it

This is the part most ghost citation advice skips, and it's the one entirely under your control. For an AI to name you, it has to connect the page it read to a business with a name. On many sites, that connection is weaker than it looks.

Common ways it breaks:

  • Three versions of the name. The logo says "Maple & Finch", the footer says "MF Home Services Ltd", and the page title says "Home Safety Tips | M&F". A person sees one business. A machine sees three possible names and no clear winner.
  • No owner stated on the page itself. The article has no author, no date and nothing saying which organisation published it. The business name only appears in an image of the logo, which carries no readable text.
  • Structured data that doesn't connect. The site describes its business in structured data on the homepage, but the article pages don't say they were published by that business, or point to a different version of it. Our guide to the Understand pillar explains how these descriptions link together, and the four ways the links break.

None of this is proven to cause ghost citations on its own, and no study has tested it directly. The reasoning is simpler than proof: an AI can't name a business it couldn't identify. Making ownership unmistakable doesn't guarantee a mention. It removes one reason for leaving you out.

The name reads like a description

Some business names are just descriptions: "Best Local Plumbing", "Premier Garden Design", "The Accounting Team". When an AI rewrites, a name like this is easy to turn into a phrase ("a local plumbing company says"), because it looks like one.

You may not want to rename your business, and you don't have to. But if your name is descriptive, consistency matters even more. Use it the same way everywhere, alongside the place or a distinctive detail ("Premier Garden Design of Bath"), so it reads as a specific business and not a category.

You're the reference, not the subject

The Semrush data shows reference sites and platforms get cited without being named far more than consumer brands do. Kevin Indig's summary of strong brands, in the Semrush findings, is that "the brand name carries on its own."

If most of your cited pages answer general questions, you're being used as reference material. That's not a problem with your pages. It's a sign that the questions you're appearing for aren't ones where a business gets recommended. The fix there isn't on the page, it's in what you publish next.

The AI hasn't seen your name enough

AI engines name businesses they recognise. The Semrush study links mentions to how consistently a brand appears across the web, through press coverage, reviews, community discussion and other people's articles. A business that only exists on its own website is a name the AI met once.

How to turn ghost citations into mentions

There's no setting that makes an AI name you. These are the changes that remove the reasons it doesn't, roughly in the order they pay off.

1. Rewrite your most-ghosted pages so the name travels with the facts

Take the pages from your check above and look at every key fact, number or finding. Wherever the page says "we found", "our data shows" or "this study", write the business name into that sentence instead. Give your research and methods names that include yours, such as "the Maple and Finch Home Safety Survey" rather than "our survey". A named study is hard to repeat without its name.

Don't overdo it. A page that crams the business name into every sentence reads badly to people, and AI engines are built to prefer pages people find useful. Name the source of each claim that matters; leave the rest alone.

2. Make ownership unmistakable on every page

  • Use one version of your business name everywhere: logo text, page titles, footer, about page, social profiles and directory listings. Pick it, write it down and fix every place that differs.
  • Give articles a visible author and a publisher, so the page itself says which business it belongs to, in words and not just a logo image.
  • Describe your business in structured data once, with a stable identifier, and make every article say it was published by that same business. On WordPress or any other site, AI Schema Gen's Business Identity settings do this for you: the business is described once, and each page's publisher points back to it. Our guide to entity grounding explains why a confirmed identity beats an AI's best guess.

This is groundwork rather than a quick win, but it serves every page on the site at once, including the ones you haven't checked yet.

3. Publish things where naming you is part of the answer

The Semrush numbers show the AI names sources far more often when the question is about comparing or choosing. You can't change which questions people ask, but you can make sure some of your pages answer the comparing and choosing kind: honest comparisons of options in your field, pricing explained, "how to choose" guides that state who you're for and who you're not for. On those pages, being named is the natural result of being useful.

Original data works the same way. A finding that only exists because you produced it is much harder to repeat without saying where it came from, as long as the page makes that clear.

4. Get your name repeated somewhere other than your own site

Mentions follow recognition. Reviews, local press, trade association listings, podcast appearances, guest articles and community answers all put your name next to your topic in places AI engines read. You don't need a PR campaign. You need your business name to appear, spelled the same way, in a handful of places that aren't your website.

5. Re-check the same questions

Run the same check again after a month or two and compare the four groups, looking for ghost citations moving into "cited and named" across several questions, not one answer changing.

Common mistakes

Counting citations as wins without reading the answer. A citation report that looks healthy can be mostly ghost citations. Read the answers, at least for the questions that matter most.

Treating all ghost citations as a problem. Being the uncredited source of a general explanation is mostly harmless. Spend your effort on the pages that carry your research, your methods and your comparisons.

Checking one engine and generalising. Engines split into ones that link a lot and ones that name a lot. A clean result on one says little about the others.

Writing "we" and "our" around every key finding. It reads naturally to people and loses the name the moment the text is rewritten.

Letting the business name drift. Every variation (abbreviations, old names, a different legal name in the footer) makes it harder for a machine to treat the page as belonging to one clear business.

Stuffing the name everywhere. Repeating your business name in every sentence makes pages worse for readers and does nothing a few well-placed mentions don't already do.

FAQ

What is a ghost citation?

It's when an AI answer lists your page as a source but doesn't mention your business name in the answer itself. The term came into wide use in 2026, through a Semrush study with Kevin Indig in June and a Writesonic study reported by Search Engine Land in July. The reader gets your information without being told it came from you, unless they open the list of sources.

How common are ghost citations?

It depends heavily on what's being measured. A Semrush study found 61.7% of website appearances in AI answers were ghost citations, Writesonic's data put it near 40%, and an Orbit Media study of three businesses' own topics found only 17 cases in 1,052 query runs. They're common for reference content and general questions, and much rarer when a question is about choosing a business like yours.

Which AI engine has the most ghost citations?

In the Writesonic data, Perplexity had the highest rate at 52%, followed by Google AI Mode at 49% and Google AI Overviews at 41%. ChatGPT was at 37%, Gemini 25%, Grok 22% and Microsoft Copilot 19%. Other studies measured the engines differently, so treat these as a rough guide, not fixed rates.

Is a ghost citation better than no citation at all?

Usually, yes. It means an AI engine read your page, trusted it and linked to it, and some readers do open the sources. It's a missed chance to be remembered, not a penalty.

Can structured data make AI name my business?

No study has shown that it does, so we won't claim it. What structured data does is state clearly which business owns and published a page, which removes one reason an AI might fail to name you. Treat it as making identification easy, not as a switch that produces mentions.

Does AI Schema Gen detect ghost citations?

Not directly yet. AI Visibility tracking records whether your site is cited in ChatGPT's answers to the questions you track, and which other sites are. It doesn't yet check whether your name appears in the answer text, so finding ghost citations still means reading the answers.

How long does it take to fix?

Rewriting pages takes an afternoon. Seeing the effect takes longer, because AI engines need to read the updated pages again and their answers vary from one ask to the next. Re-check after a month or two, across several questions.

Start with whether AI can tell it's you

Ghost citations begin with a simple failure: an AI used your page without connecting it to a named business. Our free AI readiness check shows whether AI can tell who runs your site: whether your business name, details and structured data line up, and whether your pages say who published them. Fix that first, and every other step in this guide has something to build on.

Is your website ready for AI?

Check now. Free readiness score in under a minute, no signup, no card.