What Is Ghost Citation & Why Getting Cited Isn’t the Same as Getting Named

A ghost citation occurs when an AI engine like Google AI Overviews (or ChatGPT, Gemini, Perplexity, etc.) links to a website as a source, but never actually names that brand or website in the visible answer text. The reader sees the AI’s answer, but has no idea which brand or source the information actually came from unless they specifically dig into the source links. In other words, there’s a gap between citation and mentions. 

A homeowner in Toronto asks ChatGPT why their garage door won’t close. The answer is clear, specific, and correct: check the photo-eye sensors for misalignment, look for an obstruction, test the limit settings. Underneath it sits a small link to a local garage door company’s repair guide. The company wrote every useful sentence in that answer, but its name appears nowhere. That is a ghost citation. 

That number is the reason Logik Digital treats getting cited and getting named as two separate problems rather than one. They are produced by two different parts of the system; they respond to different work, and a business can be winning the first while losing the second without ever noticing. It is a part of a wider persective of AI SEO in 2026 where businesses are cited for AI overviews.

What the Data Actually Shows

The Semrush and Indig study tagged 3,981 domain appearances across 115 prompts, 14 countries, and four AI engines. Every appearance was marked twice. It was cited if a source link was present, and mentioned if the brand name appeared in the answer text. Those two tags turn out to move almost independently.

It was found that 13.2% is the share of appearances where a business got both the link and the credit. Everything else is partial. Either the AI borrowed your expertise without introducing you, or it recommended you without pointing to your website. There’s an underlying retrieval difference in how AI search engines find and cite content, and this is the same fragmentation showing up in a different place.

Then there is what the citation is actually worth. Pew Research Center, which sells nothing in this market, tracked 68,879 searches from 900 US adults and found users clicked a link inside an AI summary in just 1% of visits. So the typical ghost citation delivers no name and, in practice, almost no traffic either. The work gets used. Nothing comes back. However, information with original research wins AI citations.

Lily Ray’s analysis at Amsive sharpened the point on a different dataset. Across B2B “best software” queries that triggered an AI Overview, Google cited a brand’s own page and then recommended a competitor instead in 69% of cases.

Why It Happens- Because It’s Two Systems, Not One 

The instinctive explanation is that the AI is being stingy with credit. The better explanation is that naming and citing are decided in different places.

Retrieval is the part that goes and finds documents. It cares about topical relevance, whether your page answers the question cleanly, and whether the passage can be lifted out without breaking. Generation is the part that writes the answer, and it draws on what the model already absorbed during training about which brands exist and matter in a category. In simple terms: the model decides who to talk about from memory, then goes looking for evidence to back the answer up.

Seer Interactive tested this across 541,213 responses covering 20 brands and found a signature that fits. When a brand was mentioned, its citation rate was 53.1%. When it was not mentioned, 10.6%. The relationship runs in the opposite direction from the intuitive one: being known makes you more likely to be cited, rather than being cited making you more likely to be known. 

What follows from it is practical enough regardless. If brand selection happens before retrieval, then publishing another twenty articles does not touch the step that decides whether you get named. It only gives the retrieval layer more to borrow.

There is a second cause. If your brand name is not grammatically attached to the fact the AI wants to extract, the AI takes the fact and leaves the name behind. A sentence like “most patients need six to eight sessions” is perfectly extractable and anonymous. That is a writing problem, and it is the cheapest thing on this list to fix.

Why Local Service Businesses Should Care More, Not Less

There is a reasonable objection here. Local businesses do not depend on blog traffic the way publishers do, so who cares about a missing link? The answer is that for local queries the naming problem bites harder than the click problem. When somebody asks an AI assistant who the best physiotherapist in their area is, the assistant names one to three providers before that person visits a single website. 

If you are the ghost, you are not competing for the click. You are outside the shortlist entirely. At Logik Digital, we covered the visibility mechanics for local queries in AI SEO for local businesses, and ghost citations are the sharpest version of that problem.

The Ghost Citation Audit: How Logik Digital Measures Your Brand’s Gap

Most businesses have never checked this, because the tools they use report citations and stop there. The audit below needs no software beyond the AI assistants themselves. Logik Digital runs a version of this at the start of every AI visibility engagement. 

• Write ten to fifteen prompts with no brand name in them

These have to be discovery questions, phrased the way a customer would: “best laser hair removal in Etobicoke,” “who should I call for a broken garage door spring in Toronto,” “do I need a lawyer for a separation agreement in Ontario.” If your own name appears in the prompt, you are testing recall, not discovery, and the result will flatter you.

• Run every prompt across four engines. 

ChatGPT, Google AI Overviews, Gemini, and Perplexity at minimum. Run them logged out where you can, and record the date and model version. Given how differently the engines behave, an average across all four hides the thing you need to see.

• Tag each result on two axes.

Cited means a link to your site appeared. Mentioned means your business name appeared in the text. Every result lands in one of four buckets: both, cited only, mentioned only, or absent.

• Calculate the ghost citation rate.

Divide the number of cited-but-not-named results by your total citations. That single percentage is the number to track, and it is more useful than any composite visibility score, because it tells you which of the two problems you actually have.

• Segment before the act.

Split the results by engine and by funnel stage. A high ghost rate on ChatGPT informational prompts means something different from a high ghost rate on Gemini comparison prompts, and the fixes are not the same.

Re-run the audit monthly. Citation behaviour shifts month to month, and a single snapshot taken on one day is an anecdote rather than a baseline.

How Does Our Three-Layer Technique Fix Ghost Citation

Nothing here is quick, and any agency promising otherwise is selling something. What follows is ordered by how directly it addresses the naming problem rather than by effort.

Layer 1: Make your name inseparable from your claims

This is the layer most businesses can start on today. The principle is simple: make your business the grammatical subject of the sentence carrying the fact, so the model cannot lift the insight without carrying the name along with it. This does not mean stuffing your brand into every paragraph, which reads badly to humans and adds nothing for machines. It means putting the name where the extractable claim is. Our guidance on how to structure content for AI search engines covers the passage-level formatting that sits alongside this.

Layer 2: Build the entity signals machines read

If naming is decided by brand recognition, the practical question becomes how a model builds that recognition. Broadly, it comes from seeing your business described consistently across many independent places. Structured data is how you make that description machine-readable rather than something the model has to infer.

• Organization or Local Business schema with sameAs. The sameAs field is the one that does the work. It explicitly ties your website to your profiles elsewhere, so the model does not have to guess that three similar listings are the same business.
• One canonical name, address, and phone number everywhere. Inconsistent details across directories fragment your identity into several weak entities.  So, apart from content pruning for AI models, the business should focus on NAP consistency for a stronger presence.
• Wikidata where you qualify. Wikidata is a structured, machine-readable database that AI systems consume directly, and it is achievable for many established local businesses.
• Author schema tying named practitioners to the organization. This matters most in healthcare and legal, where the credentialled human is part of what makes the source trustworthy. Also, schema markup helps in AI citations by making the search engine recognise the business entity, which further helps AI bots to extract your website.
•   FAQ schema with the brand name inside the answer text: Not only in the question. The answer is the part that gets extracted.

Layer 3: Earn third-party mentions in recommendation contexts
This is the slowest layer and the one that moves the needle most. Ahrefs, studying 75,000 brands, found branded web mentions correlated with AI Overview visibility at 0.664 against 0.218 for backlinks, roughly three times stronger. Even read conservatively, being talked about by name across credible sources is doing more than link volume. Reviews are the most accessible version of this for a local business. 

The Type of Content That Actually Earns Citations and Mentions

Informational prompts, the “what is” and “how does” questions that most business blogs are built around, produced an 89.3% citation rate and an 18% mention rate. That combination is the ghost citation engine in one line. Comparative prompts, the “best” and “versus” questions, produced a 43.3% mention rate. If your content plan is entirely explainers, you have optimized for exactly the format that borrows your expertise anonymously. However, there can be a consensus gap between the AI tools, as each one of them has a separate system built to provide citations. 

Prompt length pushed the effect further. Short, conversational queries produced mention rates approaching 100%, while long structured prompts produced 2% to 3%. Real customers type short questions, which is the useful half of that finding. Country patterns varied too, with Canada at a 44% mention rate, in the upper-middle of the fourteen markets tested.

The practical adjustment is not to abandon explainers. It is to make sure comparison content, decision guides, and “how to choose” pages exist alongside them, because those are the formats that produce a name.

Inside Logik Digital: What We’ve Seen in Our Own Work

We’re not claiming a fixed playbook. Model behaviour shifts too often for that. But we can point to what happens when the entity layer is built properly.

Laserlicious, an Etobicoke medspa, initially didn’t appear when prospects asked ChatGPT or AI Overviews for the best laser hair removal nearby. After implementing Organization, Local Business, and FAQ schema alongside review and citation building, it started appearing in the AI searches with 4.8 stars. 

Frequently Asked Questions

1. If a ghost citation drives almost no clicks, how would I even know it’s happening on my own site?

Most analytics tools won’t flag it directly, and AI-driven visits often get bucketed as generic traffic rather than attributed to the assistant that sent them. This is part of a broader measurement gap. You can track AI traffic in GA4 to start isolating these visits before you even get to the naming question. 

2. Why did my ghost citation rate look fine last month and worse this month, with no changes on my end?

Your AI visibility changes month to month. It is never static. The same brand can be treated differently by the same engine week to week as models update and retraining happens. This is a distinct phenomenon from the retrieval-versus-generation split covered above.

3. Can two people asking the same AI assistant the same question get different answers about which brand to use?

Increasingly, yes. As AI search personalization results are based on a user’s history and context, the same query can surface different brands for different people. This shifts the naming problem in ways worth understanding.

4. Should I stop publishing informational content altogether if it mostly produces ghost citations?

No. Informational content still earns citations and supports discoverability; it’s just not the format that reliably produces naming. The fix is to add comparison and decision-guide content alongside it, not to abandon explainers. 

5. If my brand isn’t mentioned at all right now, is there any quick way to check where I stand before committing to a full audit?

Running five or six brandless discovery prompts across two engines will give you a rough read in under an hour. It won’t replace the full monthly audit, but it’s a reasonable gut-check while you decide whether to invest in the entity and content work outlined above. 

Find out what your ghost citation rate is

Most businesses discover they are being cited far more than they are being named, and nobody has told them. Logik Digital runs ghost citation audits across ChatGPT, Google AI Overviews, Gemini and Perplexity, segmented by engine and funnel stage, with a prioritized fix list at the end. Connect with Logik Digital, and we’ll audit your brand’s ghost citation status across every major AI platform.