Your GEO Vendor's Citation Count Doesn't Prove You Are Named
A rising citation count from your GEO vendor can mean nothing changed for the buyer reading the answer. Here is the per-domain, per-engine breakdown to demand before you trust the number.
Your GEO Vendor's Citation Count Doesn't Prove You Are Named
A rising citation count from your GEO vendor can mean nothing changed for the buyer reading the answer. Here is the per-domain, per-engine breakdown to demand before you trust the number.
This is not a "why aren't I named" piece. AuthorityTech has already covered that question twice, once from ghost citations as a brand-presence problem and once from third-party recognition-versus-mention data. Both explain the mechanism and prescribe the fix: entity clarity, earned corroboration, third-party volume.
The question here is different, and it comes before that one. Before you spend a quarter chasing entity clarity, ask whether the metric your vendor reports is even measuring the thing you think it is measuring. Most GEO vendor dashboards show one number: a citation count, or a citation rate, climbing month over month. That number can rise while the rate that actually matters — how often the answer's own words say your name — stays flat or falls. If your vendor's report does not separate those two events, you cannot tell which one you are paying for.
The number vendors show you, and the number they usually don't
A citation, in most vendor dashboards, means your domain showed up in the source layer of an AI answer: a link, a footnote marker, an entry in a "sources" panel. A naming event is different. It means the words of the answer itself said your brand's name, in the sentence a buyer actually reads.
These are not the same event, and the gap between them is large. Machine Relations Index research published on Paralax measured both outcomes across 14,447 readable citations from four AI engines over a 139-day window: Gemini, ChatGPT, Claude, and Perplexity. Only 8.6% of those citations put the source's own name into the answer text. The other 91.4% were silent — the engine used the page and linked it in a structure the reader has to click open, without ever writing whose page it was.
That 8.6% is the number most vendor reports never surface, because most vendor reports collapse citation and naming into one metric. If your monthly report shows "citations: 340, up 12%," ask what fraction of those 340 put your name in the sentence. If the vendor cannot answer, the number you have been tracking is a supply metric, not a visibility metric.
Naming rate is not one number. It splits by what you are and which engine answered.
The MRI measurement did not stop at the pooled 8.6%. It broke the same outcome down two ways, and both breakdowns change what a buyer should ask for.
By source type, the range runs from 28.6% down to zero, on domains cited 100 or more times:
| Domain | Citations read | Named in the answer text |
|---|---|---|
| databricks.com | 105 | 28.6% |
| github.com | 142 | 21.8% |
| hubspot.com | 122 | 21.3% |
| crunchbase.com | 122 | 18.0% |
| deloitte.com | 110 | 13.6% |
| gartner.com | 270 | 11.9% |
| zapier.com | 116 | 11.2% |
| techtarget.com | 205 | 0.5% |
| techradar.com | 454 | 0.7% |
| arxiv.org | 405 | 0.0% |
| nytimes.com | 132 | 0.0% |
| prnewswire.com | 125 | 0.0% |
The pattern is not about source quality. Databricks, GitHub, HubSpot, and Zapier are things a buyer might choose, so when an engine cites their page it is usually recommending the product, and the name has to be in the sentence for the recommendation to make sense. Deloitte and Gartner get named because an answer citing them is reporting what they said. Arxiv, the New York Times, and PR Newswire are strong sources cited hundreds of times each, and they still reach the prose at or near zero — not because the evidence is weak, but because the job those citations do (backing a claim) does not require the engine to say who backed it.
If your brand's category role in a given query is "evidence for someone else's answer" rather than "the thing being recommended," your naming rate has a structural ceiling no amount of content work will move. That is a targeting question for which queries you chase, not a content-quality question.
By engine, the same measurement found a 2.2x spread on identical prompts and the same window: Gemini named a cited source 13.2% of the time, ChatGPT 10.5%, Claude 10.1%, Perplexity 5.9%. Perplexity's low rate holds even after controlling for how many sources it cites per answer — it stays near 5–8% across every citation-count band, while the other three engines climb as they cite more. The mechanical reason is attribution style: Perplexity credits with a numbered marker per sentence, which functions as a footnote rather than a name in the clause.
One line on what does not explain this gap: page position within the answer has already been settled elsewhere as not the driver of citation outcomes, and it is not the driver of naming rate either — this is a property of source type and engine attribution mechanics, not of where a link sits on a rendered page.
If your GEO vendor reports one blended number across engines, and your buyer mix skews toward Perplexity, you are being told a portfolio average that understates your actual naming problem on the engine where you are weakest — or overstates it, if your mix skews toward Gemini.
What to put in your next vendor RFP or QBR
A vendor citation count, taken alone, tells you that your content entered the pool an engine drew from. It does not tell you whether the answer named you, and it does not tell you which part of your portfolio — which source category, which engine — is producing the naming events your number is built from. Ask for three things before you renew a contract or approve a budget increase on the strength of a rising citation count:
- A naming rate alongside the citation rate, not instead of it. The vendor should report both numbers for every period, with the method for detecting a name in the answer text stated plainly — including whether link anchors and bare URLs are stripped before counting, which changes the result materially.
- A breakdown by source role or category, not a single blended figure. If your brand functions as a subject-of-recommendation source on some queries and an evidence source on others, the two get named at different rates and need different plans.
- A breakdown by engine, not a pooled average. A 2x spread between the best and worst engine in this data means a pooled number can hide which engine is actually broken for you.
A vendor that can produce all three is measuring the thing you are paying for. A vendor whose dashboard only goes up to "citation count" is measuring the thing that is easiest to count, which is not always the thing that matters. Watch for the failure mode directly: a citation count that keeps climbing while the naming rate — the one number the dashboard doesn't show — stays flat. That is not progress. That is a vendor optimizing the metric they report instead of the outcome you bought.
FAQ
Is a citation count from a GEO vendor a useless metric?
No. It measures something real: whether your content entered the pool an engine drew from to build an answer. The failure is treating it as a stand-in for whether the buyer reading that answer saw your name. Track both, separately.
How is this different from the ghost citation content already published on this site?
Ghost citation coverage explains why citations often arrive without a name attached and prescribes fixes like entity clarity and earned corroboration. This piece is upstream of that: it is about whether your vendor's own reporting can even show you the gap, broken down by source role and by engine, before you spend budget closing it.
Should I expect the same naming rate on every AI engine?
No. Published Machine Relations Index research found a 2.2x spread between the highest and lowest of four measured engines on identical prompts in the same window. A single blended score across engines will not tell you which engine is dragging your average down.
Does where a citation appears in the answer affect whether the brand gets named?
That has already been examined and ruled out as the driver in prior research; the naming gap in this data tracks what kind of source a domain is and which engine produced the answer, not position on the page.
What should I ask a GEO vendor before signing or renewing?
Ask for a naming rate reported alongside the citation rate, broken out by source role or category and by engine, with the method for detecting a name in the answer text stated in plain language. If a vendor can only report a single blended citation number, treat that as a gap in their measurement, not proof of your visibility.
Sources and method
The naming-rate figures above come from Machine Relations Index research published on Paralax, covering 14,447 readable citations across four AI engines (Gemini, ChatGPT, Claude, Perplexity) over a 139-day window ending September 26, 2026, drawn from Machine Relations Index release mri_score_v2.0+2026-09-26+2b779408cfda. The full per-domain table, per-engine breakdown, and methodology and limits are published at the source link and at the Machine Relations Index. Platform citation and attribution mechanics referenced here are documented directly by each provider: OpenAI's citation formatting guide, Anthropic's web search tool documentation, Perplexity's crawler and citation documentation, and Google's AI features documentation.