AI Citation Statistics 2026: How Often AI Names Your Brand

Across four AI platforms, brands get cited far more often than they get named. In a study of 3,981 brand appearances, 74.9% included a citation but only 38.3% named the brand (Semrush). That gap is the whole story of AI search in 2026, and it is why counting your citations has become its own job.

Here is the one line worth stealing: being cited means an AI used your page, being mentioned means it wrote your name, and most of the time you get one without the other.

This page collects the numbers worth knowing, where each comes from, and what they mean for whether AI actually recommends you. Every figure links to its source. We keep it current and stamp the date at the top, because a stats page that looks stale is a stats page nobody quotes.

Key takeaways

• Citation rate nearly doubles mention rate. 74.9% of brand appearances included a citation, but only 38.3% included a brand mention (Semrush, 3,981 appearances).
• Most citations are invisible to the reader. 62% of AI citations are ghost citations, meaning your page is used as a source with no brand name shown (Semrush).
• Half of brands never show up at all. 53% of brands were invisible on all four major AI platforms when asked their own buyer questions (Boring Marketing, 13,475 checks).
• The base citation rate is low. Across those buyer questions, the overall citation rate was 13.9% (Boring Marketing).
• Engines retrieve far more than they use. 85% of the pages an AI pulls never make it into the final answer (AirOps).
• A few domains own the citations. Wikipedia alone accounted for 15.2% of citations among the 25 most cited domains, and government sites made up 33% of that group (academic GEO study).

That block is the part other people will quote. Keep it near the top of your own page too, because engines lift what sits early.

What counts as an AI citation

An AI citation is when a model uses your page as a source for an answer. It usually shows up as a link, a footnote, or a source card next to the text. A brand mention is different. That is when the model writes your name into the answer itself, like "according to Acme" or "Acme recommends."

You can have one without the other, and you usually do. This is the single most misunderstood thing about AI visibility, so it is worth slowing down on.

Picture someone asking ChatGPT for the best project management tool for a small team. The answer names three tools in the text and lists nine links underneath as sources. If you are one of the three names, you won the mention. If you are only one of the nine links, you won a citation but the reader may never notice you. Your name did not make it into the sentence they actually read.

Semrush gave this its own name. A ghost citation is a source link with no brand mention attached. In their data, the split looked like this:

• 61.7% of appearances were ghost citations only, cited but never named
• 25.1% were brand mentions with no citation, named but not linked
• 13.2% were both cited and mentioned, the full win

So the complete outcome, where you are both named and linked, happened in about one appearance out of eight. That is the bar. Everything else is partial credit.

How often does AI cite brands in 2026

The honest base rate is low. Boring Marketing ran 13,475 buyer question checks across four platforms and found an overall citation rate of 13.9% (Boring Marketing). More striking, 53% of brands were invisible on all four platforms for the exact questions their own buyers ask.

Read that twice. More than half of brands never appeared, on any major engine, for their own category questions. Not ranked low. Absent.

This is the number that surprises founders the most. You can have a strong site, real traffic, and a good Google position, and still be missing from the answer an AI gives when someone asks what tool to buy. The old scoreboard and the new one are not the same game, and a good score on one tells you little about the other.

AI citation rates by platform

Different engines cite at very different rates, and the word "rate" means different things depending on what a study measured. This trips people up, so here are both lenses, kept separate on purpose.

Lens one is how often a brand gets cited at all for a buyer question. Boring Marketing measured this across roughly 3,399 checks per platform:

Source: Boring Marketing.

Lens two is the style each engine uses once a brand does appear. Semrush measured that, and the two leaders flip:

• Gemini named the brand 83.7% of the time but linked a source only 21.4% of the time. It talks like a person, dropping names into the sentence without footnotes.
• ChatGPT did the opposite, linking a source 87% of the time but naming the brand only 20.7% of the time. It writes like an academic, heavy on footnotes, light on names (Semrush).

Google AI Mode named brands at nearly twice ChatGPT's rate. And the engines did not even agree with each other. In identical scenarios, they disagreed on whether to name a brand 22% of the time.

The lesson is not to memorize one magic number. It is that "citation rate" describes different things in different studies, and neither describes your brand. The only rate that describes you is the one you measure on your own prompts, which we get to below.

The citation gap: cited but not named

If 62% of your AI citations never show your name, your analytics and your brand recognition tell two different stories. Your server logs might show AI crawlers pulling your pages. Your referral traffic might tick up. And the person reading the answer still has no idea you exist, because the model used your work without saying your name.

This is the gap the whole category is waking up to. For years the goal was a link, because a link meant a click and a click meant a visit. In an AI answer, the click is optional and often skipped. The answer is right there. So a citation with no mention buys you a little trust signal with the model and almost nothing with the human.

The wording of the question changes this dramatically. Semrush found that short conversational queries, around 16 characters, produced mention rates near 100%. Long structured prompts, around 98 characters, dropped mention rates to 2 or 3%. That is a 30 to 50 times difference driven by nothing but how the question was typed (Semrush).

Content type matters too. Informational questions pulled an 89.3% citation rate but only an 18% mention rate. Comparative questions, the "X versus Y" and "best tool for" kind, pushed the mention rate to 43.3%, roughly 2.4 times higher. If you sell something, the comparison questions are where your name has the best shot of landing in the sentence.

Why your analytics undercount AI citations

Here is the trap. You open Google Analytics, you see a trickle of referral traffic from ChatGPT and Perplexity, and you conclude AI is a rounding error. That conclusion is wrong, and the reason is the same gap this whole page is about.

Analytics only counts the clicks. A citation with no click leaves no trace in your traffic reports, and a mention with no link leaves no trace at all. When an AI names you in a sentence and the reader nods and moves on, that happened. It shaped a buying decision. And your dashboard shows nothing, because nobody clicked anything.

So the brands that judge AI by referral traffic are measuring the smallest, least important slice of what is going on. They see the clicks and miss the influence. The influence is the part that decides whether a buyer arrives already thinking your name, or arrives thinking a competitor's.

This is why AI visibility needs its own measurement, separate from your analytics. You are not counting sessions. You are counting how often the machine that increasingly sits between you and your buyer decides to say your name. That number does not live in Google Analytics. It lives in the answers themselves, which is why you have to go read the answers, at scale and on a schedule, to know it.

AI citation rates by industry

Visibility is not spread evenly. Some categories are close to silent.

Finance and insurance had the lowest citation rate in Boring Marketing's data at 2.1%. In legal and professional services, 86% of brands were never cited at all (Boring Marketing).

There is a pattern here. The more regulated and the more the answer carries risk, the more engines lean on a small set of sources they consider safe, and the harder it is for an individual brand to break in. If you sell in one of these categories, assume the default is silence and that earning a citation takes more proof, not less. A confident claim on your own page is not enough. You need third party corroboration the model already trusts.

What AI actually cites

When engines look for sources, they reach for a short list of trusted places. Boring Marketing found that across the brands they audited, Reddit appeared in AI answers 56% of the time, YouTube 51%, and Google Business Profiles 28% (Boring Marketing).

Your own pages are still in the mix. In that same data, 51.7% of citations pointed to a brand's own site, with guides and blog posts making up 14.3% and listicles 12.4%. So first party content works, but it competes with a wall of community and video content the models lean on by default.

Zoom out and the concentration is stark. In an academic study across providers, the 25 most cited domains accounted for 23.8% of all citations, Wikipedia alone was 15.2% of that top group, and government sites were 33% of it. Meanwhile 59.1% of all cited domains were cited exactly once (academic GEO study). A handful of giants get quoted constantly, and everyone else fights over the long tail.

What this means in practice: your citation strategy is two tracks at once. Make your own pages so clearly the best answer that they earn a spot in the long tail, and get named on the trusted third party sources the engines already pull, like Reddit threads, review sites, and reputable roundups.

What makes a page get cited

Pages that surface in answers share a short and boring list of traits. An analysis of how ChatGPT selects sources found the ones that get cited disproportionately have a clear H1 that matches the question, a direct answer in the first third of the page, a visible last updated date in schema, and a named author. FAQ markup, original data, expert quotes, and recency all push in the same direction, and being in Bing's index is close to a floor, because ChatGPT's live search runs on it (AirOps).

None of that is exotic. Most brand pages fail it for the same three reasons every time. They bury the answer halfway down under a warm up paragraph. They hide or omit the date, so the page reads as undated and stale. And they publish under a company byline with no human name, which reads as lower trust to a model tuned on authored content.

Remember the retrieval funnel too: 85% of the pages an AI pulls never make the final answer (AirOps). Getting retrieved is table stakes. Getting kept is the job, and getting kept comes down to answering the exact question faster and cleaner than the other pages in the pile.

Five mistakes that keep brands out of AI answers

The brands that stay invisible tend to make the same handful of errors. None are hard to fix once you see them.

1. The answer is buried. The page opens with a warm up paragraph about how the topic has never been more relevant, and the actual answer sits four scrolls down. Engines reward the pages that answer in the first sentence or two, so a buried answer is a page that gets retrieved and then dropped.
2. No date, or a hidden one. A page with no visible last updated date reads as undated, and undated reads as stale. Recency is a real citation factor. Put a clear date on the page and set dateModified in schema, then actually keep it fresh.
3. No human author. Publishing under a faceless company byline lowers the trust signal for a model trained on authored writing. A named author with a real bio and a role does better. It costs nothing to add.
4. No original substance. Pages that restate what everyone already said get skipped in favor of pages with a number, a quote, or a piece of first party data. If your page could be swapped for any competitor's without changing a fact, an engine has no reason to pick yours.
5. Never measured. The biggest mistake is not tracking any of this. Brands assume they are cited because they rank on Google, then never check the AI answer itself. The check takes minutes and the result is almost always humbling, which is exactly why it is worth doing.

Fix those five and you are ahead of most of your category, because most of your category has not looked yet.

How the wording of a question changes your odds

This one is worth its own section because it is so easy to get wrong. The same brand, the same page, and the same engine can produce a near certain mention or a near certain ghost citation depending on how the question is phrased.

Short and human phrasing tends to name brands. Long and structured phrasing tends to cite without naming. Comparison phrasing names brands far more than plain information phrasing. So when you test your own visibility, you cannot test one polished prompt and call it done. You have to test the messy, real range of ways your buyers actually ask, because the wording swings the result more than almost anything you can change on your page.

This is also why a single manual check lies. Run the same prompt twice and you can get two different answers, with your brand named once and dropped the next. AI answers are non-deterministic. One check is a coin flip dressed up as a measurement.

Why you have to measure your own number

By now the pattern should be clear. The public averages are useful for context and useless for decisions. Citation rates depend on the platform, the industry, the question type, the exact wording, and the run. No industry average survives contact with your actual buyer questions.

So the number that matters is yours, measured the boring way: take your ten or twenty highest value buyer questions, run each through every engine, and record two things per run, whether you were cited and whether you were named. Repeat each question several times because the answers move. Average it into a citation rate and a mention rate with a confidence interval, so you know which changes are real and which are noise.

Do that, and you can finally answer the questions that count. Are we in the answer or not. Are we named or just linked. Who is beating us and which sources are propping them up. What moved after we changed a page. That is a scoreboard you can actually manage.

How to check your AI citations

You can do the first pass by hand. Open ChatGPT, Perplexity, Gemini and Google, ask your top buyer question in each, and note whether your brand is named and whether your site is linked. Do it three times per engine so the wobble averages out. It is tedious, but even a rough manual pass usually surprises people, because the silence is real.

When you want it done properly and repeatedly, that is the check we built. You can run a free AI visibility check on your domain and see, per question and per engine, whether you are cited, named, or missing, with a confidence interval on the score. To test one keyword against Google's answer box, the AI Overview checker shows whether an overview fires and exactly which sources it pulled. For the wider view across all five engines, the AI visibility checker rolls it into a single score.

If you want the playbook for actually improving the number, start with how to get cited by ChatGPT and the difference between a brand mention and a citation. And if you care about how much of the conversation you own versus competitors, that is AI share of voice.

Methodology and sources

The figures on this page come from public studies, each with its own method and sample, linked at every claim so you can check them yourself. Semrush analyzed 3,981 brand appearances across 115 prompts in 14 countries on four platforms. Boring Marketing ran 13,475 buyer question checks across ChatGPT, Perplexity, Gemini and Claude. AirOps studied how ChatGPT selects sources. The domain concentration figures come from a peer reviewed measurement study of generative engine citations.

We do not blend numbers from different studies into a single false precision, because they measure different things, and pretending otherwise is how bad stats get made. Where two studies disagree, that disagreement is the point: citation rate is not one fixed number, it is a range that shifts with the question. We update this page as new research lands and keep the date current at the top. If you spot a number that needs refreshing, tell us and we will fix it.