Skip to content

ResourcesAEO Handbook

Citation share in AI search: what it measures and what a good figure looks like

Citation share tells you how often AI answers point to your pages. The figure means little until you know its denominator, its engine and the rivals beside it.

Updated Jack F, Co-founder

MeasurementAEOcitation shareAI citationsAI search

Part of the Answyn AEO Handbook.

Now readingWhat citation share is

Citation share is the proportion of the citations in a set of AI answers that point to your own domain: your domain's citations divided by all citations in the same answers. There is no universal good figure. In Answyn's own UK category tracking (6 October to 4 November 2026), the most-cited domain held 4.0% of 11,097 citations. Benchmark against the rivals named on the same questions, engine by engine, and track the direction.

A citation share figure is easy to produce and easy to misread. Two honest calculations on the same answers can give the same brand 2% and 10%, depending on what they divide by.

This page sets out the calculation, works it through on real answers, and shows how to set a benchmark that means something. A disclosure first: we build Answyn, which reports citation share, and the worked figures come from our own tracking of our own category, our own small numbers included.

What is citation share in AI search?

Citation share is the proportion of the citations in a set of AI answers that point to your own domain. When ChatGPT or Google AI Overviews answers a question, it shows the pages it drew on as links. Count those links across a fixed set of questions, count the ones on your domain, and divide the second number by the first.

We define it in two parts: how often answers cite a page at all, and how often that page is yours rather than a review site, a competitor or a third-party list. The second part is the share. The first sets its scale, because an answer that leans on seven sources leaves less room for any one domain than an answer that leans on two.

Answers carry a lot of links. Google says AI Overviews and AI Mode may use query fan-out, issuing "multiple related searches across subtopics and data sources", which lets them show a wider set of links than a classic search (Google Search Central). In our own tracking, every one of ChatGPT's 750 runs searched the web, and the 1,575 answers carried 11,097 citations between them, about seven per answer.

A citation is a link, which makes it a different outcome from a mention. A mention is your brand named in the answer's own words; a name that appears only as a cited link is a citation. The two move separately, as the figures further down show.

How do you calculate citation share?

Fix the set first: the same questions, market, engines and window every time. Then there are two calculations, and they are often confused.

  • Citation share: citations to your domain divided by all citations in the same answers.
  • Answer citation rate: answers that cite your domain at least once divided by all answers in the set.

The first counts links, so an answer that cites eleven sources weighs eleven times as much as an answer that cites one. The second counts answers, so it usually reads higher and answers a plainer question: in how many answers did we appear as a source at all?

An arithmetic illustration, not measured data: 100 answers carry 700 citations, and your domain is cited 14 times across 10 of those answers. Your citation share is 2% (14 of 700). Your answer citation rate is 10% (10 of 100).

Both are legitimate. A figure that does not say which one it is cannot be compared with anything, including last quarter's figure from a different tool.

On real answers, the first calculation looks like this. The data is Answyn's own tracking of its category: 25 questions about AI search visibility and AEO software, asked in the UK of ChatGPT, Google AI Overviews and Claude between 6 October and 4 November 2026.

Answyn's own tracking of its category, UK, 6 October to 4 November 2026: 25 questions, 1,575 answers.
DomainCitationsCitation share
semrush.com4454.0%
otterly.ai4363.9%
peec.ai4103.7%
youtube.com3803.4%
tryprofound.com3222.9%
answyn.com170.2%
All domains11,097100%

Each share is that domain's citations divided by 11,097. Semrush's is 445 divided by 11,097, which is 4.0%. Ours is 17 divided by 11,097, which is 0.2%, and all 17 came from ChatGPT.

Read the 0.2% as what it is: a new brand's starting figure, and the baseline we measure our own pages against. It is not a benchmark for anyone else.

Notice how thin the top is. The most-cited domain held 4.0%, and the ten most-cited domains held 3,015 citations between them, 27.2% of the total. The rest was spread across every other domain the answers cited.

Why one blended figure misleads

The engines do not cite the same kinds of page, so a share pooled across them describes none of them. Of the 11,097 citations, ChatGPT produced 5,554, AI Overviews 4,474 and Claude 1,069. Split by engine, the same domains look like this:

Share of each engine's citations, with the count in brackets: ChatGPT 5,554 citations, AI Overviews 4,474, Claude 1,069. Same dataset as above.
DomainChatGPTAI OverviewsClaude
semrush.com6.6% (364)1.3% (59)2.1% (22)
otterly.ai5.9% (330)1.7% (78)2.6% (28)
peec.ai6.5% (360)1.1% (50)0% (0)
youtube.com0% (0)8.5% (380)0% (0)
linkedin.com0% (0)6.6% (296)0.1% (1)
answyn.com0.3% (17)0% (0)0% (0)

Semrush's blended 4.0% is 6.6% of ChatGPT's citations and 1.3% of AI Overviews'. Fourth overall, youtube.com is first in AI Overviews, with all 380 of its citations there and none in ChatGPT or Claude; linkedin.com follows almost the same pattern. In this category ChatGPT leaned on vendors' own product, pricing and feature pages and on official documentation, while AI Overviews leaned on YouTube, LinkedIn and Reddit.

Answers also differ in how many sources they cite. Claude's 75 answers carried 1,069 citations, about 14 each, against about 7 per ChatGPT answer and 6 per AI Overview shown. Pool them and each Claude answer counts for roughly twice as much as a ChatGPT answer. Read citation share per engine, or the engine that cites the most sets your number.

Citation share is not visibility

Being cited and being named are separate outcomes, and our own figures show the gap. In the same window answyn.com was cited 17 times, while Answyn was named in the text of 2 answers out of 1,575, both in ChatGPT.

The gap runs the other way too. In 49 ChatGPT runs across 5 questions, ChatGPT's searches retrieved an Answyn page and the answer still left Answyn out. For a question about where in a buyer's research an assistant stops mentioning a brand, it read our customer journey handbook in 18 of 30 runs and named us in none of them.

So the two numbers answer different questions. Citation share asks whose pages the answers lean on. Mention rate, which we also call visibility, asks whose name the reader sees. A brand can lead on one and be missing from the other, which is why Answyn reports them as separate figures and never folds them into one score. Share of voice, like mention rate, counts names rather than links, as set out in how AI share of voice is calculated.

What earns the citation in the first place is a separate subject, covered with its evidence in how brands earn citations in AI answers.

What does a good citation share look like?

There is no universal good citation share, and a target offered without its question set, engine and market is a guess. The number depends on all three, and on how many sources each answer cites.

That last point sets a ceiling worth calculating. If answers cite about seven sources each, as ours did, a domain cited exactly once in every single answer would hold about 14%. In our category the leader held 4.0%. Where answers spread their citations this widely, a share of a few per cent can be a leading position; where answers cite two or three sources, the ceiling is far higher and the same figure means much less.

So set your own benchmark from the answers you track:

  1. Fix the set: the same questions, market and engines in every window, with each question tagged by buying stage.
  2. Calculate the share for every rival named in those answers, on the same questions and per engine. Their figures, not an industry average, are your benchmark.
  3. Split their citations into own-domain pages and third-party pages. In our category ChatGPT's most-cited sources were vendors' own pages, so owned pages can compete there; in AI Overviews the leaders were YouTube, LinkedIn and Reddit, which owned pages alone are unlikely to move.
  4. Track the direction in windows of the same length. Give a single page about four weeks and a topic eight to twelve before you judge it.

A good figure, then, is one that is rising against the rivals named on the same questions, on the engines your buyers use. Answyn's Pages screen shows which of your URLs earn those citations and, with LLM Analytics installed, which ones assistants fetch and never cite. Pages the engines already read are usually the cheapest to improve, and what counts as an update worth making is set out in the freshness handbook.

What to check when a tool reports citation share

If you are moving to a platform for its citation share tracking, ask these questions before you compare its figure with the one you had:

  • What is the denominator: all citations, or answers that cite you at least once? Ask for the formula in writing.
  • Which domains count as yours? Subdomains such as a help centre or a blog, regional domains and a separate docs site all need to be on the list, or your own citations are counted as somebody else's.
  • Can you read the figure per engine, or only as a blend?
  • Does it count only the pages an answer cites, or also pages the engine read during its searches and did not cite? Both are useful if they are kept apart.
  • Can you open the answers and links behind the number?
  • How far back are answers kept, and can you export the citations so your old baseline comes with you?

Run the old and new tools on the same questions for a few weeks before you switch platforms, so the two baselines overlap and you can see how far the definitions differ.

Google's own reporting on whether AI Overviews cite your website will not give you this figure. Search Console's generative AI performance report, announced on 3 June 2026 (Google Search Central blog) and rolled out to all sites by 31 August 2026, shows how often your URLs appeared in AI Overviews and AI Mode, by page, country, device and date (Search Console Help). It counts impressions for your own site only, so there is no total to take a share of.

Answyn reports citation share over time, own-domain against third-party citations, and the domains the engines lean on, and every figure opens back to the answers that produced it (what AEO Hub measures). Which citations count as yours follows the website on your brand profile, so if that is wrong, your own citations are missed. Access is by explicit invitation; enquire if you want your category tracked this way.

A note on the data

The worked figures come from Answyn's own tracking of its category: the Answyn project (answyn.com), 25 questions in 8 topics, asked without the brand name, in the UK market, from 6 October to 4 November 2026. ChatGPT answered each question every day (750 answers), Google AI Overviews was checked every day (750 answers, 30 of them with no overview shown), and Claude answered 75 times, last on 30 October. That is 1,575 answers and 11,097 citations, pulled on 4 November 2026. Claude is not on Answyn's standard roster, and a plan adds it only by agreement; we collect it for our own project.

Shares are rounded to one decimal place. It is one category in one market over one month, so read the shape as the finding: how thinly citations spread, and how differently the engines cite. The levels in your category will differ.

Questions people ask

What is a good citation share for a mid-sized UK brand?
There is no universal benchmark, because the figure depends on the questions you track, the engine and how many sources each answer cites. In Answyn's own UK category tracking (6 October to 4 November 2026), the most-cited domain held 4.0% of 11,097 citations. Take your benchmark from the rivals named on the same questions, per engine, and judge whether your share is rising against theirs.
How do you calculate citation share?
Divide the citations that point to your domain by all citations in the same set of answers, keeping the questions, engines, market and window fixed. Some tools report the share of answers that cite you at least once instead, which usually reads higher. Ask which one you are looking at before you compare figures.
Is citation share the same as share of voice?
No. Share of voice counts the brands named in the answer text, each once per answer, while citation share counts the links an answer cites. In our own tracking answyn.com was cited 17 times in a window where Answyn was named in 2 answers, so the two move independently.
Why is our citation share different in ChatGPT and Google AI Overviews?
Because the engines cite different kinds of source. In our category ChatGPT leaned on vendors' own pages and official documentation, while AI Overviews leaned on YouTube, LinkedIn and Reddit, and all 380 youtube.com citations came from AI Overviews. Read the figure per engine before you act on it.
We're switching from Rankscale or Brandlight. What should we check in the new platform's citation share?
Check the denominator, which domains count as yours (subdomains included), whether it splits by engine, whether it counts pages an engine read but did not cite, whether you can open the answers behind the figure, and how much history you can export. Run both tools on the same questions for a few weeks so the two baselines overlap.
Does Google Search Console show citation share?
No. Its generative AI performance report, rolled out to all sites by 31 August 2026, shows how often your URLs appeared in AI Overviews and AI Mode, by page, country, device and date. It counts impressions for your own site only, so there is no total to take a share of.
Is citation share the same as prominence?
No. Prominence is how clearly an answer names you: first, among others or only in passing. Citation share is about the links. Read both, because an answer can cite your page and still put a rival first.

Sources

Run the numbers on your own brand.

Answyn puts the questions your buyers actually ask to every major engine, daily, and records who gets named and who gets cited.