What is AI citation share?
The metric, the formula, the difference between a mention, a citation and a recommendation, and why vendors define it differently. This page states the definition used by the Citation Share method.
Neste texto
What is AI citation share?
Citation share is a visibility metric for answers written by AI systems. Imagine you run the same 60 questions about your market on one AI platform, three times each. In some of the answers, the platform lists your website among its sources. The fraction of valid answers that do is your citation share on that platform.
This page uses the definition of the Citation Share method, which Camillo Dantas created and documents on this site. The method is both a set of rules for observing AI answers in a repeatable way and a headline metric. It measures. It does not optimize anything, and it does not promise that any brand will be cited.
What is the formula?
For one platform and one date:
citation share = valid answers whose sources include the brand's domain ÷ valid answers
"Valid" means a complete answer to a valid question. Errors, refusals and empty answers leave the calculation. "The brand's domain" means the domains and subdomains listed in the scope before the measurement starts. The full rules for classifying each answer are in the measurement protocol (Portuguese) and in the English summary. Other metrics of the method, such as presence, recommendation and accuracy, are in metrics.
A result is always reported with its count, its rate and a 95 percent interval. A worked example, with a fictitious brand: 29 answers cite the domain out of 118 valid answers. The rate is 24.6 percent, and the Wilson 95 percent interval runs from 17.7 to 33.1 percent. If the same prompts are run on another platform, that platform gets its own line.
What is the difference between a mention, a citation and a recommendation?
These are three different events, and the method keeps them apart.
| Event | What the answer does | Counts as |
|---|---|---|
| Mention | names the brand in the text | presence |
| Citation | lists the brand's domain among the sources | citation share |
| Recommendation | indicates the brand as a suitable option for the request (decision queries only) | recommendation |
An answer can name your brand without citing your site, cite your site without naming the brand, or list the brand without recommending it. Collapsing the three into one "appearance rate" hides the diagnosis. If a platform names your brand but cites other sites, the question to ask is why it does not use your pages as a source.
Why does the definition vary between vendors?
There is no standard. When this page was written (October 22, 2026), the definitions found on vendor pages differed in what they count and in the denominator.
- Everything-PR defines citation share as the share of AI-generated answers across a defined prompt set in which a given brand appears. The page names five scoring components, such as frequency and crawl access, and gives no formula. It shows no publication date.
- AirOps, in a glossary entry dated August 7, 2026, defines it as how often a brand, author, domain or entity is cited relative to all citations in a defined set, with the formula (citations attributed to an entity ÷ total citations in the defined set) × 100.
Two things stand out. The first is the unit: the Everything-PR definition counts answers in which a brand appears, while the AirOps one counts citations within a set. The second is the denominator: all answers, answers that have sources, or all citations. A brand can have a 24 percent answer-based share and a 27 percent citation-based share on the same data, because one answer may cite several brands or none. Other vendors, including AI visibility platforms, publish their own wording, and this page does not describe any definition it could not open.
For that reason, a number without its definition cannot be compared with another. When you read a report from any vendor, ask four questions: what is counted, what is the denominator, which prompts and platforms, and how many runs.
How is it different from share of voice?
Share of voice is an older marketing concept: a brand's portion of the total exposure in a market, historically measured in media spend or searches. Applied to AI answers, it usually counts mentions. Citation share, in this method, counts sources.
The two can diverge. A brand can be named in 31 percent of mentions and appear in only 27 percent of the sources in the same set, which means it is discussed more than it is used. The article Citation Share and Share of Voice (Portuguese) works through a numeric example.
What sample size is needed?
The interval depends on the number of valid answers. For an observed rate near 25 percent, the Wilson interval at 95 percent covers roughly 13 to 43 percent with 30 answers, and roughly 18 to 33 percent with 120 answers. These widths assume independent answers, which is the best case, because repeated runs of one query resemble each other. That is why the method also computes the rate with the query as the unit and reports that interval.
The method asks for 50 to 100 questions, each run at least three times. With fewer than 30 valid answers per platform, treat the result as a pilot.
What are the limits?
- It is a sample. A fixed set of questions is a small view of what people ask. It does not describe every answer a platform gives.
- Answers vary. A study by SparkToro and Gumshoe.ai, published on January 28, 2026 (a vendor study, not peer reviewed, with 2,961 runs by volunteers using default settings), found that for ChatGPT and Google's AI answers the chance that two responses to the same prompt contained the same list of brands was under 1 in 100. The authors treat a visibility percentage measured across many prompts and runs as a more reasonable metric than a position in a list. That is why the method measures many prompts and runs, and reports intervals.
- Frequency is not quality. Citation share does not say whether the answer was accurate or favorable. The method tracks accuracy as a separate metric.
- Platforms differ. Each one shows sources in its own way, and a single average across platforms hides the information you need.
- It does not predict traffic or sales. A citation can send a visit, a mention can send nothing, and no source opened for this page quantifies either.
How do you measure it every month?
- 01Fix the scope: brand domains, three to eight competitors, language, country and platforms your audience uses.
- 02Freeze 50 to 100 questions, numbered and dated, with and without the brand name.
- 03Run each question at least three times per platform in a clean session, with the same settings, and log every answer with date, platform and sources.
- 04Classify each answer with written rules, and check a sample with a second person.
- 05Calculate the rate and the interval per platform, on the answer and on the query.
- 06Repeat at a fixed interval, monthly or quarterly, and compare two measurements through the interval of the difference. If it includes zero, you cannot separate a real change from noise.
A sheet template holds the log, and the calculator computes the rate, the interval and the difference between two measurements.
What is the next step?
To learn the full routine in one week, the mini-course Citation Share in 1 week walks through the first measurement, for R$ 147. The full course Citation Share: measure and prove, for R$ 397, adds the diagnosis and the action plan. Both teach measurement. Neither promises that any brand will be cited.
What is a good citation share?
There is no universal benchmark. The value depends on the question set, the platform, the market and the competitors, so the useful comparison is against your own earlier measurement with the same set and settings.
Is citation share the same as ranking?
No. A brand can rank well in classic search and be cited rarely by an AI platform that prefers other sources. Citation share looks at which sources an answer uses, not at where a page ranks.
Does a mention count as a citation?
Not in this method. A mention is the brand named in the text, and it counts as presence. A citation is the brand's domain listed among the sources. Both are logged, because the gap between them is a diagnostic signal.
Who created the Citation Share method?
Camillo Dantas created and named it, and the authorship is declared by him. Other vendors use the same words with different definitions, as shown above.
Sources
- Citation Share, Everything-PR (Everything-PR News LLC) (2026-10-22)
- Citation share, AirOps (Oshen Davidson) (2026-08-07)
- AIs are highly inconsistent when recommending brands or products (vendor study, not peer reviewed), SparkToro and Gumshoe.ai (2026-01-28)
Meça antes de otimizar
Defina o conjunto de consultas, registre a linha de base e acompanhe a evolução com o mesmo protocolo.