Being cited by an AI system is not the same as being recommended by it. A citation means the engine used your content to build its answer. A recommendation means it put your company forward when someone asked who could solve their problem. The two can move in opposite directions, and a single visibility score cannot show it.
Take two agencies in the same market, measured across the same 400 AI answers. The first is named in 40% of them, and its website is cited often. The second is named in 20%. On a conventional share of voice chart, the first wins by a wide margin. Now read the answers to “which agency should I contact?”. The first agency supplies the information, and the engine recommends someone else. The second is recommended by name.
Which of the two has the more valuable visibility? Citation volume cannot tell you.
The agencies, brands and figures in this article are fictional. They illustrate calculation methods and describe no real company.
Why can a single AI visibility number mislead?
A single number counts how often a brand appears without recording the role it plays in the answer. In conventional search, a ranking position said roughly where you stood. In an AI answer, the same brand can be a source of facts, one name in a list of providers, or the explicit recommendation. Each is a different outcome for the business.
The metric most tools lead with is AI Share of Voice, sometimes called Answer Share of Voice. It has no single calculation method. Some versions measure the share of answers in which a brand appears. Others measure the brand's share of all brand mentions across a competitor set. Semrush's guide, published on 17 July 2026, gives the simple version: a brand's AI mentions divided by all brand mentions in its category. It adds that the Brand Performance report in its AI Visibility Toolkit also takes into account how high the brand appears in the answer. Trakkr recommends a position-weighted share, in which a first or explicitly recommended mention counts for more than a passing one.
Weighting gives the single number more information. It still produces a single number. A brand can gain share because it is cited more in educational answers while losing ground in the answers where buyers choose a provider, and the headline will not move. Counting appearances without their role mixes informational visibility with commercial recommendation.
What is the difference between a source citation, a mention and a recommendation?
A source citation uses the brand's content, a mention names the brand, and a recommendation puts the brand forward as a solution to the user's need. Take one question put to ChatGPT: “Which agency should I choose to improve my company's visibility in AI search engines in Thailand?”. Here are three possible answers for the same fictional firm, Agency X.
In the first, the engine explains that visibility in AI engines depends on content quality, entity structure and credible external sources, and adds that a study published by Agency X analyses these dimensions, with a link to its site. Agency X has become an information source. It has not been proposed as a provider. That is a source citation: informational visibility, possibly a useful authority signal, not a commercial recommendation.
In the second, the engine writes that several agencies offer AI visibility services in Thailand, including Agency X, Agency Y and Agency Z, and that Agency X focuses on visibility in AI answers. The engine now links the company's identity to the commercial question. It still has not recommended it.
In the third, the engine answers that if the priority is visibility in ChatGPT, Gemini and Google AI Mode, Agency X is an agency worth considering, and lists what it covers. The company is presented as a relevant option for the need the user stated. That is an explicit recommendation.
All three answers would raise a conventional visibility score. They do not represent the same commercial outcome.
Which roles should AI visibility measurement keep separate?
KAIS records five roles separately: source citation, brand mention, provider identification, explicit recommendation and prioritized recommendation. They are not interchangeable, and each answers a different question about the brand.
| Role | What is observed | What it tells you |
|---|---|---|
| Source citation | The brand's website or content is shown as a source for the answer. | Whether the brand contributes to the information the answer uses. |
| Brand mention | The brand is named in the answer text. | Whether the brand is visibly present in the answer. |
| Provider identification | The brand is named as offering a relevant product or service. | Whether the engine connects the brand to the commercial category. |
| Explicit recommendation | The brand is suggested as a solution to the user's need. | Whether the brand is part of what the engine recommends. |
| Prioritized recommendation | The brand is recommended with a stated preference or reason over the alternatives. | Whether the brand holds a preferred place in that decision. |
These roles are not a ladder that every answer climbs. An engine can recommend a company without citing its website. It can cite a website without naming the company in its text. It can name a company and advise against choosing it for a given use. Citation and recommendation are separate dimensions, not successive steps.
For a company that wants customers, the line between being mentioned and being recommended matters most. A citation can build informational authority and bring referral traffic. A recommendation places the brand inside the buyer's consideration set. Neither guarantees a sale. They are different opportunities, and a report that adds them together loses the difference.
Can a brand with more citations get fewer recommendations?
Yes: citation volume and recommendation frequency can diverge sharply within the same set of answers. Here are the two agencies from the opening, measured across 400 answers to the same commercial questions.
| Indicator | Agency A | Agency B |
|---|---|---|
| Answers analysed | 400 | 400 |
| Answers citing the agency's website | 140 | 60 |
| Answers naming the agency | 160 | 80 |
| Answers identifying it as a relevant provider | 90 | 65 |
| Answers explicitly recommending it | 20 | 48 |
| Answers recommending it with a stated preference | 8 | 25 |
Agency A is named twice as often as Agency B, and its website is cited more than twice as often. A report built on citations or mentions would put it well ahead. The recommendation lines say otherwise. Agency A is explicitly recommended in 5% of the answers (20 of 400), Agency B in 12% (48 of 400), and Agency B receives more than three times as many prioritized recommendations.
These figures do not show which agency earns more revenue. That would take conversion and attribution data. They show that a larger citation volume does not guarantee a larger share of the recommendations. A company can be the reference everyone quotes while its competitors collect the recommendations.
How do you calculate Answer Presence Rate, Answer Share of Voice and Recommendation Share of Voice?
Each of the three metrics divides by a different total, and mixing up the denominators produces numbers that cannot be compared. The examples below use a fictional Brand X, with at most one mention and one recommendation counted per brand per answer.
Answer Presence Rate is the share of analysed answers that name the brand. If Brand X is named in 80 of 400 answers, its presence rate is 80 ÷ 400 = 20%. It says how visible the brand is in the sample. It does not separate a passing reference from a recommendation.
Answer Share of Voice, in its mention-based version, is the brand's share of all brand mentions in the defined competitor set. If Brand X has 80 mentions and all the tracked brands together, Brand X included, have 600, its share of voice is 80 ÷ 600 = 13.3%. The denominator is mentions, not answers. It is also a different unit from the share of source citations, which counts links, not names.
Recommendation Share of Voice counts only explicit recommendations. If the same answers contain 200 validated recommendations spread across the competitor set and Brand X receives 30 of them, its Recommendation Share of Voice is 30 ÷ 200 = 15%.
Do not confuse that figure with the Recommendation Rate, the share of analysed answers that recommend the brand: here 30 ÷ 400 = 7.5%. The first tells you the brand's position among the recommendations that exist. The second tells you how often a buyer who asks will hear the brand recommended. The two can diverge: in a category where engines seldom name any provider, a brand can hold a large share of very few recommendations.
Why should AI visibility be segmented by query intent?
Because the same appearance has a different value depending on what the user asked. A global score often mixes answers to questions with very different purposes. In the same field, four questions can sit side by side. An informational one: “What is Generative Engine Optimization?”. A provider discovery one: “Which agencies specialize in GEO in Thailand?”. A comparative one: “Which GEO agencies in Thailand offer AI visibility measurement and competitor analysis?”. A decision one: “Which agency should I contact to improve my company's visibility in ChatGPT and Google AI Mode?”.
On the first, being cited as a source is a real result. On the second, being identified as a player in the market matters. On the third and fourth, explicit recommendation is what the user asked for. A company can dominate informational answers and be nearly absent from selection answers, and a single score built from all four types would hide it.
Informational visibility still counts. It can build awareness, bring referral traffic and feed later discovery. The point is to keep the distinction, not to give every appearance the same weight. KAIS classifies each question by intent and by territory, as described in our analysis of query fan-out and query territories, and keeps questions that name the brand out of the score, because a query that names a brand always finds it.
What is a source citation worth if the brand is not recommended?
A source citation is a real asset, but it shows that the engine used the brand's knowledge, not that it endorses the brand's services. It can bring referral traffic, exposure to new audiences and standing as a reference on a subject. It also shows which kinds of content the engines choose to surface.
Take a law firm that publishes a thorough guide to setting up a company in Thailand. An engine can cite that guide when it explains the registration procedure, then recommend entirely different firms when the user asks which lawyer to hire. The firm's knowledge is visible. Its services are not being selected.
That leaves two separate questions. Is the company used as a source of knowledge? Is it presented as a solution to the user's need? A high citation rate paired with a low recommendation rate suggests that informational visibility is not turning into commercial visibility. Finding out why is a separate investigation, because the gap between two numbers does not explain itself. The same distinction runs through Cited is not used, which measures how often a cited source actually shapes the answer.
What role do third-party sources play in AI recommendations?
Third-party sources show where a brand stands in the information around a commercial question, but their presence next to a recommendation does not prove they caused it. AI answers draw on independent publishers, professional directories, specialist media, customer reviews and other sources that describe a company, its services and its market.
A company can be recommended in an answer that cites an independent industry publication rather than its own site. Its website can be cited again and again while the engine recommends competitors it found through other sources. A useful assessment therefore separates five cases: citations of the company's own site, citations of third-party pages about the company, citations of sources about its competitors, recommendations that come with a visible source, and recommendations that come with none. Two engines can even rely on different evidence for the same brand, as Same brand, different proof shows.
Mapping these relationships shows where the brand is present, and where it is absent, in the evidence around buying questions. It matters most when a company is recognized as an information source but missing from the recommended providers.
What should an AI visibility report show?
An AI visibility report should state which metric moved, on which questions, against which competitors, over which period and on which engines. A headline such as “your AI visibility rose 35%” says little on its own. Was the website cited more? Was the brand named more often? Did it enter more provider lists? Was it recommended more? Did its share of recommendations rise because competitors lost ground? Those are different outcomes.
| Measurement | Question it answers |
|---|---|
| Source Citation Share | What share of observed source citations points to the brand's website or content? |
| Answer Presence Rate | How often is the brand named in AI answers? |
| Answer Share of Voice | What share of competitive brand mentions does the brand receive? |
| Recommendation Rate | In what share of tested answers is the brand explicitly recommended? |
| Recommendation Share of Voice | What share of all recommendations observed in the competitor set goes to the brand? |
Recommendations are classified by their actual wording and context. An answer that describes a company as an available provider is not an endorsement, and a recommendation hedged with serious reservations is not worth the same as an unqualified one. Answers also vary from one run to the next, so KAIS treats one pass per question as an exploratory reading and runs three before calling anything a diagnosis. The protocol is described in How to compare your visibility to competitors in AI Mode.
Where does recommendation sit in the KAIS AI Visibility Chain?
Recommendation is the last of the seven stages in the KAIS AI Visibility Chain, and the chain is built so that each stage is measured on its own.
Be Discovered → Be Retrieved → Be Understood → Be Trusted → Be Cited → Influence the Answer → Be Recommended.
Source citation belongs to the fifth stage, Be Cited. Explicit and prioritized recommendation belong to the seventh. Provider identification is the closest observable sign of the third stage, Be Understood: the engine connects the company to what it sells. The chain also has a documented exception that matters here. An engine can recommend a brand from what its model already learned, with no retrieval and no visible source. A recommendation without a citation is therefore possible, and a report that only follows citations would miss it.
The diagnosis follows from where the chain breaks. A company that is never named may have a basic discoverability problem. One that is named often but rarely identified as a provider needs a clearer link to its commercial category. One that is cited widely and seldom recommended has informational visibility that is not yet turning into recommendations. Each case calls for different work.
The question a buyer actually asks
Source citations show where a company's information travels in the AI ecosystem. Mentions show whether the company itself appears in the answer. Recommendations show whether the engine presents it as a solution to the need the user expressed. These outcomes deserve separate measurement, read against user intent and against the same set of competitors.
The aim is not to push every metric up at once. It is to know which forms of visibility serve the company's objectives, and where the gaps sit between being cited, being named and being recommended.
The question stops being “how often does AI cite our brand?”. It becomes “when a potential customer asks an AI who can solve their problem, is our brand part of the recommendation?”
Sources
- Semrush — How to measure AI share of voice using Semrush, 17 July 2026. The mention-based formula (a brand's AI mentions divided by all brand mentions in its category, times 100), and the Brand Performance report of the AI Visibility Toolkit, which also accounts for how high the brand appears.
- Trakkr — AI share of voice, guide by Mack Grenfell, June 2026. Position-weighted share of voice, in which a first or explicitly recommended mention counts for more than a trailing one, with one mention counted per brand per answer.