The short answer
A mention means the answer names your brand. A citation means the answer links or attributes information to a source. A recommendation means it suggests your brand for a use case. Measure them separately: a page can be cited while a rival gets recommended, and a brand can be mentioned negatively.
Imagine an assistant recommends a competitor and links to your carefully researched industry guide as supporting evidence. Your content team has earned a citation. Your sales team has a perfectly reasonable follow-up question.
This is why a single “AI visibility” counter can leave everyone talking past each other. The useful unit of analysis is the answer, with its names, advice and sources still attached.
Read the answer in three layers
| Signal | Orbit | Northstar |
|---|---|---|
| Brand named | Yes | Yes |
| Explicitly recommended for this need | Yes | Not clearly |
| Own website cited | No | Yes |
This is a reading of a made-up example, not a universal classifier specification. A phrase such as “another option” may count as a recommendation under a broader rule. The important part is to state your rule and keep the quote so somebody else can disagree intelligently.
First position is another separate feature. The first brand in a response might appear in a warning, an alphabetical list or a comparison heading. “Named first” is a reproducible observation when defined carefully; “the AI’s favorite” is an interpretation you may not be able to defend.
A citation is a trail to inspect
Open the source. Check the exact page, what it says about the brand, whether it is current, and whether the cited passage supports the answer’s claim. A domain-level list is a useful starting point, but it can hide the difference between a product review and an unrelated help article.
Record whether the source belongs to the brand, a competitor, a publisher, a directory or a community. If several answers cite the same listicle, you have found a recurring source in your sample. You have not demonstrated that buying placement there will cause a ranking increase.
Also keep the product brand separate from the source host. An answer can cite a retailer while recommending a manufacturer. Counting the retailer’s domain as the recommended brand would distort both the competitive picture and the action you take.
Different signals call for different work
| Observed pattern | Useful next investigation |
|---|---|
| Named, but described incorrectly | Find the quoted claim and check your own public product facts. |
| Cited, while a rival is recommended | Examine the buyer criterion the rival satisfies and your evidence for it. |
| Recommended without an own-site citation | Check which independent sources support the recommendation. |
| Missing from relevant answers | Review the actual alternatives and sources before commissioning new content. |
Sometimes the rival is simply a better fit. If the prompt requires a feature you do not offer, publishing a clever comparison page does not remove the product gap. That is useful intelligence too; it belongs in a product conversation rather than an SEO ticket.
If you do fit the need but the evidence is hard to find, the remedy may be smaller than a new content program: a clear availability statement, a documented integration or a current specification with an owner and date.
A scorecard a colleague can audit
- Question and exact answer, with provider and collection date.
- Brands named, including newly discovered alternatives.
- Recommendation label plus the sentence that supports it.
- Cited page URLs, with owned and third-party sources distinguished.
- Any uncertainty: ambiguous aliases, broken sources or unclear recommendation language.
Review the most consequential answers by hand, especially before sharing an unfavorable claim about a competitor. Automated labels make a large dataset manageable; they do not make every interpretation correct.
You can use AI share of voice to summarize the set. Keep these layers underneath it. The summary should help someone find the interesting answers, not replace reading them.
Sources & context
This guide responds to a question raised in Anyone here actually improving Share of Voice in LLMs? on Reddit. The discussion informed the topic; it is not evidence of the effectiveness of any product mentioned there. Platform documentation is linked alongside the claims it supports.
Examples and figures are illustrative unless explicitly stated otherwise. Written and published by Share of Voice, a Superstellar LLC product. We sell AI visibility reports; our commercial perspective is worth keeping in mind.
Spotted something we should correct? Let us know.