When an AI answer includes a company, product or website, it is tempting to record a single “visibility” win. But the answer may be doing three different things: mentioning an entity, citing a page as evidence, or recommending an option for a particular need. Those signals answer different questions and should not be collapsed into one score without a clearly stated method.

Three signals, three questions

1. Mention: were you named?

A mention is an explicit appearance of a brand, organization, product or person in an answer. It shows that the entity appeared in that response. On its own, a mention does not establish that the system consulted your website, relied on your material, or expressed a positive view.

2. Citation: was a source linked?

A citation is a source reference attached to an answer—often a link, source card or footnote. It indicates that the platform presented a page as supporting material. A citation is not automatically an endorsement, proof that every statement on the page was used, or evidence that the cited source changed the answer.

3. Recommendation: were you proposed as an option?

A recommendation occurs when the answer proposes a brand, product or provider as a suitable choice for the user’s stated criteria. Recommendations may include a citation, but need not. They can also be conditional, comparative, qualified or negative. Record the context and tone rather than treating every appearance as a recommendation.

Why the distinction matters

Research on Generative Engine Optimization studies how content may be surfaced or cited in generated answers, while research on search and recommendation systems examines selection under a query and a set of user needs. Together, these fields suggest a useful measurement discipline: observe the answer-level outcome directly instead of inferring endorsement from a link or commercial intent from a name-check.

For example, a company could receive many neutral mentions but few citations. Another might be cited as a source in an explanatory answer but never recommended. A third may be recommended for one narrow use case, while not appearing for broader queries. A combined count would hide these differences.

A simple measurement protocol

  1. Define the query set. Use a stable set of real questions, grouped by intent (informational, comparison, and provider or product selection).
  2. Record the context. Save the exact prompt, date, language, location where relevant, platform/model, and whether web search or citations were available.
  3. Code signals independently. Mark mention, citation, and recommendation separately. A single answer may contain zero, one, two or all three.
  4. Capture quality, not just presence. Note sentiment or stance, accuracy, recommendation rationale, cited URL and whether the source actually supports the associated claim.
  5. Repeat and report uncertainty. Answers vary over time and between runs. Report sample size, frequency and examples; do not present a small snapshot as a universal ranking.

What this can—and cannot—tell you

These observations can help identify whether a brand is discoverable for a particular set of questions, whether its pages appear among surfaced sources, and which themes or information gaps deserve review. They cannot guarantee future inclusion, establish a platform’s internal reasoning, or prove that a website change caused a change in answers. Platform outputs are dynamic, and visibility depends on factors beyond a site owner’s control.

For practical work, use the three signals as separate diagnostic lenses. Improve the underlying site and content for usefulness, clarity, accessibility and evidence; then re-run a documented query set. The objective is not to manufacture mentions. It is to make accurate, relevant information easier to find and verify.

Where Bilveo fits

Bilveo’s AI Visibility reviews are designed to help monitor how a site or brand appears in selected AI-assisted discovery checks. Interpreting mentions, citations and recommendations as distinct observations gives a more useful view than a single undifferentiated count. Results are observations—not guarantees of placement, citation or recommendation.

References

  1. Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K. and Deshpande, A. (2024). GEO: Generative Engine Optimization. Proceedings of the ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’24). arXiv:2311.09735.
  2. Yang, K.-C. (2025). News Source Citing Patterns in AI Search Systems. arXiv:2507.05301.
  3. Martinez, O. (2026). Optimizing Visibility in Generative Engines: A Critical Survey of Generative Engine Optimization (2023–2026). arXiv:2607.14035.
  4. Google Search Central. AI Features and Your Website.