Define what you are counting

A mention is text naming a business or product. A citation is a visible source reference. A recommendation says a product fits a need. These events can overlap, but none automatically implies the others. Build separate fields before comparing brands. An answer may recommend a competitor while citing your educational article for a definition. Your domain has source visibility, but your product has not won that recommendation. Calling both events a ranking gain conceals the distinction your marketing team needs. Decide whether the report concerns your company, its products, or its published material.

Inspect the surrounding claim

Read the sentence around every detected name. Is the business presented positively, neutrally, as an example, or as something to avoid? Does the source support a product claim or background context? A simple string match cannot answer these questions reliably. Create a short annotation guide with examples. Review ambiguous cases manually, especially short brand names and common words. Preserve the original excerpt and URL so a second reviewer can check the classification without reproducing the entire search experience. Keep uncertain findings visible rather than forcing them into a flattering category.

Use an explicit counting rule

For an answer-level report, count a domain once even if several of its pages are linked. In a page-level report, preserve every distinct cited URL. Label the units clearly. An answer rate and a URL count are different measurements and should not share an unlabeled column. Hypothetical example: ten answers contain three mentions of Acme, five links to Acme pages, and one explicit recommendation. If three answers contain those five links, the answer-level citation rate is 30%, not 50%. The recommendation rate is 10% for this particular sample.

Check the destination

Open cited links and follow redirects. A familiar display name may lead to a directory, an old domain, or an unrelated company. Record the final destination and retain the original citation URL. This protects the report from identity mistakes and later URL changes. OpenAI advises checking cited sources because citations can be incorrect. Bing distinguishes citations from traffic. Apply those boundaries: linked evidence is observable, while trust and influence require additional evidence. Check whether the linked page actually supports the claim, especially when an answer combines statements from several sources.

Choose a response to the finding

If mentions are frequent but citations absent, inspect whether the site offers clear supporting material worth linking to. If citations are frequent but recommendations absent, examine whether the cited pages explain relevant product use cases. If recommendations contain mistakes, fix public facts first. Do not manufacture positive sentiment by prompting the model to praise the business. Keep naturally phrased buyer questions in the baseline and reserve brand-specific diagnostics for a separate investigation. The actionable report shows what happened, where it happened, and which part of the answer needs attention.

PUT IT TO WORK

Understand the difference between being named, linked, recommended, and visited in AI search.

Sources & further reading

Official documentation informs this guide. Examples are illustrative, and product interfaces or limits can change.

THE WORKSPACE WE'RE BUILDING

From search signals to a useful first draft.

SERP.so is in development. Join the waitlist for a workspace that connects research, content review, and publishing.

Join early access ↗