How to audit brand aliases in AI visibility reports

Before investigating why visibility moved, check what the tracker recognizes as your brand. A renamed product, shorthand or shared word can change the count without any change to the underlying answers. This guide proposes an attribution audit. The examples are fictional classification fixtures, not captured engine responses or panel results.

01 / Define the entity you intend to measure

Write a one-sentence scope: this report measures the named company, a particular product, or a defined group. List which subsidiaries, regional sites, former names and acquired products belong in that scope. Do not automatically count every product owned by the same parent as a mention of the product being evaluated. Keep competitor definitions at the same level.

Alias settings are a real configuration choice. Mangools documents a brand-variation field, while Ahrefs lets a brand entity contain multiple names and websites. Neither feature determines your editorial scope for you. Mangools' setup guide . Ahrefs' entity grouping documentation .

02 / Build a small alias register

For each candidate name, record the canonical entity, exact spelling, domain if relevant, language, reason for inclusion, ambiguity risk, owner and effective date. Include spacing and punctuation variants only where they plausibly identify the same entity. Put generic words and initials in a separate review list until the surrounding context can disambiguate them.

Keep website rules separate from name rules. Decide whether regional domains, documentation hosts and subdomains belong in your citation scope. Record former domains and redirects explicitly. A reseller mentioning your brand may count as a brand mention; a link to that reseller is still a third-party citation.

03 / Test positive and negative examples

Fictional fixture: the brand is 'Birch Ledger', a business software product. A sentence naming 'Birch Ledger' with its fictional product context is a positive candidate. A sentence about a birch tree beside a ledger should remain negative. The initials 'BL' alone need context. These fixtures illustrate review decisions; they are not a vendor benchmark.

Add cases for a domain-only reference, a former product name, a competitor with a similar name, a quoted passage, an unrelated acronym and a regional spelling. Have two people label the fixtures independently as in scope, out of scope or uncertain, with a short reason. Resolve disagreements before treating the rule as stable.

Do not assume that a tracker can accept custom fixtures or implements exact, case-insensitive or semantic matching. Ask the vendor to explain its method and show the same edge cases where possible. When the method cannot be inspected, disclose that limit and manually review a retained sample.

04 / Keep four observations separate

For an actual saved answer, record whether the answer text mentions the in-scope entity; whether it contains a linked citation to an in-scope owned site; whether it cites a third-party page discussing the brand; and whether any apparent match is ambiguous. A brand mention without a hyperlink is not an owned-site citation. A citation to a page you own does not necessarily mean the answer recommends your brand.

Retain the exact text span, cited URL, prompt ID, engine, market and capture timestamp with each decision. Inspect surrounding context before approving a match. An engine can cite an owned page as evidence while criticizing its product; do not relabel that as a positive recommendation.

05 / Review errors without hiding them

Review both reported positives and a sample of reported negatives. Positives can reveal accidental matches; negatives can reveal missed aliases. Log a false positive when the tracker claims an in-scope mention that the evidence does not support, and a false negative when your agreed rule identifies a genuine mention the tracker missed. Keep uncertain cases visible.

If you calculate an error rate, state exactly which retained sample was reviewed and how it was selected. A convenience sample of unusual cases is useful for finding bugs, but does not estimate accuracy across all answers. Do not extrapolate a handful of fixtures into a platform-wide accuracy claim.

06 / Treat a rule change as a measurement change

Version the alias register and preserve the old rules. If retained answers can be reclassified, compare both rule versions on the same records before attributing a count change to the market. If historical answers cannot be reprocessed, mark the rule-change date on the chart and avoid presenting the discontinuity as organic growth.

Our suggested change-log entry records the old and new alias, the reason, affected entity and markets, reviewer, effective date and whether past answers were reprocessed. Apply the same discipline to competitor aliases. Adding every competitor's parent company to one side of a comparison can alter its scope.

Use the audit in a buying pilot

Ask a vendor to demonstrate an explained match, an explained exclusion and an unresolved case. Choose the workflow only when its evidence is sufficient for the report you plan to deliver. Start with the Mangools research review for a concrete configuration example, then audit the export for the fields required to retain those decisions. Mangools AI Search Watcher review . Export audit . Citation measurement guide .

Sources checked 19 October 2026. No commercial relationship with Mangools or Ahrefs was disclosed to this reviewer. The classification rules, fixtures and workflow here are our proposed method; this article reports no completed prompt run and adds no score. Commercial disclosure . Panel evidence requirements . All field guides .