A dashboard can report that a brand appears in more AI answers while the commercial picture barely changes. The brand is named halfway through a response, its product is described using an old specification, and the link sends the shopper to a model unavailable in their country. The mention is real. The useful recommendation is missing.
This is the distinction a D2C AI visibility audit needs to expose. Brand presence, source attribution and purchase suitability describe different events. Treating them as interchangeable makes the reporting simpler and the subsequent investment harder to justify.
Read the answer as a shopper would
Consider a hypothetical Australian shopper asking which air purifier suits a small bedroom and has affordable replacement filters. An answer could name a brand, recommend its largest model and cite a retailer selling the previous version. A monitoring system might count a mention, a recommendation and a citation. A customer would still need to resolve size, ongoing cost and availability.
The audit should preserve that contradiction. Record the brand's appearance, then examine the actual recommendation: which product, for whom, in which market and on what evidence? A factual mistake that changes the buying decision matters more than whether the brand appeared first or third.
There is another easily missed distinction. An assistant can cite a brand's educational article while recommending a competitor. It can recommend the brand using an independent review without linking the brand's website at all. Domain citation rate and brand recommendation rate therefore need separate definitions.
A rate is only meaningful with its question attached
Suppose a hypothetical panel produces 40 valid buying-intent answers. Twelve name the brand. Five recommend a suitable product using accurate, market-relevant facts. Under those stated definitions, mention rate is 30% and accurate recommendation rate is 12.5%. Neither number estimates market share. Both describe the observed panel.
The panel itself deserves scrutiny. A question containing the brand's name gives the assistant a different task from an unbranded request for alternatives. Combining the two can inflate apparent discovery. Likewise, a highly specific scenario may reveal a valuable product advantage but tell us little about broader category coverage.
Build the initial panel from actual customer questions, support conversations and search behaviour where those are available. Record the wording, platform, search setting, language, market and date. Repeat observations across sessions and days. Preserve failed or unanswered tests separately rather than quietly excluding them to improve a percentage.
The source behind the mistake tells you where to work
A vague answer may point to missing comparison content. An incorrect specification may originate in an outdated retailer listing. A recommendation for an unsuitable use may reflect an ambiguous product claim. Those are different problems with different owners.
Microsoft's June 2026 AI Performance update adds intent, topic, citation-share and comparison views in preview. That context can help identify patterns within Microsoft's reported experiences. It still takes a person to decide whether a recommendation respects the buyer's constraints.
For Frontier, the valuable output of an AI Visibility Audit is the connection between an observed answer and a feasible intervention. If the fault lies in variant information, the next brief belongs with the product team. If local availability is unclear, regional commerce content deserves attention before another general category article.
What improvement should look like
A better answer may recommend the brand more often. It may also stop recommending the product in a situation it cannot serve. Both can represent progress in accuracy. The commercial objective is a credible shortlist position with suitable buyers, followed by a workable purchase journey.
That makes recommendation quality a more demanding measure than simple presence. It also makes the audit useful: the team leaves knowing which buyer decision is being lost, which evidence is missing and who can change it.