Combine available platform reports, repeatable observations and commercial outcomes. Keep each metric’s definition intact instead of merging incompatible signals into a success score.
A screenshot showing your brand can be a useful signal. It does not establish how many people saw it, whether they clicked or whether they became customers. The trouble starts when that isolated moment becomes a supposedly permanent ranking.
By October 2026, some platforms offer dedicated reports. This guide proposes a practical way to combine those reports with your own observations, without claiming to measure every AI experience or promising that visibility will produce growth.
Define the event before measuring it
Keep four events separate: a brand mention, a linked citation, an attributable visit and a commercial opportunity. An answer may name your brand without linking to it; it may also cite an educational guide without recommending your services. Label those events differently instead of treating every appearance as an endorsement.
Our suggested starting point is one decision: improve a service explanation, resolve a comparison or strengthen a page that already attracts inquiries. Write down what evidence would justify action and who will review it. This prevents the team from changing its objective after discovering a flattering screenshot and overlooking observations that tell a different story.
Read each report according to its scope
Search Console’s generative AI report records impressions of site links in supported Google features, with dimensions including page, country and device. Bing AI Performance reports citation activity and cited pages across its supported experiences, alongside sampled grounding queries.
Do not combine these numbers into a claim about unique users across all AI systems. Keep the source, period and metric definition in separate columns. If a report is unavailable or has little data, record the limitation. Missing information does not prove that your brand is absent everywhere, and an estimate should never be presented as an observation simply to fill a dashboard.
Sources: Google Search Console: Generative AI performance report (Search) · Microsoft Bing: Introducing AI Performance in Bing Webmaster Tools
Build a sample you can repeat
Start with questions from genuine sales conversations: the problems your services address, criteria for choosing a provider and implementation requirements. Separate branded questions from unbranded questions. Otherwise, an apparent improvement may simply mean that you began asking the system about your own company.
Record the exact wording, language, country where configurable, date, platform, conversation context and cited URL. Repeat the set on a sustainable schedule and preserve answers that do not mention you. Describe the outcome as an observed sample, not total market share. When the questions or conditions change, start a new series or explicitly document the difference before comparing results.
Illustrative example: a citation with little commercial value
Imagine a consulting firm that offers implementation and training. In a sample, an AI system cites its glossary to define a term but never cites its service page when the question asks for professional help. This is a hypothetical scenario, not a Patagonia client result.
The next move would not automatically be twenty more glossary entries. We would first check whether the service page explains fit, deliverables, prerequisites, limitations and the next step. Relevant glossary content could then link to that page. The change aims to make the journey more useful; it does not promise that the system will choose the new URL or that a citation will turn into a sale.
Connect observations to the commercial journey
Where analytics identifies referrals from AI platforms, examine entry pages, subsequent actions and completed inquiries. Compare those records with opportunities qualified by the sales team. A visit or form submission can have weak purchase intent and should not carry the same weight as a valid opportunity.
An optional question about how someone found the business can add context without replacing analytics. Self-reported discovery is imperfect too. Keep available attribution, the contact’s own explanation and the team’s inference separate. This allows you to recognize useful signals without presenting a level of precision that the evidence does not support or claiming that a single touchpoint caused the entire purchase.
End each review with one bounded decision
Choose a page and a hypothesis for each review: clarify a condition, document a method, correct outdated information or add authorized evidence. Record the implementation date and keep other controllable variables stable. If campaigns or the offer also change, note that before interpreting the next period.
The useful deliverable is a decision log: what was observed, what changed, what happened next and what remains uncertain. At Patagonia, research tools including Moz support this work; their samples are not platform certifications. If the evidence is still weak, the right next action may be better measurement rather than additional content production. That is a valid outcome, not a failed reporting exercise.
REFERENCES
Sources and review
Source documentation consulted on October 24, 2026. Platforms and features may change; check the official source before implementation. The English adaptation retains the source review date of the original article.
- Generative AI performance report (Search) Google Search Console
- Introducing AI Performance in Bing Webmaster Tools Microsoft Bing
This guide provides general guidance. It does not guarantee search rankings or replace a review of your business, locations and data.