Start with the decision, not the dashboard
AI-search reporting is easy to make noisy. A brand mention might be interesting, but it does not automatically show whether the right audience found a useful page, whether that page was eligible to appear, or whether visibility supported a commercial outcome.
Begin with one decision the measurement should support. It might be whether to improve a technical page, expand a topic cluster, protect a high-value query, or validate that a launch is reaching the intended market. That decision determines which data is useful and what can be ignored.
Use Search Console as the first source of truth
Google's Generative AI performance reporting gives a dedicated view of impressions from generative features in Search. It can be broken down by page, country, device, and date. That makes it a useful starting point for finding which pages are appearing and where visibility is changing.
The report should sit beside the normal Search performance report, not replace it. The strongest interpretation comes from comparing the page's AI-feature visibility with its broader impressions, clicks, query patterns, and the action the page is meant to support.
- Which pages appear in generative features and whether those pages match priority topics.
- Which countries and devices show the visibility, especially when a launch has a clear market focus.
- Whether a change in impressions lines up with a technical release, new information, or improved internal linking.
- Whether the pages that gain visibility also move visitors toward an enquiry, sign-up, product view, or other intended action.
Separate presence from performance
Presence is evidence that a page was eligible and selected for an AI-search feature. Performance asks whether that presence supports the business goal. The two measures are related, but they are not interchangeable.
For a service business, a sensible scorecard combines search demand for the topic, indexation and page quality, visibility for the important questions, qualified visits, and the downstream signal that matters to the team. That might be an enquiry, a demo request, or a sales conversation with the right context.
What not to optimise for
Avoid treating a third-party mention counter as a rank tracker. AI answers can change with the question, location, model, and source set. A tool can be useful for discovering patterns, but it cannot replace first-party Search Console data or commercial analytics.
Also avoid manufacturing a separate page for every possible prompt variation. Clear topic ownership, useful examples, and technically accessible pages are more durable than an expanding library of thin answer pages.