30 conversions in the prior 30 days before running an experiment
4-12 weeks before evaluating experiment results
8% average incremental conversion lift, per Microsoft

Topic Insights, and the vocabulary around it

Topic Insights groups AI citations by subject rather than listing them flat. The output is which topics AI systems associate with a brand, how often the citations occur, and which subjects are missing coverage entirely.

Three terms carry the reporting. Grounding queries are the retrieval searches an AI system runs before it produces an answer. Citation share is how frequently a domain turns up as a cited source. Share of authority is how often one domain is cited against the sources competing with it.

The grounding queries are the part with reach beyond this dashboard, because they are search behavior that no keyword tool reports. Compared against an existing search-terms report, the overlap and the gap are both informative.

What the experiments require

Two experiment types arrived for Performance Max. Uplift experiments measure what adding Performance Max alongside existing campaigns does. Upgrade experiments compare an existing Search or Shopping campaign against Performance Max once it has been migrated.

The conditions are specific: at least 30 conversions in the previous 30 days, bidding targets and product groups and campaign settings held constant between test and control, and four to twelve weeks before results get read. Microsoft attaches an average 8% increase in incremental conversions from Performance Max campaigns to the pitch.

Separately, the Ad Preview Hub grew past Audience ads to cover Performance Max and Bing search-results previews, with shareable links for pre-launch review. That is a continuation of the same tool Microsoft brought to Performance Max in July.

What to do about it

Check the floor before planning the test

Thirty conversions in thirty days is the gate, and it disqualifies a lot of accounts that would otherwise be first in line for an uplift test. Run that number for the campaigns you would want to test before designing anything, because the alternative is committing four to twelve weeks to a comparison that will not separate signal from ordinary week-to-week variation.