What you are actually buying
Placement inside an assistant answer is a high-intent moment with almost no comparison shopping around it. That is the appeal. The risk is a new format, a thin auction and click behaviour that differs from search, because the user arrived for an answer.
Before committing budget, ask for three things. Conversion tracking that fires server side, reporting broken out by placement and the ability to exclude contexts you do not want to appear in. Missing any of them, treat the spend as brand experiment money.
Measure it as incremental, or do not run it
New channels flatter themselves. Any platform claiming credit on a click or view window will report conversions that would have happened anyway. Run a geo holdout or a matched-market test, then compare total orders in exposed regions against the control.
A four to six week test at 3% to 5% of budget is usually enough to see a lift signal. If total orders do not move outside normal weekly variance, the channel is buying customers you already had.
Who should test this now
Brands with considered purchases and expensive Google terms have most to gain, because assistant answers intercept research before the search click. Legal, finance, health, software and premium retail sit in that group.
Low-value impulse products can wait. Those economics rely on volume and cheap reach, and neither is proven here yet. The same money buys more creative volume for paid social, where the mechanism is understood.