Google Ads Support: Gemini AI Agents Resolve 75% of Queries

What happened

Alphabet used its Q2 2025 earnings call to disclose that Gemini-powered AI agents now autonomously resolve 75% of support queries submitted through Google Ads, according to CX Today. The figure was presented as a concrete measure of how deeply generative AI has been embedded into one of Alphabet's largest customer-facing support operations.

The disclosure ties Gemini's rollout directly to a named business line — Google Ads support — rather than a general claim about AI capability, giving analysts and CX practitioners a specific, attributable benchmark for large-scale automated query resolution.

Why it matters

Most vendor claims about AI customer service are vague or self-reported without context. A 75% autonomous resolution rate, disclosed on an earnings call and tied to a specific, high-volume support channel, gives the industry a rare, accountable data point to benchmark against. For CX leaders weighing how far to push automation, this is the kind of figure that will shape budget conversations and vendor pitches for the next several quarters.

It also signals a shift in how automation success is being measured: not by deflection rates or cost savings alone, but by resolution — whether the customer's issue is actually solved without human intervention. That distinction matters for behavioral economics of service, where perceived effort and closure, not just speed, drive satisfaction and repeat trust in a support channel.

By the numbers

• 75% of Google Ads support queries are now resolved autonomously by Gemini-powered AI agents, per Alphabet's Q2 2025 earnings disclosure.

The Renascence take

The headline number is impressive, but the more useful question for operators is what happens in the remaining quarter of cases — and whether Alphabet's definition of "resolved" matches what advertisers themselves would call a good outcome.

A 75% autonomous resolution rate is a strong signal of technical maturity, but resolution and satisfaction are not the same metric, and Alphabet hasn't disclosed which one it's optimising for. The real test of any AI support layer is what happens at the 25% edge — how gracefully it recognises its own limits and hands off, because that failure-handoff moment is where most trust is won or lost, not in the easy 75%. Operators chasing similar deployments should instrument the escalation path as rigorously as the resolution rate, since a system that resolves fast but frustrates on failure will erode the very efficiency gains it claims.