The Maze: Google's AI-written search summaries appeared for 88% of a fixed set of 400 US desktop commercial search phrases in August 2026, up from 45.75% a year earlier. Semrush, which sells search-measurement tools, published the study on October 21. It shows a change in the answers shoppers encounter, without measuring purchases.
Source: Semrush. AI Overview trigger rates by commercial search phrase type, September 2025 to August 2026, in Semrush’s tracked panel.
The Details:
The denominator matters. Semrush followed the same phrases across banking, software and ecommerce from September 2025 through August 2026. These included product reviews, prices, alternatives and brand comparisons. A keyword here means a search phrase, not an individual shopper. The study's 88% figure counts how many tracked phrases produced an AI Overview, Google's generated answer above ordinary search links. It does not describe 88% of all Google searches. Ecommerce's share rose from 32.62% to 85.11%. Mobile results and countries outside the US sit beyond the sample. The study also leaves the number of keywords in each subgroup undisclosed.
Comparison pages now feed a broader answer. A brand traditionally builds a comparison or pricing page to attract a visitor from a search listing. An AI Overview adds an earlier step: Google can summarize the choice before that visit, using several sources. In Semrush's example comparing the email-marketing services Mailchimp and Klaviyo, the answer cites a Klaviyo guide and outside websites, including software-service Zapier. A brand's own page can therefore contribute without controlling the whole recommendation. This example demonstrates the mechanism, but one screenshot cannot show how often any source wins across the panel. There is no panel-wide citation-share measurement.
More answers do not automatically mean fewer orders. The reported limitations are consequential: this study has no click, traffic or conversion data. A conversion means the action a business wants, such as a completed purchase. The counts cannot establish whether customers stopped visiting retailers, found better information, or bought elsewhere. They also do not explain Google's reason for expanding coverage. Semrush recommends monitoring AI visibility and sells tools for doing so. That commercial interest belongs beside its recommendation; it does not make the measured counts disappear.
MarketMaze Take: Search position and the first explanation a shopper reads are becoming different management questions. This panel supports checking both. It does not supply a revenue-loss estimate. For teams outside the US, the useful lesson is to inspect who shapes their comparison answers locally, rather than import the American coverage rate.
Accuracy has more than one address. The comparison example shows why fixing only a brand's own page may leave an outdated third-party description in the answer. Checking the actual cited pages can reveal which information needs attention. That is a specific audit of claims about a product, not evidence that paying for more monitoring or creator partnerships will generate sales.
Keep the sales test attached. Google's website-owner guidance says ordinary search eligibility still applies; no special AI markup is required, and inclusion is not guaranteed. It includes AI-feature traffic within overall Web search reporting. A retailer therefore needs its own visits and order data alongside answer checks. An appearance count tells the team where to investigate, not what the investigation will find.
Sources: Semrush study · PPC Land · Google Search Central
