This article was AI-generated as part of an experimental historical-content project. The date reflects the period being analyzed rather than the date the article was originally written.
Two stories about Google’s AI Overviews ran this week, and they barely touched each other.
On Thursday afternoon Alphabet reported first-quarter results. Search and other revenue grew 10 percent to $50.7 billion. Sundar Pichai said AI Overviews now has more than 1.5 billion users a month. Philipp Schindler, Google’s chief business officer, called the first quarter its largest expansion yet, with the feature now in more than 15 languages across 140 countries, and gave investors the line they were listening for: “For AI Overviews, overall, we continue to see monetization at approximately the same rate.”
A day earlier, people had been typing invented sayings into Google with the word “meaning” and posting the results. WIRED documented AI Overviews explaining phrases like “you can’t lick a badger twice” as if they were established idioms, sometimes with reference links. Google’s spokesperson explained that for “nonsensical or ‘false premise’ searches,” its systems try to find relevant results from the limited content available.
What the earnings line actually says
“Approximately the same rate” is a statement about economics. It tells investors that putting a generated answer above the links has not, so far, cost Google money. Schindler noted that ads now appear inside AI Overviews on mobile in the U.S. When an analyst asked about click-through and conversion, he declined: “I don’t think this is the moment to go into the details.”
That is a reasonable thing for a company to say on an earnings call. It is also a reminder of what the metric does not measure. It does not measure whether the answer was right, or what the answer said about the companies it described.
Where trust is actually decided
Trust in a generated answer gets settled in screenshots, news stories and the slow accumulation of examples that people remember. The badger idiom is harmless. The mechanism behind it is not.
A false-premise question about a company looks like this: “Why did [company] recall its product?” when no recall happened, or “Why was [executive] fired?” about someone who left on good terms. The same tendency WIRED described, building a plausible explanation rather than questioning the premise, could produce a confident paragraph explaining an event that never occurred. Rumors often arrive as questions of exactly this kind.
The publisher side of the trust ledger also sits outside the revenue line. In February, Chegg sued Google, arguing that AI Overviews cut the traffic it depends on. Every one of these disputes is part of how the feature is perceived, and none appears as a line item.
Why this matters for finance and IR teams
Investors and analysts use Google too. An AI Overview at the top of a search for your company’s name is now part of the research process for at least some of them, in front of an audience Google measures in the billions. I argued in 2023 that markets move on filings but reputation hardens on the results page. The results page now has a paragraph written by a model at the top.
So investor relations teams have a reason to check what AI Overviews say about their company, and to try the false-premise versions of the questions they fear most. Is there an overview? Does it accept the premise? Which sources does it cite?
Alphabet has shown that the economics of generated answers can hold. That settles the question for Alphabet’s shareholders. For the companies those answers describe, the open question is the one the earnings call was never designed to answer.