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.
Alphabet reported fourth-quarter results today, and the headline is the one Sundar Pichai led with: annual revenue passed $400 billion for the first time. The earnings release puts 2025 revenue at $402.8 billion and fourth-quarter revenue at $113.8 billion, up 18 percent. Google Search & other grew 17 percent in the quarter, to $63.1 billion. Anyone who spent the past two years expecting generative answers to hollow out the search business has a number to reckon with.
The figure I would read twice sits further into Pichai’s prepared remarks. Google was “able to lower Gemini serving unit costs by 78% over 2025.”
Why a cost line is a reputation number
Investors will hear that as margin. It also describes a change in the information environment around every company.
When generating an answer is expensive, a platform rations it. It shows generated text on some queries and not others, sends only the hardest questions to the strongest model, and keeps long conversations behind a subscription. When the cost of an answer falls by more than three quarters in a year, that rationing logic weakens. More queries get a generated answer. More follow-ups get answered at length. More surfaces get a model that can describe things.
Pichai’s other numbers point the same way. The Gemini app has over 750 million monthly active users. Gemini models process more than 10 billion tokens a minute through direct API use by customers, up from 7 billion last quarter. In the U.S., daily AI Mode queries per user have doubled since launch. AI Mode queries are three times longer than traditional searches, and nearly one in six is voice or image rather than text. Alphabet expects 2026 capital spending of $175 billion to $185 billion. And Pichai said Google is collaborating with Apple to develop the next generation of Apple Foundation Models, based on Gemini technology.
Taken together, this is a company building capacity to produce far more descriptions of the world, in more places, at a lower cost per description. Some of those descriptions are of your company.
What the slides don’t measure
None of these metrics say anything about accuracy. Daily queries, tokens per minute and unit costs measure volume and efficiency. They do not tell you whether the answer about a regional insurer’s claims record is right, whether a summary of a CEO’s departure has the correct year, or whether a comparison table has a fund’s fees right.
That is not a complaint about Alphabet’s disclosure. Earnings calls report what drives revenue. But it leaves a gap that investors and issuers should keep in view. The financial case for generative search is now large and well documented. Its reliability on any given company is not documented anywhere, because nobody is reporting it.
In 2023 I argued that markets move on filings, but reputation hardens on the results page. The same split now applies one layer up. Alphabet’s filing tells the market how many answers it can afford to produce. What those answers say is settled query by query, out of sight of any slide.
What follows
The longer queries deserve attention. A question three times as long is more specific, and specific questions about a company reach deeper into the sources: old litigation, a forum thread, a niche trade article. Cheap generation means those questions get full answers rather than a list of links.
For investor relations and financial-services communications teams, the response is familiar but more pressing. Keep primary sources current and dated. State plainly, somewhere authoritative, the facts most likely to be asked about at length, such as leadership history and how material events were resolved. And test long, specific questions, not only the company name, because that is where the volume is heading.
For investors the point is simpler. A wrong answer about a holding is now cheaper to produce and easier to repeat at scale. That makes the information around a company a little more like a risk factor, even though no filing will list it as one.