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.
Yesterday at I/O, Google made Gemini 3.5 Flash the default model in AI Mode “for everyone globally.” Google also says AI Mode has passed one billion monthly users a year after launch, a figure Sundar Pichai repeated in his keynote. That will be the headline, and it matters: the model writing the first generated paragraphs about a company has changed again, for every user, without anyone opting in.
The announcement that deserves more attention from reputation teams is a different one. Google is rolling out what Elizabeth Reid calls the biggest upgrade to the Search box in more than 25 years. The new box expands as you type, accepts text, images, files, videos and Chrome tabs as inputs, and “helps you formulate your question with AI-powered suggestions that go beyond autocomplete.” It is starting to roll out in all countries and languages where AI Mode is available.
Autocomplete was always a reputation surface
Long before generative answers, the first thing many people saw about a company was not a result. It was a prediction. Type a company’s name and the suggestions underneath told you what other people were asking: about its CEO, its stock, its layoffs, whether it was a scam.
Google has documented how that works for years. Its help page on autocomplete says predictions “reflect real searches that have been done on Google,” shaped by language, location, trending interest and a user’s past searches. It also lists policies, including one against predictions that associate disparaging or sensitive terms with named individuals, and one against predictions that read as accusations of serious malevolent acts without well-established evidence. Those rules exist because a suggested query can work like a claim, even though Google says predictions “aren’t assertions of facts or opinions.”
From predicting queries to proposing questions
Autocomplete completes a string. A box that helps formulate the question is more ambitious. It proposes what the user might really want to know, in natural language, before any answer is written. Google has not published how those suggestions are generated or which policies govern them, and it would be a mistake to assume they work like autocomplete. That is the point. The question layer used to be narrow and well documented. It is now broader and, for the moment, less explained.
It matters because a question frames its answer. “What does this company do?” and “Is this company in trouble?” pull different sources into AI Mode, and the second tends to surface the coverage a communications team would least like to lead with. If the box offers the second framing to someone who typed only a name, the answer that follows can be accurate and still unfavorable. Nothing in it needs to be wrong for the first impression to shift.
The new inputs add a wrinkle of their own. A user can now hand Search a file: an investor deck, a contract, a screenshot of a rival’s comparison chart. Whatever the user supplies becomes part of the question, and a company’s materials may reach the answer through someone else’s framing of them.
What to watch
Google says people will “continue to get a range of results,” and follow-ups from an AI Overview into AI Mode are now live on desktop and mobile worldwide. So results are not disappearing. The path into them is getting longer and more guided.
Teams that track only rankings, or only the text of AI answers, will miss that first step. It is worth recording, for the company’s name and its executives’ names, which questions the new box suggests, in which markets, and whether those suggestions lean toward the facts the company is best documented on or toward its controversies.
In 2023 I argued that Gemini might change who summarizes you first. This week suggests a further step. Before Google summarizes you, it may help decide what you are being asked about.