2026. That’s the year Google stopped selling a search engine and started selling a decision-maker. At I/O 2026, the company reframed Search The pitch, in Google’s own framing, is that you can ask whatever’s on your mind and Search will do more for you than ever before.
Read that pitch again and notice what’s missing. Nobody says “find.” Nobody says “results.” The verb is do.
A search engine used to have one job
The old contract was boring and beautiful. You typed a thing, you got a list of places that might have that thing, you clicked, you judged for yourself. Google was a librarian with a bad attitude about ads, but it was still a librarian. The judgment stayed with you.
The new contract collapses that. When Search does the deciding, it also does the narrowing, the ranking, and the editorializing before you see a single option. You are not looking at the web anymore. You are looking at a summary of someone’s model of the web, filtered through personalization you cannot inspect and monetization you cannot audit.
I review AI agents for a living. This is the same architecture as every agent product I’ve tested this year, just wearing the most trafficked URL on earth as a costume.
The part where publishers get to be right
When Google first started putting AI-assisted results in front of links, news publishers warned the impact would be catastrophic, because much of the industry depends on Google search to send readers to their sites. That warning was easy to dismiss as incumbent whining. It is harder to dismiss now that Search has been openly repositioned as the destination rather than the doorway.
The mechanics are not complicated. If an assistant answers the question, there is no reason to click the source. If there is no reason to click, there is less reason to publish. If there is less reason to publish, the assistant’s training and retrieval diet gets thinner. Nobody at the keynote has to be evil for that loop to hurt. It just has to work as designed.
Weird is not the same as bad, but it is the right word
What makes this feel strange isn’t the technology. It’s the tonal whiplash. Google spent two decades training a few billion people to treat the search bar as a neutral utility, like a light switch. Now that same box wants to know your preferences, weigh your options, and help you buy things. The interface did not change much. The relationship changed completely.
And the trust hasn’t caught up, for a reason Google earned honestly. Users have been complaining for years about basics, including the classic gripe that searching for a restaurant near you still won’t reliably surface its hours. That’s the unglamorous, structured-data end of the product. If the simple lookup is shaky, “let me handle your purchasing decisions” is a big ask.
What this means if you actually use AI tools
My read, having pushed a lot of assistant products past their demo scripts:
- Assume the answer is a product, not a fact. Any system that recommends purchases has a commercial gradient, whether or not it’s disclosed. Treat confident shopping advice the way you’d treat advice from a salesperson who’s very nice to you.
- Keep a second door. Have a habit, a tool, or a bookmark that gets you to primary sources without a summary layer. Redundancy is cheap insurance when one company controls the front page of the internet.
- Verify anything with a number in it. Prices, hours, availability, dosages, deadlines. Summaries are lossy, and the loss lands hardest on details that change.
- Notice when you stop clicking. That’s the real behavioral shift, and it happens quietly. If a week goes by and you haven’t visited a site you chose yourself, the assistant is no longer helping you research. It’s doing your thinking.
The honest verdict
I don’t think this was a mistake on Google’s part. It’s the logical move for a company watching users take open-ended questions to chat interfaces instead of a search field. Defending the query box by turning it into an assistant is the obvious play, and the distribution advantage is enormous. If assistants are how people navigate information now, Google would be foolish to keep shipping a card catalog.
What I’m skeptical of is the framing that this is purely a gain for users. A tool that finds things for you and a tool that decides things for you are different products with different failure modes. The first one fails visibly, with bad links you can ignore. The second fails invisibly, with good-sounding answers you never question.
Google didn’t get weird overnight. It got weird the moment the most-used utility on the internet started having opinions about what you should buy, and most of us didn’t stop to ask who those opinions serve.
🕒 Published: