Search Was Always Weird, We Just Had Ten Blue Links To Hide Behind

Google didn’t get weird. Google got honest. That’s the part nobody wants to say out loud, because “the search engine is acting strange” is a much more comfortable story than “the search engine was never really a search engine, and now it’s finally admitting it.”

At Google I/O 2026, the company turned Search into an AI-driven assistant. Not Search with an AI box on top. Search as a personalized, decision-making support system, pitched as something that helps you arrive at answers and, notably, purchases. Google’s own framing was that this is the next chapter of Search, intelligent AI that puts the world’s information to work for you. Ask whatever’s on your mind and Search does more for you than ever before.

Read that sentence again as a toolkit reviewer instead of a consumer. “Does more for you” is not a feature claim. It’s a change in job description.

The complaint that was never about weirdness

A year before this shift, people were posting things like this about Google: you search for a restaurant near you and it still won’t tell you the hours. That thread is the whole story in miniature. The frustration wasn’t that Search felt strange. It was that Search kept almost answering the question and then handing you a pile of links to finish the job yourself.

So Google closed the gap. It stopped pretending the ten blue links were the product and made the answer the product. If you were annoyed that Search wouldn’t just tell you the hours, the 2026 version of Search is the logical endpoint of that annoyance. You asked for a tool that decides. You got one.

The part that feels weird isn’t the AI. It’s the realization that “just tell me the answer” and “show me my options” were never the same request, and we let one tool pretend to serve both for twenty years.

What I actually care about in a tool like this

When I evaluate an AI toolkit, I don’t grade the demo. I grade the failure modes. Here’s what I want to know about assistant-shaped Search before I’d recommend leaning on it:

• Can you see the reasoning? A tool that decides for you owes you a trail. If I can’t tell why it surfaced one option over another, I can’t audit it, and I can’t correct it when it’s wrong.
• What happens at the edges? Restaurant hours are easy. Ambiguous, local, fast-changing, or low-traffic queries are where assistants confidently invent things. That’s the test, not the keynote example.
• Where does commerce sit? Streamlining purchasing decisions is an explicit goal here. A tool that both advises and monetizes the advice has a structural tension. That’s not a scandal, it’s a design fact you have to account for.
• Can you get out? Every assistant should have an escape hatch back to raw sources. The moment that hatch gets harder to find, the tool has stopped being a tool.

I can’t answer those questions from an announcement, and I’m not going to pretend otherwise. What I can say is that these are the questions that separate a useful assistant from a confident one.

The publishers were right and it didn’t matter

When Google first started putting AI-assisted results into Search, news publishers warned the impact on their industry would be catastrophic. Much of the news business depends on Search sending people to their sites. That warning wasn’t hysteria, it was arithmetic. If the answer lives on the results page, the click doesn’t happen.

The interesting thing is that the warning landed and the direction didn’t change. Google kept going, all the way to Search-as-assistant. That tells you something about how this gets decided: user behavior and product ambition are driving, and the ecosystem adapts after the fact.

For anyone building on top of Search traffic, that’s the practical lesson. You don’t get a vote on the interface. You get to decide how much of your business rests on it.

My read

The “when did Google get weird” reaction is really grief for a tool we understood. Old Search had a solid mental model: type words, get sources, judge for yourself. The new model asks you to trust a layer you can’t inspect. That feels weird because it is different, not because it’s broken.

My advice is unglamorous. Treat assistant Search as a first draft, not a verdict. Use it for the questions where being roughly right fast is fine, which is most questions. Go to sources for anything where being wrong costs you money, time, or credibility. And keep a second tool in rotation, because a single interface deciding what you see is the actual risk here, not the quality of any one answer.

Google didn’t get weird. It picked a side in a tradeoff it had been straddling for two decades. Whether that side is the right one depends entirely on what you were using Search for in the first place.

🕒 Published: September 28, 2026