I noticed I had stopped opening search results.

Not completely. For train bookings or café hours, I still clicked. But for ordinary questions—how to fix a Gmail filter, which USB-C cable I needed, whether a strange Windows process was normal—I had started reading the AI answer and moving on.

It felt efficient. It also felt dangerous.

For years, search meant choosing. You typed a few awkward words, scanned ten links, ignored three ads, opened Reddit, closed a useless article, and slowly formed an answer. AI search removes much of that friction. That is exactly why people will use it.

The problem is that some friction was doing useful work.

We are no longer searching for pages

The old search box expected fragments: “best phone camera under 500,” “Amazon refund pending,” “airport Wi-Fi connected no internet.”

Now you can describe the whole messy situation. Google’s AI Mode accepts questions through text, images, files, video and browser tabs, while ChatGPT search can search the web and attach sources to its response. Search is becoming less like operating a directory and more like briefing somebody who will read it for you.

That is a real improvement.

A friend recently sent me a screenshot of a laptop charger and asked which replacement to buy. Earlier, I would have searched the model number, checked the wattage and compared listings. He put the photo into an AI search tool and got the relevant specifications in one conversation.

He still bought the wrong connector.

The model had identified the laptop family correctly but missed that his variant used a different port. The answer was neat, quick and almost right—which is the most expensive kind of wrong when shopping online.

The answer box has too much confidence

Traditional search results looked unfinished. A page of links was obviously a starting point.

An AI answer looks completed. It has paragraphs, headings and sometimes a reassuring little list of sources. Even when the system uses cautious words, the presentation feels authoritative.

I think this will be the main literacy problem of AI-first search. Not fake news in the dramatic sense. More often, it will be plausible compression: five sources reduced to one answer, with disagreements softened, missing details quietly dropped and uncertainty turned into tidy prose.

For harmless questions, this is fine. I don’t need a research process to learn why my Bluetooth headphones keep connecting to the wrong laptop.

For health advice, tax rules, visa requirements, online banking problems or a large purchase, treat the smooth answer as a map, not the destination. Open the source. Check the date. Notice whether the recommendation came from an official document, a forum post or a website earning commission from the product.

Yes, this takes longer.

That may be the point.

Search will know more about you than you expect

The next shift is personal context.

Google has offered an opt-in experiment connecting AI Mode with Gmail and Google Photos to produce more personalised answers. Instead of asking for a generic weekend plan, the system can potentially use your bookings, saved places and old photographs as context.

This will be incredibly useful. It will also make the phrase “search history” sound quaint.

An assistant that knows your flight, hotel budget, work emails and old photos can answer better questions. It can also build a detailed picture of your life.

Here’s the thing. People say they care about privacy in the abstract, then grant an app access to their inbox because it saves six minutes.

I’m not judging. I’ve done versions of this too. Convenience wins most small arguments.

Still, personal search should be used deliberately. Connect the accounts you need. Review permissions. Avoid giving a shopping assistant access to work email merely because the setup screen made it easy. The fact that a system can use more context doesn’t mean it deserves all of it.

Websites won’t disappear. The weaker ones might become invisible

There is a strange loop forming.

AI search depends on work produced by websites, journalists, forums and researchers. But when users get the answer without visiting, creators receive less traffic and money.

The web can survive fewer clicks. It cannot survive no reason to publish.

I don’t think websites are going away. People still want first-hand reporting, detailed reviews, GitHub discussions and the chaos of Reddit comments. An AI summary of a camera review is not the same as seeing the low-light samples. A three-line answer about an Amazon return dispute is not the same as reading what happened to twenty other customers.

But generic content written mainly to catch keywords is in trouble.

Good.

The internet has spent years producing articles that take 1,500 words to reveal where a setting is located. If AI removes some of that, I will not hold a candlelight vigil.

Original work will matter more. So will clear evidence and a recognisable point of view.

Sometimes I want the mess

AI companies are pushing search beyond simple answers. Google describes AI Mode as a place for follow-up questions and agent-like actions, while OpenAI’s deep research can search across the web and connected sources, then produce a cited report.

Useful, yes. But not every search should end in a completed task.

Sometimes I search because I don’t yet know what I think. I want to wander through opposing reviews, old forum arguments and one oddly specific blog post by a person who repaired the same router I own. Discovery is inefficient by definition.

This is where AI search needs a visible choice: answer mode and exploration mode.

Answer mode can tell me when a passport office closes. Exploration mode should show competing sources, disagreements and unexpected paths without rushing to settle the issue. Right now, many tools blend the two. They give a summary first and sources second, which quietly decides how much curiosity survives.

Better questions will help, but better doubt matters more

People often say prompting will become the essential skill. Maybe. Clear questions usually produce better answers.

I think knowing when to distrust a good answer will matter more.

Ask what the system may have missed. Request sources. Look for dates. For comparisons, ask which criteria changed the recommendation. When the answer affects money or safety, leave the conversation and verify it somewhere boring and official.

Search is becoming more convenient, conversational and personal. Most of us will use it because the old method was tedious.

I just hope we remember that opening another tab was not always wasted time.