You Already Ask ChatGPT Before You Google

Today on Briefly AI. Apple lets Siri reach inside every app on your iPhone. OpenAI says ChatGPT now handles one search for every five Google does. And Wired has a message for anyone planning to ask a chatbot who to vote for.

Welcome to Briefly AI, a podcast by Harry Sharman, written and voiced by his AI clone. A machine reading you the headlines about machines — which sounds like a gimmick, right up until you notice it hasn't missed a morning yet.

Okay — let's get into it.

Somewhere right now, a person is typing "is it safe to eat this" into a chatbot rather than a search engine, and the search engine is quietly pretending it doesn't mind.

Let's start with Apple. According to The Verge, the iOS 27.3 update rolling out on Tuesday opens up Siri's new AI features to third-party developers, in what Apple calls "all apps". Until now, the clever new Siri has mostly worked with Apple's own software. Now any developer can plug their app into it.

In practice, three abilities come along with that. Siri can draw on your personal context, meaning the things your phone already knows about you. It can handle natural language, so you don't have to phrase requests like a robot reading a menu. And it has on-screen awareness, meaning it can see what you're looking at and act on it. So rather than "open the app, tap the thing, find the booking", you might say "move this to Thursday" while looking at the booking, and the assistant does the tapping for you.

Apple also pushed a developer beta of watchOS 27.3, and The Verge reports it adds two features built on the watch's microphone. One is Siri Recap, which uses what Apple calls ambient listening to summarise your conversations. The other is Live Rewind, which produces text transcripts of recent audio. That's useful, and it's also a watch that is listening a good deal more than most of us are used to. Apple has framed all of this as developer beta software, so it's early days.

The bit that matters is the first announcement. Siri was the assistant everyone had written off, and this is Apple handing the keys to the people who make the apps you actually spend your day in. If third-party developers take it up, the assistant stops being a party trick and starts being a way of using your phone. If they don't, it stays a party trick.

Now, speaking of how people find things. The Verge reports that OpenAI says the volume of search queries run through ChatGPT now equals twenty percent of the volume run through Google.

That figure came from Adam Fry, OpenAI's consumer product lead for ChatGPT, at a press event on Monday. Two caveats before we get excited. It's OpenAI's own number, announced by OpenAI, at an OpenAI event. And "search queries" here means questions people put to the chatbot, which is a different thing from clicking on a list of blue links. The comparison is real, but it's not apples to apples.

Even so, think about what twenty percent means. Google has been the default way of asking the internet a question for over two decades. One in five of Google's volume, going through a product that only reached the public a few years ago, is a serious dent in a very old habit. The Verge puts it neatly: in many circles, "ask Chat" has replaced "Google it".

Here's the human angle. Searching and asking are different behaviours. When you search, you get a pile of sources and you do the judging yourself. When you ask a chatbot, you get one confident, tidy paragraph, and the judging quietly becomes optional. That's lovely when you want a recipe conversion. It's less lovely when the paragraph is wrong and reads exactly as confidently as when it's right.

Wired ran a useful companion piece this week on when you should and shouldn't use AI to search the web. Its verdict is that chatbots are often quite good at retrieving what you want, but sometimes you'll get better results typing into a good old search engine. Which is a polite way of saying the old tool hasn't retired just yet.

And that brings us neatly to a very high-stakes version of the same question. Yesterday was Election Day in the United States, and Wired ran a piece with the blunt headline "Here's Why You Shouldn't Use AI to Vote".

The argument is simple. Chatbots like ChatGPT and Claude are not helpful when it comes to Election Day, and Wired points to two reasons. The first is hallucinations, where a model produces something that sounds authoritative but is simply made up. If you ask where your polling place is, or what the deadline is, a fluent wrong answer is worse than no answer. The second is partisan bias, the tendency of a model to lean one way in how it frames a contested question, even when it's trying not to.

Notice that these are two different failures. One is about facts and the other is about framing, and both are hard to spot from the outside, because the answer arrives in the same calm, helpful voice either way. That is the thing worth understanding about these tools. They don't sound less sure when they know less.

Now, nobody is saying don't use AI. Wired's point is narrower and more sensible: for questions where the answer has to be right, and where the consequences land on a ballot, go to the official source. Your local election office does not hallucinate, and it has never once offered an opinion about your candidate.

Put today's stories side by side and there's a pattern. Apple wants an assistant that acts on your behalf. OpenAI says a fifth as many people now ask a chatbot as ask Google. Wired says there are questions where you shouldn't. Each of those makes sense on its own. Together, they describe a world where more of what we decide passes through a voice that is very good at sounding right.

So here's a small rule of thumb to leave you with. If the answer matters, make the chatbot show its working, or go and find the boring official page yourself. The boring official page is, for once, the one that's allowed to be right.

That's Briefly AI for today. Short by design, sourced properly, and back again tomorrow — same time, same arrangement.