As soon as it was launched, almost 2 years ago, ChatGPT raised an alert at Google, which until then was comfortable in the position of leader — synonymous, even — of “internet search”. Chat signaled that at some point in the future, super-smart artificial intelligence models could radically change the way we search for information on the web, offering ready-made answers instead of a list of links.
It can be said that the future arrived for many people on Thursday (31).
To understand how, compare these two screenshots, which I just took from my cell phone. Google on the left, ChatGPT app, with Thursday’s updateon the right side. The search was “I want to eat Neapolitan pizza”.
We are so used to Google that we barely realize how far from ideal the service is nowadays. Despite knowing precisely where I am, the search engine preferred to place a sponsored link for a pizza that is not considered the best and is not close to my house, but which I paid to appear up there.
Then came another sponsored result, which was not a pizzeria, but a list (behind a paywall).
READ MORE: AI Studio – Google’s powerful artificial intelligence that few people are using
Then comes a map with different options, a list of related questions, such as ‘what is the difference between Neapolitan and traditional pizza?’ and 4 videos on the topic — two of them very old.
In other words: Google is not exactly “solving my search”, but enjoying what he knows I want so he can offer me related products. If by chance you get a matchcool; but this is not always possible, given that it will always be at the service of three clients: me, advertisers and other Google services, such as maps and YouTube.
It’s that story: if you’re using something for free on the internet, there’s a good chance that the product is you.
But look at ChatGPT’s answer again. The service immediately mentioned the only valid option (it is both the best Neapolitan pizza and the closest to my house), and explained in a few lines why that was the case, with a link to the place. The answer is not perfect — the number of reviews and the average rating seem false — but it is an answer close to what I, a fan of the Naples roundabout, would expect.
How is this possible? Thanks to the “web search” functionality within ChatGPT, launched on 10/31. Initially only for those around 11 million subscribers (costing US$20/month), it consists of very quick internet searches to add to the results generated by AI.
It works more or less like this: ChatGPT transforms what you wrote into several mini-searches; throw them into tools like Bing or direct access to content sites to find results; go to the websites, get the texts, summarize what you found and only then write a response, making reference to the sources.
All of this typically happens in less than two seconds, thanks to a new model fine-tuned for the task, which has just debuted and has impressed everyone with its speed.
With these innovations, ChatGPT is close to resolving two of the three major disadvantages it had in relation to Google: sometimes outdated knowledge and the occasional invented response, the so-called “hallucination”. With access to the web and the ability to check facts, answers tend to be significantly more relevant and complete.
See, for example, the answer to a search I did about Ambev’s results in the third quarter, an hour after the conference with investors:
Now consider the way old to get this information on Google. With luck, after typing the search, you would see in the first results a link to the company investor relations pageI would click there and download the presentations. Or I would search directly on the news sites for something about it, without the right to follow-up questions, which you can ask on ChatGPT.
The AI-generated canned response experience is definitely faster. But it can be limiting. As ChatGPT normally uses news sites as a basis, it is likely that it left out some interesting elements, focusing on the lowest common denominator, what everyone reported. In the case of Ambev, ChatGPT did not think about the interesting challenges of beers in the Chinese market, an angle that we cover here at InvestNews.
Is this the future? Will web users get used to versions that are perhaps less complete and full of nuance, but delivered more quickly and in short? Aravind Srinivas, founder of Perplexity, bets that this is the natural way of things. The reason: he says that “people are lazy. A good product is what allows them to be more lazy, no less.” We talk about this in a recent video from our IA: How to Use:
AI: Why “Google” will be a thing of the past
Perplexity has been looking for a solution similar to SearchGPT since before Chat was launched. And investors are confident that it has a chance to be a worthy rival to Google and OpenAI. This week, it was valued at US$8 billion. At the beginning of the year, it was worth US$3 billion.
At Perplexity, a search also brings up a ready-made answer, sometimes with instantly generated infographics. If you think about it, it seems like personalized news for you, written by an AI in seconds, after “reading” several websites. Looking exclusively from the user’s side, it is something apparently superior.
But this AI-personalized web model assumes that we will have, for example, journalistic companies working for technology giants. SearchGPT or Perplexity need the raw material to give good answers. It turns out that, if the answer is satisfactory, it is possible — even likely — that the person will not visit the original source.
Given that a significant portion of traffic to news sites comes from Google searches, and this traffic generates money via impressions and ad clicks, what will happen to the web news ecosystem that depends on it? Or with the industry to make websites appear better on Google (so-called SEO)?
Nobody knows. Google, which seems to be aware that the winds are blowing in the direction of AI-assisted search, it is also increasingly investing in its version of SearchGPT, with summaries generated by Gemini. They prominently link to the source, and according to the companygenerate more clicks than normal searches.
But I get the impression that this is because Google’s “smart” summary is designed to be incomplete, almost a clickbait. Despite being willing to a certain level of cannibalization, Google still wants people to click on links and see other results. After all, it is a profitable business, which made the company grow.
It’s a safe bet it won’t be for long, however. Partly because the company – one of the tentacles of the Alphabet holding – no longer depends solely on searches. Cloud services are growing at a rapid pace — more than 30% in the last balance sheetand YouTube already generates more sales (in ads and subscriptions) than Netflix. If none of this works, the autonomous car arm it finally seems to be taking off.
So maybe it’s better to get used to this new way of searching on the internet. But don’t forget to check the primary source if the AI is not hallucinating. We thank you.