This week, Google announced a new feature that will roll out to all its search users over the next few months.
This new feature is the most impactful application of generative AI ever. It’s an even bigger deal than ChatGPT.
Google’s new AI-powered feature will change everything about how we use the Internet. And a lot of people are angry.
SGE Evolves
Although OpenAI gets most of the hype, Google has been in the generative AI space for far longer than the burgeoning startup.
As a gigantic company, Google initially played it safe with generative AI, building powerful systems like DeepMind, but refusing to release them to the public.
When OpenAI and ChatGPT rose to prominence, their obscene popularity basically forced Google’s hand. They needed to stay relevant, so they launched Gemini (nee Bard), their own Large Language Model (LLM) powered chatbot.
But they also began testing something much more powerful.
Initially dubbed Search Generative Experience (CGE), Google’s new product used generative AI to summarize web pages and present information right at the top of the search results page.
Instead of the classical “10 blue links”— or newer features like visual featured snippets or YouTube videos — Search Generative Experience directly answered users’ questions instead of sending them on a rabbit hole of research.
I beta-tested Search Generative Experience in its early days. While lots of people felt that it was too crude to be of much use to actual people, I immediately saw that it would change everything.
Why? SGE is a perfect example of the psychological concept of “satisficing”.
This cognitive psychology term refers to people’s tendency to seek out the fastest decent answer to a question, rather than the absolute best one. It’s a portmanteau of “sufficing” and “satisfying.”
If you enter a query into Google, the SGE result is often mediocre, at best.
But for many queries, it’s good enough that I expected people would use it instead of going through the 2 to 3 minutes of research required to click on multiple traditional web pages and get the best possible answer.
It appears I was right. Google says that people have used SGE “billions of times” just in its Beta phase. Maybe users don’t always love it. But their behavior suggests they seem to find it good enough.
And crucially, SGE is ultimately good for Google’s bottom line. Rather than sending users off their platform and onto external web pages, SGE keeps the user on Google, where Google can continue to serve them lucrative search ads.
SGE for Everyone
Google spent much of last year and early 2024 refining SGE through beta tests. Now, at their annual I/O event, they announced that it’s rolling out to the general public as AI Overviews.
When you enter a search into Google, you’ll now get an AI-generated summary for many queries, whether you want to or not.
It’s hard to overstate how big a deal this is for the way that people use the Internet. Google has more than a 90% share of the search market. That means they serve at least 1.5 trillion searches per year.
For trillions of questions people are asking on the Internet, Google is therefore the gatekeeper who determines the answer they receive.
With the rollout of AI Overviews, that gate is now manned by a generative AI Large Language Model.
If Google indeed rolls out AI Overview for everyone and keeps them in place, much of the information people receive on the Internet — which, let’s be honest, means most of the information people receive period— will be mediated or outright created by AI.
Google says that “this week, hundreds of millions of users will have access to AI Overviews, and we expect to bring them to over a billion people by the end of the year.”
That colossal reach makes AI Overviews the biggest and most impactful application of generative AI ever released.
Yes, ChatGPT is powerful and popular. Yes, it was the fastest-growing app in history and has an incredible user base, given how new OpenAI is.
But even with that success, ChatGPT still only brings generative AI to less than 200 million people.
With AI Overviews, Google is about to expose the entire world to the outputs of its generative AI systems, whether they like it or not.
The Backlash
Many people and institutions definitively don’t like it.
In the lead-up to the launch of AI Overviews, Google has rolled out several crushing updates to their algorithms that have demolished many independent blogs.
Many travel blogs, food blogs, local news sites, and more have lost 80 to 90% of their traffic since September of last year.
Google says that AI Overviews will include prominent links to publishers’ websites. But it remains to be seen how many people will actually click these links, versus just getting their answers directly from the AI and moving on.
Others have raised concerns about the accuracy of AI Overviews. Especially for sensitive medical queries or queries that relate to hazardous questions, mistakes could have dire consequences.
When one user asked the system how to clean their washing machine, AI Overviews reportedly told them to combine bleach and vinegar, which can potentially release deadly chlorine gas.
At least AI Overviews correctly says that the combo could be fatal!
There are also many concerns about the source of the data on which AI Overviews were trained.
Many people in the publishing space are asking how Google can legally train its AI on information from the same websites that the AI could potentially replace.
Answering these questions — and understanding AI Overviews’ broader impact — will take time.
The only thing that’s certain is that — with the launch of such a widely available AI feature — Google is unleashing a powerful new technology that will change business models, alter people’s default mode of accessing information, and ignite a firestorm of controversy and discussion around the benefits and limits of generative AI.
Will It Get Used?
For many creators, there’s still a tendency to think that maybe this is all just in passing.
Maybe users will hate the new AI Overviews, demand that they go away, and force Google to return to the classical 10 blue links.
As a web publisher, I would love that! But I also know that it will never happen.
Again, AI Overview may not be perfect, and experts can certainly nitpick and find problems with them. Over the next few years, lawsuits and closed-door licensing deals will sort out the legality of this kind of generative AI training.
But in the meantime, people will absolutely use AI Overviews because people have limited time, rarely care about the absolute accuracy of information, and are happy to go with something that quickly gets the job done, even if it’s imperfect.
Perhaps for deeper topics like researching a big trip or a toe fungus, people will be willing to click through to multiple travel blogs to find better answers than AI Overviews provide.
For many simple queries, though, they won’t.
What To Do
As creators, we need to understand this new reality.
If you create content that could be replaced by an AI Overview answer, it ultimately will be.
Only content that serves a different purpose — or that can’t be quickly replicated by generative AI — will survive in this new landscape.
As consumers, we also have to be aware of the fact that this is happening, and the potential implications for how we consume information.
AI Overviews are still a highly experimental feature. They may be rolling out to an audience of billions, but the path to becoming a mature technology will be long and circuitous.
Google may pull AI Overviews back temporarily if the system malfunctions, much as they did with parts of their Bard image generator.
They may face regulatory or legal scrutiny. There will be a lot of angry tweets.
Through all this, we need to remember that these AI Overviews are ultimately generated by machines — albeit disturbingly intelligent and human-like ones.
For some queries, it’s probably okay to accept the AI output. If I want to know the capital of Wyoming, I can probably rely on Google’s AI answer.
If I’m researching a medical issue or anything related to health and safety, though, I absolutely should not trust information from any AI system.
I should drill down and look for authoritative sources, consider multiple opinions, and seek out well-cited information to answer these kinds of life-and-death questions.
In short, the release of a powerful new feature like AI Overviews means we all need to get even better at media literacy.
With the flood of deep fakes and partisan news, we’ve all gotten pretty good at recognizing when a human is trying to scam or mislead us.
Now, we need to get better at recognizing the benign inaccuracies of a machine.
If you use Google (as all of us do) educate yourself about how AI Overviews work. Recognize the potential and the limitations of the technology, and decide for yourself when it’s okay to accept an AI answer, and when you need to keep scrolling.
Teach your kids, too. It’s never too young to begin a discussion about media literacy, and to help kids learn to spot misleading and inaccurate information online.
For now, there’s no easy way to switch AI Overviews off. Even if you’ve hidden under a rock and stayed away from any generative AI system so far, the tech is coming for you anyway.
It’s time to understand that it’s here, figure out what that means, and decide what we as consumers and creators want from this technology — and what we won’t accept.
I’ve tested thousands of ChatGPT prompts over the last year. As a full-time creator, there are a handful I come back to every day that fit with the ethical uses I mention in this article. I compiled them into a free guide, 7 Enormously Useful ChatGPT Prompts For Creators. Grab a copy today!
This story originally appeared in The Generator.
