Who ruined Google? The answer points to Prabhagar Raghavan, the former head of advertising at Google. This man led an internal coup to seize control of Google search, ultimately leading to the site's downfall.

This statement is not groundless. The evidence comes from Google search itself: if you search "Who Ruined Google" on Google, the search results will pop up an AI-generated summary that quotes He wrote several articles, including one titled "The Man Who Killed Google Search."

It all started on May 14, 2024, when Google made a surprising decision: to fully introduce generative expressions in search results AI. This new feature, called Search Generative Experience (SGE), fundamentally changes the way users obtain information. On some search pages, traditional web links are replaced by AI-generated overviews. This AI will scrape the text content of other websites and automatically generate answers so that users do not have to visit the original website and of course will not generate any advertising revenue.

This approach itself is puzzling. It’s a well-established fact that generative AI is often wrong and of questionable reliability. However, Google seems to be completely ignoring this. In an interview* with The Verge reporter Nilay Patel this year, Google CEO Sundar Pichai even claimed that these changes would actually help the development of the Internet.

* Welcome to review this interview: "Failed AI Search, Is Google Dead?" 》

This statement obviously cannot withstand scrutiny. As Wired's Lauren Good pointed out in a recent article about Google's shift to AI, by choosing when and where to present these AI overviews, Google is effectively deciding what is a complex problem and what is not. . More importantly, they are deciding what web content should be included in their summaries and what users will ultimately see and learn. However, these analyzes are based on a very naive assumption: that Google really cares about building and maintaining a good search engine.

How absurd this assumption is is evident from the performance of Google AI. Just a few days before this article was written, if you asked on Google if there are any countries in Africa that begin with the letter “K”, you would be given this generative result: “As of September 2021, there are no countries in Africa that begin with the letter K” country.”The answer cites two sources: one is a 2021 forum post that quotes a hallucinatory answer from ChatGPT; the other is a website called "Countries starting with the letter K" whose first sentence is Kenya was mentioned.

Such errors are not isolated. Google's generative search results also suggested using white glue to prevent cheese from slipping on pizza, claiming a dog once played in the NBA. While these glaring errors were quickly corrected after prompting massive criticism, the very occurrence of these issues points to a serious problem.

These incorrect answers come from a modified version of Google’s Gemini AI, their counterpart to ChatGPT. Answers are generated based on web content, which can come from anywhere: news sites, random blogs, Reddit posts, you name it. As Liz Reid, Google's new head of search, puts it, it's "letting Google AI do the search for you." However, there are serious problems with this strategy.

Generative AI has a fundamental problem: "Illusion," that is, completely false information presented in an authoritative tone. This is because these models don't actually "know" anything. Whether it’s Google’s Gemini, ChatGPT, Anthropic’s Claude or Meta’s Llama, they all just predict the most likely correct answer based on training data. This means that Google’s AI-driven search is actually searching for information for you without any understanding or intellectual judgment, and without any real understanding of the content itself.

All the AI ​​can do is say: "Based on the math, these phrases seem to make up what might be the correct answer, and these links seem to contain those phrases, so I guess that's ok to answer." That's why Google's AI search once suggested people eat a stone a day - because it used content from satirical news site The Onion when generating the answer.

When The Verge reporter Kylie Robeson questioned these issues, Google’s response was shocking. They called these "very rare queries and not representative of the experience of most users." When Robeson used her own real-life experience as an example of Google telling her that "the mammal with the most bones is the python" (pythons are reptiles, although they do have bones), Google still stood by this stance.

However, behind these seemingly comical mistakes lies a larger crisis: Google’s AI-driven search is seriously damaging the entire Internet ecosystem. By summarizing links from other sites to generate answers, Google is effectively robbing the internetNet, randomly picking and choosing parts of the content it deems worthy of showing, and then providing a highly hallucinating rough summary.

It’s important to note that Google was already positioning itself as the gatekeeper of the Internet when it invented the PageRank algorithm. In the original PageRank paper, they used the word "importance" to describe how to rank web pages. The theoretical value exchange at that time was: we get reliable, secure results that actually answer our queries. But AI-powered search turns Google into a source of truth — and a deeply unreliable one. They are using a technology that is known to produce errors.

Some people may think that Google will be aware of this problem, and even if it is aware, it will not deliberately introduce such a hallucinogenic technology, but this idea is too naive. In an interview with Nilay Patel, Pichai explicitly admitted that they were fully aware of the hallucination problem, that it was still unresolved and that it was an inherent feature of large language models. On this point, Pichai is right. Inexplicably, however, the Google CEO then said that large language models are good for exactly this reason, a property that actually makes them very creative. This statement is confusing: "What are you talking about? I don't need Google to be creative, I need Google to answer my questions accurately!"

When you say that a large model is "creating When you say they are good at lying, you are actually saying that they are good at lying.

The dangers of this strategy are obvious. Generative AI could hallucinate or even authoritatively spread misinformation on important issues such as ways to deal with chemical fires, responses to mental health issues (and Google AI has already done this, only to have it quickly removed). Billions of regular users, those who are not tech-savvy and not active on social media, rely on Google to answer questions every day. They would take it for granted that a multi-trillion dollar tech company wouldn’t hand over the world’s most visited source of information to an unreliable AI.

Business Insider’s Peter Kafka offers a pithy metaphor for this situation. It's like "the difference between being given a map and being given instructions that will send your car off a cliff," he said. The analogy hits home: Google’s AI-driven search would be a disaster not just for the internet, but for society as a whole. By choosing which queries to summarize and content to pull from which websites, Google both centralizes and polarizes the world’s information while depriving actual content creators—the real humans who feed search engines—of traffic.

This problem is bad enough on its own. However, when Pichai told Nilay Patel (who barely questioned this): “No, it’s actually good for the web because people will get answers., and then they’ll click through to see more.” Let’s use social media to illustrate how wrong this statement is. Look at what people do when they see clickbait content: They click on the link To verify? No, they will just forward the content and express anger about it, rather than being curious. All users have already assumed that Google search will not be full of nonsense and will not spit out AI-generated and easy-to-generate information. Summary of Hallucinations - What else would I do with it?

To deeply understand the complexity of this issue, we need to listen to the industry Insider Voice The author interviewed Lily, an expert with 15 years of SEO experience. Ray, provides a unique perspective for this article.

Q: What is SEO?

Ray: Search engine optimization is the attempt to make a specific website and its pages more visible in search engines. The practice of being discoverable, specifically, is making your website visible on Google, Bing, or other targeted search engines. of the top positions because those are where most people will click.

Q: Does this practice manipulate search engines?

Ray: There are different methodologies. are Google's own guidelines on how to do it in a way that I think is actually good for the internet SEO. What I and professional SEO practitioners like me do every day is make websites more discoverable, more accessible, faster to load, and easier for people to find what they're looking for. There's a lot of technical work involved. . But of course, there's the other end of the spectrum, and that's people trying to exploit situations, spamming, and manipulating search engines. So it depends on what type of SEO you're talking about.

Q: SEO. What does a professional do on a day-to-day basis?

Ray: Companies come to us because they obviously want more visibility, so we have to consider all the different factors that impact how people find content. For example, a lot of our work is focused on page speed, overall accessibility, and if an image is uploaded to the site without image alt text (which is actually required for site accessibility), search Engines have traditionally been unable to understand content on the image. So we have to add the right descriptors. This is not only about technology, but also about content orientation. Sometimes people may write an article in a certain way. When people go to Google and search, they can never find this article, so we help companies build content in a way that allows people to actually find what they're looking for.

Q: How do search engines provide signals and. Guidance?

Ray: SEO The field is largely groping forward becauseA search engine is essentially a black box. We mostly use third-party data or take advantage of analytics tools provided by Google and other search engines where possible.

In terms of technology, Google's communication is quite sufficient. It's in their interest - to have a fast website, a website that's easy to access, a website that has a good user experience. Google holds many developer conferences, which provide a lot of communication on the technical level. For example, John Mueller often hosts webmaster video conferences at Google and answers many people's questions almost every month.

When it comes to content, however, the story is completely different. While Google does provide some guidance, such as a questionnaire that website owners can follow, and the so-called "Search Quality Assessment Guidelines," a 160+ page document that guides Google's human search quality assessment team in judging the quality of content bad), but in terms of specific questions like “Is this content good or bad for Google?” and “What does helpful content actually mean?” These questions have become increasingly difficult to answer over the past few years.

They (Google) can't tell us everything because people will use whatever information they can to create spam - so it almost becomes a war. But this vague guideline had an unintended consequence: search quality dropped significantly. There are a lot of different factors at play. The reason why SEO practitioners in general get a bad rap is that those who are best at creating spam have gained a lot of visibility lately. Particularly on the AI ​​side, a lot of people using crappy means really make it to the top. But there are also a lot of people doing really meaningful work that really makes the internet more accessible for everyone.

Q: What are the recent changes in the SEO field?

Ray: The biggest change may be the emergence of ChatGPT. It’s not just ChatGPT itself, but tools like it that enable spammers and SEO practitioners alike to find new ways to speed up their work. This works surprisingly well over a year, as it takes time for search engines to develop new algorithms and spam policies to deal with this kind of content. By the end of 2023, you will see a lot of AI automatically generating spam in search results, and people are using various new methods to accelerate their spam production.

Q: Can Google control search quality?

Ray: They've made it clear in this series of updates starting in March 2024 that they're not kidding. They do crack down on a lot of websites that use various SEO tactics and spammy tactics. So I do think they're regaining control. But I also think they may have gone a little too far in some areas and caused a lot of collateral damage. That's why you hear a lot of people say, "This destroyed my business," because there are a lot of people who have no idea they're getting involved.

Q: Why does it take Google, a multi-trillion-dollar company, so long to make changes?

Ray: One reason for this is that not all of these are easily discovered by algorithms. For example, if a brand starts working with third-party copywriters who provide product review content, these relationships aren’t always completely clear enough that the algorithm won’t necessarily recognize it. So Google has to use a variety of different methods to degrade this content and understand these commercial relationships. You can't solve every problem just through algorithms.

In the past few years, almost all large websites with strong authority, big brands with long SEO history on Google, have discovered new SEO opportunities, and these opportunities have brought them Amazing success. For example, you may have noticed that over the past few years, nearly every major brand on the Internet has dabbled in product reviews. This is because it is a huge source of income.

This phenomenon has caused widespread dissatisfaction. People complain that every time they search for keywords like “best running shoes for women,” they always see the same few big brands. Interestingly, in the past two or three months, Google has begun to crack down on this behavior in a large scale.

Q: But why did it take Google so long to figure out how to address this problem?

Ray: These brands have a lot of content that users really like. There are plenty of signals to Google that these sites are producing extremely useful content in many different categories. So it's not easy to say that this category of the site is bad for users and that category is good.

To further complicate matters, these are very large brands. You can’t remove or demote a brand from Google that people are looking for and expect to see. This explanation itself betrays a self-fulfilling prophecy: These sites are likely “popular” because of Google’s longstanding algorithmic preferences that have made them popular.

This problem is particularly obvious in actual searches. For example, what results would a search for "best laptops" yield? First it’s a tech site that relies on affiliate marketing for funding, then it’s a link to Best Buy, then it’s another affiliate site, then it’s Best Buy again, then it’s a series of clickable questions like “Which brand of laptops is the best?” ”

When clicking on the first question, it leads to a website called “New Indian Express Deals,” which is full of affiliate marketing links and devoid of any real journalism. The content may be entirely AI-generated and filled with links to MacBooks as well as laptops like the Techno Megabook and even the Honor Magic Book that you simply can’t buy in the United States.

This example highlights the role a search company should play: it should search the web for you and find things you might be interested inOr insightful content that might get overlooked in the sea of ​​pages on the internet. It should highlight independent publishers who produce quality content that might get lost in the din of the internet. However, the reality is that Google wants to highlight the same companies and the same publishers, doing the same things in the same way.

As in "The Man Who Ruined Google Search" ” article, Google’s revenue department now controls the company, and they demand more query volume – that is, the number of searches on the platform. Even if it makes the platform worse, they won't hesitate. Because when you optimize a platform to get people to spend more time on it and click on more content, you're not actually solving the query or problem. "I know they're looking for more queries, but they're not talking about solving queries, they just want more queries so they can show more ads."

When we look at Google In the current situation, the toxic effects of these decisions are clearly visible. Google Search will arbitrarily move its menu positions based on what you're searching for—features like shopping, images, news, and videos are constantly changing. Search results are littered with sponsored content, YouTube videos (YouTube is their own product, of course), and haphazard forum content. To make matters worse, they intentionally obscure the identification of paid links. The SEO industry has eaten away at it, and now it's hard to tell whether Google lacks the ability to fix it or just doesn't care that much.

However, there may be another reason for the poor state of Google search: Google deliberately distances itself from the search optimization community. While SEO is an area that indisputably hurts Google’s search results, it’s worth considering that beyond technical criteria (like how fast a site loads or whether it has a sitemap), Google is also deliberately hiding information from the SEO community. They fear that optimizers will use these criteria to manipulate search results. Ironically, however, this happens. Rather than clarifying what “good” content is, Google is vague: “Yes, you should do it, but I’m not going to tell you how.”

As a result, there are still people in the control system.

This strange phenomenon is puzzling: a multi-trillion-dollar company behaves like a startup when it comes to actual quality control. As Ray said: "They are worth trillions of dollars, and they definitely have enough people. I'm not a search ranking engineer, so I don't know why. I also think you know more about what's going on inside their company now than I do. I think they There's a lot of confusion internally right now. ”

When it comes to Google's recent handling of AI search, Ray is more critical. She explained that this feature has been tested in Google Labs for more than a year: "SinceIn theory, AI should be improving and learning every day. But over the past year, I and many others have been raising the alarm about quality issues. We also feel that it often doesn't provide much of an improvement over existing search results. For example, Google already has featured snippets. If they want to display a piece of text or an excerpt from someone's website directly in the search results, there are already mechanisms in place to do that.

Nevertheless, Ray still tries to maintain a certain optimism: "I want to recognize that this is a transitional period and talking about where it is now is not reflective of where it will be six months from now." In theory, these AI big language models should learn, improve, and keep getting better. ”

A fundamental problem currently facing Google’s AI search: once we know its accurate error rate, (whatever the actual error rate is, but based on the number we have seen over the past week , indeed quite a lot) Will we continue to trust it? What does this mean for Google's overall trust?

For featured snippets, you can say that the snippet results are only from the website. When displaying data, you can also blame that website. But with AI. The generated hallucinations, you can't tell where the specific wrong references are.

What particularly worries Ray is that the health topic seems to be one of the categories with the most AI hallucinations. This situation is particularly dangerous. Because the accuracy of health information is directly related to people's safety, when asked how Google should improve, she made several key suggestions: "I think they should re-evaluate. The concept of ‘helpful content’ because I think they’ve inadvertently caused a lot of sites that were really making experience-based helpful content to lose a lot of visibility over the past year. We still see a lot of big brands appearing for seemingly every possible query, and we don’t hear enough from smaller brands. We’re not giving small brands enough opportunities to have their content compete. ”

The core of the problem is that the trillion-dollar technology company with a monopoly on Internet information has shown weakness in the face of websites designed specifically to deceive it.

What’s even more frustrating is that the introduction of generative AI has made Google searches even more unreliable, yet Google’s CEO Still able to swagger into interviews and ramble about how he values ​​independent sources and more authentic voices, while his service continues to fail at delivering unique, interesting, or useful content

The Verge. That interview got a lot of praise for being a "tough" interview with Pichai, but it was just another opportunity for Google executives to blur their positions, allowing them to equivocate. To show that they care about search quality without actually taking responsibility, the reporter should prepare several specific examples of search results to show him on the spot.Look, ask him to explain why this is happening. If he could get on camera and say, "Actually, I think it's pretty good," that would say it all - it would either show that he's lying or that he doesn't care about the quality of the product.

Pichai’s cronies, such as Prabhagar Raghavan, now in charge of multiple divisions, and head of search Liz Reid, have been churning out shoddy software in the pursuit of sustainable growth. These people are part of a larger problem in the tech industry: Tech companies are no longer building products for customers, and both big tech and too many startups are creating some kind of token capital that investors can bet on, as in It’s the same as at the roulette table: “Maybe it’ll be red this time? No, maybe black, maybe double zero?” These products exist simply to show off astounding growth numbers—10x, 20x, 30x. returns that actually have nothing to do with the services they claim to provide.

In the not-too-distant past, the tech industry was exciting and fun. New product launches, whether subtle or dramatic, are changing our lives, making things better, helping people connect, or making our alone time more meaningful and substantive. The developments then were real progress.

However, today’s tech industry seems focused on launching products that no one is asking for, trying to solve problems that don’t exist, while also asking people to applaud a future that has not yet been realized. Cryptocurrencies, the metaverse, and now generative AI are all pseudo-religious movements built to sell products that fail to deliver on their promises, and they defend themselves by accusing those who disagree of being “not optimistic enough” when this The industry has repeatedly made people lose hope for the future.

However, the real dangers of this strategy go far beyond business failure. Regular users — the billions of people who are not tech-savvy and not active on social media — who rely on Google to answer their questions every day would take it for granted that a multi-trillion-dollar tech company wouldn’t make the world accessible hand over the largest source of information to an unreliable AI. We can laugh at the absurdity of Google advising people to eat a rock a day, but what if someone was searching for advice while trying to put out a chemical fire? What if a depressed person asked Google for help and the system recommended some weird, unproven treatment plan? Similar serious mistakes can happen at any time.

We must each remain vigilant and think critically until the next change occurs. In a world where even the most basic factual inquiry can yield wrong answers, we cannot blindly trust any single source of information, even if it comes from a trillion-dollar tech giant.