The introduction of AI means how we interact with the internet is poised to undergo the single most significant shift since the decline of the AOL web portal and explosion of new web browsers. And no, I’m not talking about [the death of web cookies], although that will have its own ramifications. Instead, I’m referring to one of the core paradigms of the internet experience: search.
The underlying mechanics of search are relatively straightforward, and have remained unchanged since 1993 with the launch of Jumpstation. A web crawler goes around the internet, indexes the pages, and then allows users to search through that database to find what they’re looking for. Back then, search engines ranked pages based on how many times the user’s search phrase appeared on a webpage, but that approach was upended with the creation of Google.
Google’s rise in prominence
Google, officially launched in 1998, quickly rose to prominence due to their innovative approach to algorithmic search. Rather than simply listing all the relevant pages to a search term, Google used the number of backlinks a page had to assess its quality, and that value would then determine how high it appeared in a search result.
Google’s popularity really exploded when, in 2000, they became the client search engine for Yahoo, at the time one of the most popular sites on the internet.
The same year, Google launched AdWords, and by 2002 it had become a money-making machine – it turns out advertisers loved being able to place ads in search results. The pay-per-click advertising model, coupled with Google’s automated bidding and ad performance tracking, made campaign management easier and more efficient.
By harnessing the power of advertising, Google became an absolute powerhouse. The revenue they generated from advertisement contributed to their successful IPO in 2004, and afforded them the capital to then further invest in their technology, and go on a purchasing spree.
People might look at YouTube or Android as the two most important purchases to Google, but I would argue that it was in fact DoubleClick in 2007. DoubleClick was the missing piece to Google’s advertising business, with a focus on display and video ads, and a thriving ad marketplace with pre-existing relationships with big publishers. This acquisition gave Google control of the whole ad tech stack – owning both the leading publisher ad server and the leading advertiser ad servers.
The economics of search
Search is a very profitable business for Google. In 2022, search ads accounted for $162.45bn, or 58.1% of Google’s total revenue. Search is so valuable to Google, it even pays Apple 36% of all of its search revenue from Safari in exchange for being the default search engine on Apple devices – a deal expected to cost over $18bn a year. Google also happens to control over 90% of the search business.
From this, we can make two assertions:
Search is very important to Google
Google is first and foremost an advertising company
Ok, so search is profitable, but what does this have to do with AI?
The inevitable quality decline of any service
To explain my scepticism about the application of AI to search, I refer to the term 'enshittification', wonderfully coined by blogger and author Cory Doctorow.
Enshittification, or platform decay, is the declining quality of a digital platform that acts as a two-sided market as it tries to exploit both sides. Doctorow has highlighted this decline in a number of services, including most social media platforms, Reddit, Amazon, and Google. The commonality: advertisement.
We see it time and again. A service starts off free for users, then advertisement is introduced, monetising your attention to businesses. Then, once they’re entrenched, the platforms start to squeeze advertisers as competition increases on the platform, while the user experience simultaneously gets worse. The inevitable end is a messy interface full of adverts because the platform has tried to maximise profitability from every available surface.
The prospect of ‘normal people’ using anything other than Google would have been largely unthinkable 10 years ago. Bing was typically viewed as a punchline, and DuckDuckGo was just for the privacy wonks. Now the search experience has become so horribly cluttered that we are actually seeing paid search services like Kagi begin to gain traction in certain online circles.
That should be a warning.
The harms of profit-driven behaviour
Google already exerts so much control over the web, and how website owners structure and present their data. The SEO industry was reportedly worth $75bn in 2023, and set to grow further in 2024. That’s $75bn on trying to get your website to rank highly on search engine result pages (SERP), and doesn’t even factor in paid advertisement.
The web has undoubtedly gotten worse because of SEO, and to see that encapsulated you only need to look at any recipe website. Who cares if this was the soup your grandmother made when you were 6 and off school sick with tonsillitis? Nobody, but SEO and time-on-page metrics has now incentivised keyword stuffing. Generative AI is only making the situation worse, allowing publishers to fill their sites quickly and cheaply with SEO-rich text that says nothing original, but performs well. Fake news sites are being created with this technology. Worse, we’ve already seen examples of feedback loops where generated text with factual inaccuracies begin to perform well online, and then get scraped by other AI models, until it becomes fully detached from reality.
What changes are websites going to implement to be more AI friendly, and is it going to really benefit the end user? At this time, we don’t know, but I’m yet to see anything that convinces me the experience of browsing the web is going to improve.
If anything, I fear it might actually get worse.
Where’s the money, Lebowsky?
OpenAI, Perplexity.ai and Google are all trying to create the next generation of search. They believe it should be multimodal, and browse the web for you to find the answer and present it on page. That’s potentially great for the user, but what about the websites being scrapped for their data?
Good content takes time and talent to create, and there needs to be financial remuneration for that effort. Adverts are the principal economic driver for the internet. Everything from news sites, to social media, and now even the streaming platforms (that you already pay money to) consider advertising as their primary means of making money. Subscriptions can work, but without a radical shift in user behaviour, it’s unlikely that putting a paywall in front of services will ever bring the returns that adverts can. But, if people aren’t visiting websites because they no longer need to, where is the incentive to create unique, well-researched, and compelling content?
Note also the word ‘potentially’ in my assessment of the user experience. By disincentivizing the free, ad-supported model of distribution, we risk creating tiers of internet citizenship, where good information is paywalled, and generative copy is then regurgitated to those who are unwilling or unable to pay for it.
And, returning to Cory Doctorow’s enshittification, do we actually want to cede even more control of our browsing experience to big companies who have every incentive to create a sticky product that can then prioritise their own services and profits? Do we want Google, Microsoft, or a Jeff Bezos-backed Perplexity.ai deciding what it is we see online, rather than just presenting us with the options that we can pick from ourselves? Isn’t this algorithmic approach to content delivery exactly what turned social media into cesspools of misinformation and teenage depression?
Certainly, early signs indicate that concerns around the deterioration of search results seem founded. Following an exceptionally AI-heavy I/O developer conference, Google announced that they are ending their Search Generative Experience (SGE) experiment and rolling out the finished product, known as AI Overviews, to users in the US. Social media has been awash with examples of how badly the product is failing, including stating that snakes have the most bones of any mammal. Worse, it has recommended adding glue to pizza to make cheese more stringy. Literally a week in and people are being advised to eat glue. AI search simply doesn't have the guardrails in place to protect from the trolling of the internet, and when to date it's been viewed as the de facto authoritative source of information, users aren't prepared for this catastrophic failure.
AI is not cheap to develop, and none of the big companies have found a way to successfully derive a profitable return on their investment yet. Investors, whether public or private, will only tolerate losses for so long, and will eventually expect to see these projects pay off – and when that pressure mounts, you better believe companies will enshittify the service to placate the stakeholders that really matter.
So is AI search a bust?
Believe it or not, I am not an AI sceptic. I do believe that both will have genuine utility for a number of people in varied use cases. AI driven search engines will better understand intent and meaning, which allows a user to more quickly find the information they need – even when they aren’t quite sure what it is they’re searching for. Multimodality means users can use their voice, images or even videos to search, offering both flexibility and granularity that will significantly boost productivity. I am also hopeful that AI will make the internet more accessible, giving more freedom to people with disabilities to engage with the internet on their terms.
There’s also the potential to better educate users around news and web literacy. As a society, we need to develop the skills to critically evaluate the information we consume online. That should start in schools, but technology companies will have a role to play too.
Search engines could contextualise generated answers with quality assured sources directly linked in a manner that encourages people to click-through. They could use AI to identify and filter out bad actors, deranking fake news and misinformation to ensure only true and valuable information is discovered. Using browsing habits they could even build a profile of our biases, and try to offer sources that take us out of our own echo chambers, reversing the downward spiral of political discourse.
Shopping online will be easier through AI powered search; users will be able to upload a photo or video of a product and quickly find it online. In fact, TikTok has already started trailing AI product identification that pushes users to its shop. But that also poses the biggest risk to the experience. Search providers need to resist the temptation to immediately flood users with products to buy, else they risk annoying potential buyers. TikTok are already learning this lesson the hard way.
Given how new this technology is, whether it ends up being a net good or net negative is yet to be seen. There is still much unknown about how it will be implemented, and more importantly how it will be monetised. Its development is inevitable, so it is on us to participate in it to help shape it in a way that isn’t harmful for users, or the internet as a whole.
