Will AI eat the internet?
Online search has changed in the ChatGPT era. Instead of using Google, workers can query AI as they would a colleague at the next desk and get the answer without having to even open another browser tab.
To some, the danger is that as we look at less original web content rather than content AI references, we’ll have no reason to create original web content anymore.
Critics argue that as sources and links gradually degrade, there will be nothing left but AI models endlessly regurgitating each other’s content, steadily decreasing in value because of amplified inaccuracies and biases. It’s been called everything from the dead internet to the death of Google, and it’s even been studied mathematically as ‘model collapse’.
If you think the word ‘extinct’ is a bit drastic, consider this. A 2024 Cornell study suggested if LLMs keep developing at the current rate they’ll run out of human-generated data between 2026 and 2032. As far back as 2022, Gartner was reporting that synthetic data generated by algorithms to closely mimic real-world data would be the main form of information AI uses by 2030.
And when even the search companies are in on the act, what’s going to stop the decline? Alphabet CEO Sundar Pichai told 2024’s Google I/O conference the company would move away from lists of links to a summary response to a query, the cornerstone of its AI Overviews initiative.
But Bharath Thota, Kearney Partner, says social media will thrive driven by our inherent curiosity about what others are doing and saying, and companies will still maintain a web presence for communication and branding. “Overall, we’re moving from a web designed for browsing to one shaped around intelligent systems acting on our behalf.”
Shoring up sources of knowledge
There are already movements and practices emerging to stem the tide. One simple fix might be to have AI get better at presenting sources, leaning on techniques such as retrieval augmented generation (RAG) and web rephrase augmented pre-training (WRAP). Julian Wiffen, chief of AI and data science at Matillion, thinks the notion of citing sources and peer review like we do in scientific publishing might prosper.
“AI could weight sources used in training … to support responses based on a calculated trustworthiness score,” he says. “I see that evolving the same way a search engine’s search algorithm weights its results rankings.”
Human-generated content also isn’t going anywhere anytime soon. Nikola Baldikov, founder of InBound Blogging, says because technologies like social media are ever popular, AIs will continue to have plenty of publicly available training data, something he calls ‘genius in and of itself because people will never stop communicating with each other’.
Then there’s the idea AI might actually help humanity generate more content. “[AI] is another shift that lowers the barrier to contribution,” says Roger Williams, partnerships & community manager at Kinsta. “If we treat AI as a collaborator and not a replacement we don’t lose the web, we reinvent it.”
Manuj Aggarwal, Founder and CIO, TetraNoodle Technologies, thinks it might even boost other types of human-made media that are truly original (so far). “Content like podcasts may become more popular and valuable,” he says.
We also can’t forget the platforms that stand to lose if the internet eats itself are very smart and very well resourced. Many search platforms detect and demote low-quality content. “Google regularly ‘punishes’ AI-generated spam websites, making sure search results shows trustworthy, verified material,” says Baldikov.
Nor do AI companies themselves want human-generated content to end, because it’s what their products are trained on. “As long as human insight remains part of the process, we can expect innovative solutions,” Baldikov adds.
Some people, like SEO expert and AI engineer Vincent Schmalbach, think that will lead to a market correction that will make human-generated content even more valuable. “Organizations with strict editorial standards and original research will thrive,” he says.
Thota also sees a future where content created by humans is locked behind paywalls, only accessible through subscriptions or private communities. “As a result, traditional web search may become less useful for genuine human insight or creativity since that kind of content won’t be indexed or available in the open web any more.”
Above all, Aggarwal points out AI hasn’t scoured the internet for knowledge to the extent search engines have, which puts a premium on human-made content. “We’ve only indexed about two percent of the internet,” he says. “There’s still a treasure trove of unexplored content – ancient texts, offline books, private databases. Search … may have a larger reach, so it’ll be relevant in the foreseeable future.”
Is the web approaching a reckoning?
So where will it all end? We’ve already seen the statistics about AI running out of original data, but when will catastrophic model collapse render the entire web unusable?
Derren Nisbet, CEO at AI testing company Virtuoso, refers to a National Bureau of Economic Research study saying 39% of Americans use AI and 89% use Google every day. That puts Google far out in front, of course – except that it’s been going since 1998, whereas, as Nisbet says; “ChatGPT was launched in November 2022 and here we are in early 2025 seeing it used by over a third of the US. The period in which the AI companies have made progress is astonishing.”
A recent report by the management consulting firm Bain & Co report mentioned above also says 80% of consumers now rely on AI-written results for at least 40% of their searches, reducing organic web traffic by 15-25%. It adds that 60% of searches now terminate without the users clicking through to another website.
“Customers will engage along two paths – direct to the brand website for loyal, returning, word of mouth customers or intermediated through AI,” says Natasha Sommerfeld, partner at Bain & Company’s Technology, Media & Telecommunications practice. “There’s still a role for the website, but we believe the ‘browsing’ and ‘comparison’ steps are going to be more quickly disintermediated as AI agents learn preferences and can digest far more information and return recommendations. [Websites] playing both roles – effective database and personalised experience engine – will remain important in the near and medium-term.”
So after Web 2.0, cloud computing, the mobile revolution and big data, AI is indeed the latest paradigmatic shift on the internet. But the jury is still out on if it could be the revolution that kills it off altogether.
Source link
