ChatGPT will kill ad-driven websites

and that's a good thing

How do you get your information?

For almost 25 years, Google has dominated the way we find information. The last few years have seen cracks. How often do you look beyond the first page of search results? How often do you even look beyond the first few search results?

What do you find when you go to most sites linked from Google’s search results? More often than not, it’s a site plastered with advertising and written in SEO-friendly language that’s optimized for higher rankings in the search engine. More often than not, I tend to click away from those sites as soon as I land on those ad-infested pages.

Chat as Interface

On first thought, the last thing that most of us would want to do is interact with a chat bot as we search the net. We think of one of those slow chatbots that lurk in the corner of so many websites that might or might not have a real person in the background slowly typing responses back at us.

If you’re somewhat following tech conversations on the net, you’ve encountered ChatGPT that is, essentially, a chat interface to the Internet. The experience is remarkable. ChatGPT upends the search-and-retrieval paradigm through a conversational approach that mimics talking with an extremely knowledgeable assistant who can help you in all sorts of ways.

The chatbot destroying every search engine in its path (image via Midjourney)

ChatGPT is based on a large language model trained on content scraped from the Internet. Unlike Google Search, ChatGPT is not a real-time search engine. ChatGPT’s AI model has no content published after 2021. For many purposes, that’s not a problem. But we all know the dubious nature of the Internet with its content farms and then there are all the forums with very opinionated, unfiltered talk.

Can we trust chat AI?

Not always. But that’s the same with Google search results. There’s a lot of hand wringing about the trustworthiness of ChatGPT. But we have the same problem in the real world: can we trust the news media? Can we trust teachers? Can we trust ministers and priests? We certainly cannot trust politicians. Whom do we trust? Do you trust what you read in Reddit?

In the last few years, we’ve seen a lot of angst about misinformation. And we’ve seen a lot of public actions that have been linked back to misinformation.

Evaluating the credibility of information is a skill. And we need to put as much emphasis into developing that skill in individuals as we do in fighting institutions and individuals that promote misinformation.

So, why is AI any less trustworthy?

Because we think algorithms can fix the trust problem. Will AI-based information systems provide probabilities about trustworthiness? Can we trust those systems if the models are not open to us for inspection?

The race for chat domination

We’ve seen ChatGPT as a proof-of-concept. How well will it scale? That’s an engineering problem that can be solved. ChatGPT comes to us from OpenAI, which is a deceptive name. OpenAI has a business partnership with Microsoft. When will Microsoft release the re-imagined version of Bing as a chat interface to the Internet?

And what about Google? Can they survive this innovator’s dilemma and transition to a chat-oriented interface. We know that Google has invested heavily into machine learning, and AI already provides a significant foundation to Google’s searching algorithms.

I have little confidence in Google committing to a new product. Google does search well; they do email okay. YouTube is clearly Google’s bes product, though it’s a product they purchased. Incidentally, YouTube is also the second most popular search engine. Indeed, in the last year, I’ve found myself going to YouTube more often than Google when needing to learn how to do something.

For those of us who have been around for a while, we will remember Google killing off Google Reader, which is directly tied to the demise of blogs.

How is Google working for you lately?

ChatGPT will become shorthand for how we refer to AI using large language models applied to the Internet, even if the ultimate victor is Google, Microsoft, OpenAI or some other company.

Creating content for AI

Trustworthy content on the Internet is not going to get better over the next decade. Where will future versions of AI get its content? Will large language models continue to train AI on material scraped from the Internet?

ChatGPT will kill SEO when chat becomes the dominant information retrieval method on the Internet.

Then people can return to writing quality content on the Internet because they want to do so rather than making a few hopeful bucks from ads. But getting compensated for one’s writings is not a bad thing. We’re seeing mechanisms like paid newsletters, podcasts and subscription sites stepping up to replace ad networks.

And the YouTube model works well. I admit that I pay for the monthly premium YouTube subscription to avoid the ads. And that monthly price makes a huge difference in improving my experience of YouTube.

Scenarios for the next few years: I believe that there will be multiple types of ChatGPT specializing in different types of topics. That allows the language model to be finetuned for a specific data set.

Part of that finetuning could be a different type of content farms. Rather than sites hiring writers to produce search-friendly content, writers and editors are hired to write quality content that targets the AI models.

Or, a likely scenario is that the providers of chat AI services will hire writers and editors to directly prepare responses that finetune a model. In this scenario, the content only exists within the chat AI service.

The business model of chat AI

How does the chat product make money? Ads? Subscription. Most likely a combination of ads and subscriptions similar to YouTube’s business model. Even the simple existing UI of ChatGPT would easily allow for incorporating unobtrusive ads on the side of the page or elsewhere. Then the option for a subscription to remove the ads or a subscription to remove a limit on the amount of searching you could do per month.

A major concern would be the insertion of sponsored content into chat responses. That will invariably happen but labeling of such sponsored content must be clear.

Chat, GPT, and every other large language model will impact how we organize knowledge and retrieve information.