ChatGPT officially launched its search feature to paid subscribers on 31st October 2024. Given the timing, I expected a wave of write-ups, videos, and excitement about it. Yet, even after a week — despite the significant launch and the US elections happening in between — there has been surprisingly little buzz.

During this quiet period, I tested, used, and overused ChatGPT Search — and, to be honest, I am not impressed in the slightest.

Familiar Territory for Some Users

This feature is, in a way, not new for paid subscribers. If you were already familiar with the "/" backslash hack, which allowed users to perform searches. For those who thought this might rival traditional search engines like Google, Bing, or DuckDuckGo or others, I have some bad news: these generative AI tools are not just behind — they are nowhere close.

Whether it's ChatGPT, or others (e.g., Perplexity etc.) , they all fall short on the most fundamental principle of search: the ability to effectively find relevant information.

Understanding the Search

Search is an extremely simple concept: Finding Relevant Information. While these three words might seem straightforward, they contain a lot of depth and nuance. Let me make my case first before I talk about the limitations of Generative AI search solutions :

Finding relevant information is key to effective searching. It involves locating the right sources, with the right data, in the right (your) context. Search engines are influenced by factors like SEO (Search Engine Optimisation), which allows businesses to control their visibility. Human users can differentiate between genuine content and paid promotions, a nuance that AI-powered searches currently struggle to replicate effectively.

When you search, you have the ability to choose the sources you want to explore, determine what's relevant, capture what you need, and iterate. Sometimes you adjust your search when the initial results aren't satisfactory. It's a simple yet powerful human process — a unique human algorithm of sorts.

The Limitations of Generative AI in Search

Generative AI, despite its aspirations to achieve AGI (Artificial General Intelligence), can't replicate this process — at least not yet. AGI aims to mimic general human cognitive abilities, but effective search involves nuanced decision-making, context evaluation, and adaptability — areas where AI still falls short compared to human intuition and judgement. What AI-powered search solutions like ChatGPT Search do is take your prompt, search on your behalf, scan the content of the first few results (perhaps even clicking on paid links without you knowing), summarise it, and then present the results. While this approach may seem logical, it is fundamentally flawed.

You lose control over selecting sources, clicking on paid results, and determining how much information to extract. Yes, ChatGPT Search gives you a list of sources it found (though not necessarily consulted), but you have very little control beyond that.

Moreover, many searches are simple and straightforward. Often, you're just checking the weather or looking up your favourite band's concert dates — things you can easily find on dedicated apps or platforms. In contrast, more complex searches — such as understanding the impact of recent economic policies or comparing different medical treatments — require human judgement to evaluate the credibility of multiple sources and synthesise nuanced information. You rarely need Google or ChatGPT for those. In fact, a large proportion of top Google searches are brand names, such as YouTube or Amazon. Are you really going to use ChatGPT to search for those?

The Content Conundrum

Thanks to the capabilities of Generative AI, everyone is now a blogger or YouTuber, and the amount of content online is growing rapidly — much of it generated by AI. The result? Less reliable, less relevant information. Imagine: a machine generates content, and then that same machine searches through it. It's like writing your own notes, only to search through them later, hoping to find something new. This cyclical process lacks the critical discernment that a human brings, creating an echo chamber of potentially unreliable information. Without the ability to evaluate context and verify credibility, this loop perpetuates errors and undermines the reliability of the results compared to human judgement. With so much AI-generated content out there, the internet is becoming even more cluttered — and no one is taking responsibility for it.

In a nutshell

You wouldn't use ChatGPT Search for simple, everyday information. When it comes to research, where accuracy and context are crucial, the ability to control the sources and determine relevance becomes even more important. You'd be better off manually finding reliable sources and then using ChatGPT to summarise them. That's what it's best for — making sense of well-chosen material.

Some people argue that the key is writing the 'right prompt' to get good information. However, not everyone is going to become an expert in crafting perfect prompts (adding a link below to my other post related to this). And let's not forget — neither ChatGPT nor other tools are free if you want anything truly useful. You must subscribe to their paid plans.

There's also the question of whether these tools are reading from paid links. If so, we might be unknowingly 'clicking' on paid links, which in turn skews the results and makes search even less trustworthy. Paid links often prioritise advertisers' interests over genuine relevance, leading to biased information that can mislead users rather than provide objective insights. Companies paying for top results could gain more traffic — and your data — and target you with ads across the internet.

Where Does This Leave Us?

I'm not here to argue for or against generative AI search solutions. I'm trying to make a larger point: we need more control over how we conduct our searches. Even if we decide to outsource the process to a machine, we should be able to set boundaries and manage the aspects that matter to us.

I use ChatGPT heavily for many things, but I will not use it for search until I have better control over how it works.

What do you think? I'd love to hear your thoughts — whether you agree, disagree, or have your own experiences with these tools. Feel free to share your perspective in the comments.