I don't need AI to summarise a news article I'm already reading. I don't need an AI to help me switch phone plans. I don't need AI when I'm searching up an actor's name - I just want a list of their films! In fact, there are remarkably few situations in which I truly need a user-facing, conversational AI.
Let's be clear; AI is an incredibly powerful and useful tool - behind the scenes. I am not a luddite by any means. I recognise there is reason for the momentum behind the recent and sudden upward trend of AI. But please, for the sake of our humanity, keep it behind the scenes. Sophisticated recommendation algorithms, supply chain analysis, fraud detection - AI excels at helping humans at complicated analysis tasks. But this functionality is a world away from most people's experience with AI nowadays.
TOBi, Vodafone's AI helper
Nothing annoys me more than those 'helpful' AI assistants on websites nowadays which are actually just glorified ChatGPT wrappers with almost no ability to answer my question, much less act upon it. There's no hope of me trying to tell the Vodafone bot that I want to switch plans - I just clumsily dismiss the pop-up window and try to find the menu on their website like I used to.
DPD's chatbot, via @codfish246 on Twitter
In cases where I am forced to interact with this powerless AI, I will resort to every tactic in the book to get connected to a real human, or get rid of it as soon as possible. If it genuinely is just a badly-made wrapper of an LLM, you can often get connected to a human with some sneaky prompt engineering.
The user finally being connected to a human
That said, simple functionality without these AIs on websites seems to be harder than ever to find, in this new era of minimalist settings pages and navigation bars.
Please, Bring Back Documentation
A true functional documentation seems to be a thing of the past. Forget the AI chatbots - it would be so much better for the sanity of both your developers and your users to instead build a comprehensive documentation and provide a search function. If you have to force AI into your app to please your VC funders, make it an "intelligent" search, to make it even easier for me to find the right walkthrough document. But for goodness sake, please have documentation on your site. And in no cases should the only way to access a feature be through your vibe-coded, last-minute-thought AI assistant which you've named "Stanley" or "Gemma", something attempting to inject a synthetic personality into your company.
Ironically, I'm currently building a webapp (CashCat), but as always, I'm trying my best to follow all my own rules, and in fashion, I have prioritised a large set of documentation pages publicly available for every possible action in the app I can imagine a user doing. And there's no useless chatbot to talk your ear off. Because why would there be? What possible benefits could one have?
Consider, for one, the developer time and financial resources poured into building these AI chatbots. You can either resort to a low-quality outsourcing and produce simply a wrapper of a popular service like ChatGPT, or you can train and perpetually refine your very own chatbot. This isn't a one-and-done project; it requires continuous data feeding, error correction, and model updates. Compare this to investment in comprehensive documentation - which, while demanding an initial investment in terms of writing, provides a far more stable, predictable, and long-term solution.
Not only that, but documentation possesses inherent scalability that those AI chatbots often lack. A single, well-written article can be instantly read by millions of users, providing each of them with a solution to their problem that is guaranteed to be correct (because you wrote it). Whereas, the computational cost and resource consumption per interaction with your 'chatbot' does not scale well. The chatbot's answer to each user's query, even if they asked the exact same question, could be subtly different, and cause inconsistent user experiences and unexpected problems down the line. They also have a tendency, unless you have trained them exceptionally well, to falter on complex issues, which will frequently force users back into the human support queue they were initially designed to replace. They are expensive to host, can go down for technical reasons, and can be difficult to use on various devices and browsers.
Indeed, the reason behind all of these AI chatbots isn't genuine user need or operational efficiency - it's simply the illusion of innovation. As more and more companies adopt AI, product managers and research analysts feel 'left behind' if their company is seen as behind the times. They direct their developers to adopt AI, to get with the modern world - and so Stanley the AI is born. But no users ever asked for Stanley, and no users ever prefer Stanley.
I think it's absolutely time for a re-evaluation of our priorities as a society. We really don't need to pour more energy, money, and development time into what is essentially a solved problem, and has been for many many years. Documentation, in its purest form, is an extremely powerful tool to provide to users - treat us like literate adults, please. Not only that, but it genuinely will bolster our reading and comprehension ability. Reducing everything to AI summaries and information retrieval will slowly chip away at our comprehension level until we rely on them irreversibly. In the book Nineteen Eighty-Four, George Orwell writes about this exact situation - books being slowly reduced to summaries, which slowly collapses the population's literacy and mobility.
They Bring No Value - But They Might Take Some
It's increasingly well-known that LLMs continuously train on user data during conversations. Most companies make vague statements of denial in relation to this, mentioning that their specific AI does not use sensitive customer data in its training set. That's a good start; but it's not rock-solid. The data is stored somewhere, temporary or not, and everything suddenly becomes more vulnerable and valuable as a target for attacks and data breaches. This is simply not a problem with a documentation portal.
Beyond the practical downsides I just mentioned, the consistent frustration with ineffective AI chatbots also, in my view, damages a company's brand reputation. I, for one, think less of Vodafone after experiencing their chatbot - do they not value us as customers enough to provide us with human support? They are leaving us with the scraps of a poorly-trained LLM which has barely any ability to do anything except for convince you to purchase one of their upgraded products, just as it is instructed to. Users quickly develop resentment for systems that waste their time and feel dismissive of their actual needs.
A well-structured and easily searchable documentation portal can reduce support needs, increase user engagement, and result in a more educated and independent user base.
An example of a knowledge base (Userpilot)
So please, let's bring back the art of documentation. Let's get rid of Stanley and Gemma, and bring back that beautiful knowledge base, providing me everything I could ever want to know about your app. It'll be better for everyone, I promise.