ChatGPT has become a competitor to Google search, but this is on the scale of the entire internet, where there is only one top spot – hence, entering this race might seem futile. However, there’s a vast number of websites with tons of content, each of which wouldn’t mind having its own version of ChatGPT for its content. And this opens opportunities for many! And this thing could look something like this.
Project Essence
Direqt has created a platform where an online publication (magazine, blog) can create a chatbot that their readers can interact with.
The essence of the chatbot is that readers communicate with an AI trained on the publication’s content. The chatbot window can be inserted anywhere on the site – even in the middle of an article, so that the reader can ask questions about the published material and its characters.
For instance, while reading an article about David Beckham, one could ask when David first met his future wife Victoria, or how many children they have.
Or, one can simply ask the chatbot what the most interesting current news is in the field of technology or any more specific topic that the publication covers – and then the chatbot will provide links to the latest popular articles on that topic.
To train the chatbot on the publication’s content, all published materials must first be uploaded to it, and then new articles must be timely added. Currently, this is done by informing the chatbot of the publication’s RSS feed address, from which it pulls links to publications.
This is not the most convenient method, but it is somewhat universal.
By learning from the publication’s content, the chatbot also adopts the publication’s style of speech. Moreover, it can be further trained in a specific communication style, given a set of instructions on how to interact with readers, what phrases to use, and which ones to avoid.
The publication can choose one of two ways to pay for the platform. The first way is to pay a certain amount each month, depending on the number of messages processed by the chatbot. The second is to agree to have ads from the startup’s partners displayed in the chatbot window and pay nothing, even earning a share of the money the startup received for placing ads in the publication’s chatbot.
Chatbots created on the Direqt platform have already been placed on their websites by 75 publications, including Vogue, GQ, Wired, ESPN, Men’s Health, The Sun. Over the last 30 days, readers of client publications have exchanged more than 10 million messages with publication chatbots.
Direqt was founded in 2017, but initially, they developed regular chatbots for publications – albeit with varying success 😉 They decided to create an intelligent bot based on GPT in 2021, for which they raised $1.4 million in 2022 and $4.5 million now.
WHAT’S INTERESTING
An intelligent chatbot capable of conversing with publication readers is, of course, an interesting technological gadget. But the question arises – “why”? 😉
It turns out that such a chatbot improves one of the most important metrics of an online publication – the time readers spend on its site. One of Direqt’s clients reports that after installing the chatbot, readers began spending three times more time on their site, asking the chatbot various additional questions about the read content.
If we look at the situation more broadly, the chatbot can be called one of the first swallows of a more powerful paradigm shift in online publications – the creation of interactive content.
The problem for any author is that they always have to balance between the brevity and completeness of the article they create.
An article is not a book, which is initially intended for the most comprehensive disclosure of the material. Thus, an article cannot be very long. Especially now, in the era of the flourishing of “clip thinking,” when most people can no longer (or do not want to) maintain attention on something for a long time. Therefore, articles need to be made as short as possible.
On the other hand, if the reader is interested in the topic of the article, they want to know and understand more details and specifics – without this, they will consider the article bad because the “topic is not disclosed.” At the same time, these details and specifics will strain all the others, for whom the article with such details and specifics will seem too long and complicated.
As a result, authors have to create “average” articles in length and depth, which in the end do not particularly satisfy either one 🙁 To those who want shorter, the average article will seem too long. And to those who want more details – too superficial.
One of the solutions to this problem may just be the concept of interactive content. When an interested reader can ask the article (the chatbot sitting under the hood of the publication) to tell them more details, focusing only on the aspect that interests them. And everyone else will read the short material, be satisfied with it, and move on.
By the way, if the publication has already trained AI on all its content, this can give readers another useful tool, which today’s Direqt may not even have thought of yet.
But another startup, Objective, thought of this and raised $18 million in its first funding round last month. They offer websites to install their developed “AI search.”
The problem with “regular” search engines is that they look for texts containing words entered by the user. But what if the user doesn’t remember the exact phrases from an article they once read, which they suddenly decided to find – “something like… kind of…”? And what if they just want to find articles on a topic whose name doesn’t necessarily have to appear in the article’s text?
Objective helps find texts not by exact word matches – but by meaning, which can be expressed by different words and scattered across different parts of the text. This became possible because the Objective AI machine first learned the content of the site, and after that, it became able to answer questions about this content – not as a chatbot but as a search result.
By the way, Objective can search in the same smart way not only for texts but also for images – for this, it is enough to describe in words what should be depicted on them.
In terms of search capabilities for images, this is similar to the platform Kive. Kive helps creatives maintain a catalog of photographs and other visual content. The platform itself marks objects in pictures along with details of their placement, allowing one to find in the catalog, for example, “photos of adult men with a sad face, made in the style of ‘noir,’ in which the model’s head is slightly turned to the right and raised slightly upwards.”
The starting point for Direqt and Objective is essentially the same – training the AI machine on the site’s content. They just use it so far for different things. But online publication readers may have both types of tasks – to learn more details about the material they just read and to find something by words they don’t remember exactly or by a topic they can only formulate in general.
And this means that sooner or later, these two functionalities will merge either within Direqt or within Objective, or as a result of the efforts of someone else who also decides to sell AI tools to content publishers.
Where to Run
This may become the main direction of movement – creating modern AI platforms and tools for content publishers and readers.
Two specific tasks that such tools can already clearly solve are:
Asking questions and getting additional information on the topic of the read article. Searching for articles not by exact word match but by meaning, including searching for articles on the necessary topics, concepts described in them, and other general concepts. The main point – these tools may be complex in terms of technology but must be extremely simple to install and use. For example, they could be supplied as plugins for WordPress, which, by the way, is used to create half of all internet sites. Including FastFounder.
By the way, I would definitely not refuse such a plugin 😉 As would many other bloggers.
These problems are old, but only now can they start to be effectively solved using AI technologies. So the moment for creating such platforms seems very timely.