The Chatbot That Knows Every Article in Your Publication

01 /The Concept

Direqt built a platform on which online publications – magazines, blogs – can deploy chatbots for their readers.

The chatbot talks with readers through an AI trained on the publication's own content. The chat window can be embedded anywhere on the site – including mid-article, so readers can ask follow-up questions about whatever they just read.

For example, while reading a profile of David Beckham, a reader might ask when he met Victoria, or how many children they have.

Or a reader might simply ask the chatbot what the most interesting recent stories on technology are – or on any other topic the publication covers – and get back links to the latest popular pieces.

To train the chatbot on a publication's content, all previously published material is ingested first, with new articles added as they go live. Currently this is done via RSS feed, which the chatbot monitors itself. Not the most elegant solution, but reasonably universal.

As the chatbot learns the publication's content, it also absorbs its editorial voice. That voice can be further fine-tuned with a set of instructions: what tone to use with readers, which expressions to favor, which to avoid.

Publications choose between two pricing models. The first is a monthly fee based on message volume. The second: allow Direqt's partner ads inside the chat window – no fee, and a revenue share from what Direqt earns on those placements.

Chatbots built on Direqt are already live on 75 publications, including Vogue, GQ, Wired, ESPN, Men's Health, and The Sun. Over the past 30 days, readers across those publications exchanged more than 10 million messages with their chatbots.

Direqt was founded in 2017, initially building conventional chatbots for publishers – with mixed results. The decision to build an intelligent bot on top of GPT came in 2021, supported by $1.4 million raised in 2022 and a new $4.5 million round now.

02 /Why It Matters

An intelligent chatbot that converses with readers is technically impressive. But the natural question is: why?

The answer is one of the most important metrics in online publishing: time on site. One Direqt customer reports that readers spend three times longer on their site after the chatbot was installed – asking follow-up questions about articles they just read.

Zoom out, and the chatbot is one of the first signals of a deeper paradigm shift: the move toward interactive content.

Every writer constantly navigates the tension between brevity and depth.

An article isn't a book – it's not designed to cover everything exhaustively. It also can't be too long. Especially now, in an era of shortening attention spans, articles need to be brief.

But readers who are genuinely interested in a topic want more detail – and without it, they feel the piece is shallow. Meanwhile, all that additional context overwhelms readers who wanted just the highlights.

The result: writers produce "medium" pieces that don't fully satisfy either camp. Too long for the skimmers, too shallow for the curious.

Interactive content could resolve this. An engaged reader can ask the chatbot – the AI trained on the publication's full archive – for more detail on exactly the aspect they care about. Everyone else reads the short piece and moves on.

If a publication has already trained an AI on all its content, that creates another capability worth noting – one that Direqt may not have fully explored yet.

A startup called Objective raised $18 million in its first round to do just that. It offers sites an "AI search" that finds content not by keyword matching, but by meaning. Traditional search engines look for the exact words the user typed. But what if you don't remember the exact phrasing from an article you read months ago? What if you want to find pieces on a topic whose name doesn't appear anywhere in the text?

Objective finds articles by semantic meaning – which can be expressed in different words and spread across different parts of the text. It also applies the same approach to images: describe what's in the picture, and it finds matching visuals.

Something similar was [covered previously](/review/fig-chego-najdjosh) – Kive, a platform helping creative teams manage catalogs of photos and visual assets. Kive auto-tags objects and details in images, enabling searches like "noir-style photo of an adult male with a slightly upward-tilted head turned fractionally to the right."

Direqt and Objective both start from the same premise: an AI trained on a site's content. They just use it differently – for now. But readers of a publication often need both things: more detail on what they just read, and a way to find something they half-remember. Sooner or later, those two functionalities will merge – whether within Direqt, within Objective, or through someone new entering the space of AI tools for content publishers.

03 /Opportunities

This is the central opportunity: building modern AI platforms and tools for content publishers and readers.

Two specific problems are clearly defined:

- Asking questions and getting more detail on the topic of an article just read.

- Searching for articles not by exact keyword matches, but by meaning – including searching by topic, concept, or theme without knowing the exact terminology the article uses.

The key point: these tools can be technically sophisticated, but they need to be dead simple to install and use. Delivered, for instance, as plugins for WordPress – which powers roughly half of all websites on the internet. Many bloggers would use them immediately.

These are old problems that AI technology has only recently made solvable effectively. The moment to build these platforms is now.