One such development is a conversational AI system designed to curate and contextualize news in real time. And we think this could be one of the possible paths news media will go in a very near future to meet the audience's expectations.
Have a look on our new feature — Chat — that mimics the future of news media.
Moving beyond chatbots
In general I think about chat interfaces as the most convenient way to communicate with the audience. I don't believe people need news formats, styling etc. I believe that people need information to make decisions or be informed. And of course to be entertained, educated etc. Chats can do all this in a best way — just answering user's questions and offering something. And here we have a lot of variations of how we can organize chats: by topics / output formats or universal, tone of voice etc. But the general idea is simple: when people get used to use chats (messengers and services like ChatGPT or Claude), those big noisy webpages of news media become less and less convenient to use. And also Google that kills search traffic with Overviews etc — all this make us think that people need summaries from several sources (even inside one news media, but from different pieces of content) to get full context in a glance. And news media can easily provide it. In chat.
The technology? Well... I spent 7 hours to build this system. I didn't do event-driven RAG system to work with entire database to optimize context window for LLM requests (just because my database is not big enough), but it won't take a lot of time and efforts. And I think this will be the next step as I want tot test it for my another project — Startupt.in.
How the technology works
At its core, the system parses user queries — such as “How are news organizations in Scandinavia using AI to create content?” or “How is AI applied in Southern European newsrooms for management?” — and retrieves examples from a curated database of AI initiatives. Each response is attributed and presented in a conversational summary.
A central feature is its semantic search and keyword-mapping capability. For example, when asked about “Southern Europe,” the system expands the query to include countries such as Spain, Portugal, and Italy. It also interprets terms like “management” to cover areas such as newsroom workflows or editorial oversight. The system recognizes aliases for large language models as well. For instance, mapping “Claude” to Anthropic and “ChatGPT” to OpenAI. This same logic is applied consistently across all content elements.
The query processing follows a structured flow: the system breaks down each request into two components — the main intent and associated keywords (such as brand names, regions, or technical terms). Keywords are categorized by type, with weights and priorities assigned based on their relationships. For example, if a user asks, “How is AI used in Portugal to earn money?” the location is prioritized first, while “earn money” is matched to “monetization” in the database. If no examples are available specifically for Portugal, the system broadens results to similar initiatives in other regions and offers an index for further exploration.
This approach goes beyond basic keyword matching, aiming to interpret queries in ways that reflect how journalists make connections between context, terminology, and geography. The goal is to provide a more natural, dialogue-based user experience. Future integration with retrieval-augmented generation (RAG) could further improve this capability.
What is under the hood
Currently, the platform runs on large-scale models such as Llama 3.3 70B Versatile and Mistral Saba 24B, hosted on Groq, both for the chat interface and the underlying content management system. These models are cost-efficient and perform strongly in text-based tasks. However, the developers note that smaller models could potentially deliver similar results, since the system’s context requirements are relatively modest — raising questions about the balance between performance and efficiency.
User experience
The interface is designed as a straightforward chat window. It supports multiple AI providers, allows users to supply their own API keys, and includes transparent tracking of token usage and costs.
Responses are structured to answer questions directly while also offering broader context, with links to further initiatives for readers who want to explore more. And also user can download chat in simple txt file. And by the way, we don't store a thing in database (API Keys and user's requests) — everything is happening in user's browser.
Implications for news media
Advocates of this approach argue that conversational systems could represent a new model for news consumption — one that prioritizes dialogue over passive reading. Instead of encountering a single, static article, audiences could engage in an interactive exchange, asking follow-up questions and receiving context that adapts to their interests.
If developed at scale, such interfaces might provide not only real-time updates but also historical background, comparative case studies, and expert perspectives. This could point toward a form of news media organized around a single conversational hub — whether a webpage or a voice assistant — where personalization and context are central.
Another consideration is independence. Because the system can operate without relying on major technology platforms, it offers the possibility of protecting newsroom content from being absorbed into large-scale corporate AI training pipelines, an issue that continues to generate debate in the industry.