Overview
A large public organization in the Netherlands saw an opportunity to improve its content production process, particularly because the journey from raw project information to published website articles was lengthy and labor-intensive. Project data scattered across dozens of documents, years of archives, and strict guidelines for tone, accessibility, and inclusivity: the process demanded a great deal from editors. As AI became increasingly visible as an option, the question arose of how it could be deployed without sacrificing quality or human oversight.
The challenge
The organization regularly produced content for its website, a process that worked well but required significant time and manual effort. As AI continued to gain ground, the question emerged whether there was a smarter way: to shorten the long path from project information to published text and make it less labor-intensive, without compromising on quality or control.
The obstacle
What initially appeared to be a relatively well-defined assignment turned out, on closer inspection, to be more complex. The organization needed a tool for content generation, but lacked a functioning process to underpin it. There was no structured way to collect, organize, and prepare project information for use. Without that foundation, any AI solution would be built on shaky ground. This meant that Latitude’s task was not simply to build a technical solution, but first to think through and establish the underlying process together with the organization.
The solution
Latitude developed an application that supports the entire workflow. Project managers set up a project and add relevant documents. An AI agent analyzes those documents and distills the key points into a single, coherent knowledge document. Based on that, a first draft article is automatically generated.
This is followed by a layered review process carried out by a second AI agent:
- First, the article is checked against internal guidelines and rewritten where necessary.
- Next, fact checks are performed and the text is assessed for bias and inclusivity.
- Finally, readability and language use are adjusted so that the article meets the required writing style.
If an article fails a check, it is automatically rewritten until it meets the standard, or until it becomes clear that human intervention is needed.
In this way, human oversight remains a deliberate part of the design at all times. The system is not built to publish content directly, but to make the editor’s work as focused and purposeful as possible. Where the system is uncertain it flags a warning, for example; a fact that warrants verification or a guideline that has not yet been fully followed. This way, the editor knows exactly where their attention is needed.
The results
What began as a question about content generation grew into a fundamental improvement of the underlying information process. By getting that process in order first, a solid foundation was created upon which AI could truly demonstrate its value. The result is a workflow that is faster and more consistent, and one in which ultimate responsibility is deliberately controlled by people, instead of AI agents.
