Detecting changes in authorship

Problem

Changes in writing style can be subtle, contextual and difficult to distinguish from normal variation. We explored whether they could be detected and explained across short-form communications, including writing influenced or generated by AI.

What we did

Tekh worked with academia throughout the project, forensic-linguistics expertise together with our data engineering, machine learning and product development capabilities. This multidisciplinary approach helped ground the technology in established authorship research while keeping the outputs useful and understandable for human analysts.

Over 12 weeks, we developed a containerised research demonstrator combining stylometric, statistical and machine-learning methods. The work covered three seperate languages, with explainable outputs designed to show the evidence behind an alert rather than present an opaque automated judgement.

Result

The project demonstrated technical feasibility and produced reusable research assets, including code, evaluation methods and datasets. Results varied across scenarios, reinforcing that subtle authorship change remains a frontier research problem and that expert interpretation is essential.

Tekh retains ownership of the project intellectual property, giving us a strong foundation for continued research and future product development in authorship analysis and AI forensics.