The problem of (semi-)automatically turning text into hypertext is one that has been identified as important to the growth and development of hypertext as a way of organising information. In this paper we describe an approach we have developed to semi-automatically generate a hypertext from linear texts. This is based on initially creating nodes and composite nodes composed of ‘mini-hypertexts’. Following this we then compute node-node similarity values using standard information retrieval techniques. These similarity measures are then used to selectively create node-node links based on the strength of similarity between nodes. What makes our process novel is that the link creation process also uses values from a dynamically computed metric which measures the topological compactness of the overall hypertext being generated. Thus link creation is a selective process based not only on node-node similarity but also on the overall layout of the hypertext. Experiments on generating a hypertext from a collection of 846 software product descriptions comprising 8.5 Mbytes of text are described. Our experiments with a variety of IR techniques and link creation approaches yield some guidelines on how the process should be automated. Finally, this text to hypertext conversion method is put into the context of an overall hypertext authoring tool currently under development.