The problem with meaning in the texts generated by the AI?

Cm. here is the list.
You need to be able to do the same: Demo

the Markov chain always generate random data (garbage), just restricted to certain rules (dimensions of probability).
to raise the quality using competitive approaches GAN (to generate this approach is now at the forefront is).
read more here:
\rhttps://ru.wikipedia.org/wiki/%D0%93%D0%B5%D0%BD%D...

Well, as it AI we have is still there. There are some neural networks that work on specific tasks. But the main problem for you, the fact that neural networks do not understand the meaning of the text.
So, the task is currently not feasible. This requires "network", which grew up listening to tales from my mother in childhood, passed the stages of development of the personality, filled "bumps" so to speak.
Easier to hire a rewriter - man.