Where do language models fall short in the task of Natural Language Generation?
Natural Language Generation heavily depends on the purpose, just like how your written communication varies with the purpose. For instance, you will need to apply a lot of creativity while writing a blog post; curation requires that your content and world general knowledge are based on the input document; while reports are mostly templates & use Natural Language Generator to realize the sentences and Natural Language Generator for chatbots requires NLU/intent engines.
To conclude, language models fall short of bringing the purpose into consideration. Even after NLG shifted from templates to the dynamic generation of sentences, it took the technology years of experimenting to achieve satisfactory results. As a part of NLP and, more generally, AI, natural language generation relies on a number of algorithms that address certain problems of creating human-like texts.
If you are worried about where to get product descriptions for 1000+ products, then try. It costs you around 25 to 15 paisa per word (40 to 20 cents) depending on the number of descriptions you order. EPDG costs 1/3 rd of the cost of hiring a content writer (in India) and 1/20th of the time required by content writers. The final output will be proofread by experts and hence is of first-class quality.