You can not solely rely on AI. I have been working with AI based content generation tool, long before GPT3. I tried all the models. They suffer from the same problem. The models can not remember context beyond 1200 to 3000 characters. The models are not meant for generating coherent long form articles. The tools that sell AI writing services are using few tweaks and in some cases custom GPT-J open source models, which is not powerful like GPT-3.
There are only few parameters you can tweak to generate good (subjective) content.
1. temperature - (this decides the "creativity" or entropy)
2. top_k - controls diversity of the vocabulary
3. top-p - controls the probability of the next word in the current sentence
4. beam search - which controls the ngram (this is not good for open ended text generation , only useful for summarisation). In some cases it also reduces the risk of missing out on
hidden high probaility word sequences.
The tools that you see are not suitable for BH purpose. I am considering making a tool that uses custom models with
parameters that are open to tweaking. The tool in question should take a bunch of questions and generate articles in bulk for it.