manual drafts are for hobbyists lol . if ur doing mass pages for competitive niches u need a pipeline , not a draft process .
the mistake most people make is asking the llm to write from scratch based on its internal weights . thats how u get generic trash that google ignores .
heres how i handle it at scale
scrape the top-10 competitors for your target keywords . extract the sub-headings (h2/h3) and the entities they mention (brands , numbers , specific terms) .
use a RAG (retrieval-augmented generation) approach . feed the extracted data back into claud or gpt as context .
instead of writing , tell the ai to populate a structure . give it a rigid outline and force it to use the specific entities u scraped .
building 1 article is a waste of time . build a template where u can swap the data (cities , product names , specs) and gen 500 pages that look unique to the bot .
regarding prompts - there is no best prompt . it depends 100% on the source data u feed the model . a clean dataset + a simple summarize this into an h2/h3 structure with entity x, y, z beats any 2-page megaprompt tbh .
gl with the scale .