Please use models that support long memory for writing, such as KIMI, Claude, etc. Because in step 4 you need to bring in at least 3-5 pieces of content from the search results to reference and create.
In the creation process due to the length of the input text is too large, the output may be interrupted, the above two models by inputting "continue", generally can give full and complete output results.
This method, in addition to manual operations, provides automated execution code: https://github.com/mshumer/ai-journalist
1. Give the topic of writing
{topic}
Example: bitcoin price trend in 2024
2. Extended Theme
Please provide 5 search terms related to '{topic}' for research and article writing. Please reply as a Python parsable list, with commas separating the keywords.
Note:Output Python to facilitate engineering and output data formatting, manual operation can ignore this part of the prompt.
3. Gathering information
Search using the prompt words generated by the 2. extension theme. Then manually filter the content for quality or use the following prompt words to filter the content.
Search Results:
\n{search_results_text}\n\n
From the search results above, please select the URLs/text blocks that are most relevant and informative for writing articles on this topic.
Please reply as a Python parsable list, with numbers separated by commas.
4. Writing articles
Bring the collected information to {combined_text}
Subject: {topic}\n\n
Article text: \n
{combined_text}\n\n
Write a high-quality, New York Times-compliant article on the given topic based on the article text provided. The article should be clearly structured, informative and engaging. Make sure the article is at least as long as the NYT cover story - at least 15 paragraphs.