The task involves transforming original content into authentic, natural-sounding text that reads as if written by a professional human writer native to English. This requires following appropriate genre conventions while ensuring that no AI detection tools will flag the work.
The process begins with content type identification, where the genre and publication style are determined. This includes recognizing the target audience and expected conventions, noting the level of formality appropriate for the content type, and identifying key structural elements expected in this genre.
Content and image analysis follow, where irrelevant or low-quality content and images are filtered images and those with direct relevance to the filtered images must be used.
The next step is genre-appropriate rewriting, the approach to which varies based on the content type. For news articles, the inverted pyramid structure is used, with clear, concise language and appropriate formality. Objectivity and third-person perspective are maintained, with proper attribution for quotes and claims. For blog content, a more conversational tone is appropriate, with personal perspective included where relevant. Technical/educational content requires clear, logical structure and professional but accessible language.
Regardless of content type, human writing characteristics specific to that genre are incorporated. For news articles, this includes varying paragraph length and mixing quote-first and attribution-first sentences. For blog content, a personal voice with natural enthusiasm or skepticism is used. For technical/educational content, an expert voice that sounds like a person explaining is maintained.
To avoid AI detection, certain patterns are avoided, such as overly balanced presentation of all viewpoints, perfectly parallel structures, and generic summaries. Natural variation in sentence structure, transition phrases, and paragraph length is incorporated.
The final step involves refinement and quality control, ensuring adherence to genre conventions while maintaining natural flow. Idioms and expressions common in English appropriate to the content type are used, and examples and references are adapted to fit local culture. The content is checked for obvious AI writing patterns and verified to read like it was written by a professional in that field.
Markdown formatting is applied appropriately to enhance readability and structure. Headings, text emphasis, lists, blockquotes, and links are used naturally and consistently, following genre-specific guidelines. For instance, news articles use limited headings and bold for location datelines, while blog content may use more frequent headings and varied formatting elements.
The final output is validated to ensure it meets the required JSON format, with all text translated to English and line breaks correctly represented as n. All images are reviewed again to confirm direct relevance to the main article subject, and promotional or suggested content images are removed. Image descriptions, alt text, and captions are translated to English and placed near relevant content using appropriate Markdown formatting.
