High AIGC detection rate
AIGC (full name: Artificial Intelligence Generated Content) means “artificial intelligence-generated content.”
The core task of AIGC detection is to determine whether text was generated by AI. The AI generation probability (AI rate) expresses as a percentage the likelihood that text was generated by AI: 0% means entirely human-written, and 100% means entirely AI-generated.
We use a variety of AIGC detection tools to test generated book content and have found that it still has a relatively high AI rate (interestingly, when we selected some articles and materials published before 2010 that were definitely written by humans and tested them, they also still received high detection rates).
If you want to directly use books created by this project without human edits and achieve a lower AI rate, this platform may not be suitable for you, and we make no commitments in this regard.
OCR text extraction quality for PDF files needs improvement
Extracting text from images (or PDF files with only an image layer) using OCR technology is a technical challenge that many top technical teams around the world have studied for a long time and still cannot perfectly solve.
This project uses a combination of third-party technical solutions such as Baidu PaddlePaddle, Microsoft MarkItDown, and Alibaba Cloud Intelligent Document Parsing. However, none of these solutions can perfectly extract text; the extraction quality is closely related to factors such as the layout complexity and image clarity of the original file.
If you use reference materials in PDF format for AI writing, it may affect writing quality.
There are limitations in non-Chinese writing capabilities
This project supports uploading multilingual materials and having the AI read them, but testing has shown that the AI may be unable to generate long books written entirely in English. This issue has not yet been further investigated or resolved.
Limitations of image insertion
When creating a project, if you select “Use images”, the AI will automatically insert images into the book while generating it. How it works:
- First, all images in the reference materials are extracted, and description information for each image is inferred and generated from context. The images and their descriptions are then stored temporarily in the library;
- While generating the book, if the AI determines that the context of the text passage it is writing corresponds to an image, it inserts that image into the text.
This process is not always reliable. Because the position and format of caption text differ across reference materials, the AI cannot make perfectly accurate judgments and associations. As a result, there is a very small chance of the following problems:
- An image that should have been inserted is missing
- An incorrect image is inserted
To address this, you need to check the illustrations in the generated book and handle them manually:
- Option 1: Export the Word file, then edit it and insert images separately;
- Option 2: Use the online book editing feature. Click the “Insert image” button on the editor toolbar, open the image library, select an image, and click the insert button.
Issue: SVG images converted to PNG images in Word files
Because Microsoft Word files (DOC / DOCX) do not natively support embedded SVG images, when a book is exported as a Word file, the current approach is to convert SVG charts into PNG images and insert them into the body text.
Limitations of citation references
According to the GB/T 7714-2025 standard “Information and documentation — Rules for bibliographic references and citations to information resources”, citations should include page numbers.
Here is a correct example:
In the body text: Qian Mu discusses this view in detail in Outline of National History①.
Footnote: ① Qian Mu, Outline of National History (revised edition), Beijing: The Commercial Press, 1996, p. 38.
However, in this project, because the reference materials used for AI writing are mainly files without page numbers in formats such as EPUB, DOC, and Markdown, some files may not even have clear chapter features that AI can recognize and distinguish.
For reference materials in PDF format, although most original files have page numbers printed on them, because their positions and formats vary, current OCR text extraction tools find it difficult to intelligently recognize and extract them (unless custom specialized OCR software capable of locating the page number area is used).
Unless users first edit and convert reference files in various formats into neatly typeset, paginated files such as PDFs with standardized page number formats, indicating page numbers in citations is an almost unsolvable, complex, and enormous engineering problem, far beyond the scope of this project.
For the complex reasons above, in books currently generated by this project:
- It is impossible to determine and mark the page numbers of citations;
- For citations whose chapter cannot be identified, it is also impossible to mark the chapter.
You can manually insert or edit citations in the exported Word file (DOC / DOCX). Although this also takes considerable time, the overall workload is still far less than manual writing.
Other
At present, the test results for nonfiction book writing are fairly satisfactory, but we have not yet tested the ability to write fiction.
Based on the technical principles of this project, we speculate that satisfactory results may not be achieved when creating fiction.It is not recommended at this time to use this project to write novels, prose, autobiographies, and other works of a literary or fictional nature.