Physix FrontierAfter three months with WorkBuddy, I admit it solves only 60% of my paper issues—but the remaining 40% makes me more anxious
Conclusion first: WorkBuddy is the closest thing I've used to the definition of a "full-process paper writing tool." But it's not a panacea, and the parts it can't solve happen to be the most critical aspects of a paper.
I'm a third-year master's student. My advisor keeps asking me to revise the sixth version of my thesis, ranging from "adjust the framework of the introduction again" to "check if this formula derivation is wrong." Each revision feels like a new world. I've tried Zotero for literature management, EndNote for citations, Little Green Whale for translation, and Scholard for peer review—in the end, I realized the problem isn't a lack of tools, but the lack of connection between them. Notes for a paper are in Zotero, translations are in Little Green Whale, citation formats are in EndNote, and when writing the paper, I still have to switch between windows one by one.
WorkBuddy did improve this sense of fragmentation. But after using it for a week, I discovered its true boundaries.
The 60% It Does Well
Regarding literature management, I previously wrote a post saying WorkBuddy made me uninstall Zotero and EndNote. Looking back now, that judgment was a bit aggressive, but the direction was correct. WorkBuddy's literature management module treats "reading literature" and "writing papers" as a single task.
Specifically, it does three things well, in my opinion:
First, literature excerpts can be dragged directly into the paper draft. Select a paragraph, right-click "Add to Current Document," and it automatically generates citation markers. No need to remember reference numbers or manually check formats. I tried processing 20 references simultaneously, comparing how the same theory was expressed in different articles, and it saved about 40% of the time.
Second, the format templates are rigid, but rigidly correct. IEEE templates, National Standard GB/T 7714, APA 7th—these format details are standardized and don't require creative freedom. In my tests, WorkBuddy's template matching required manual adjustment in only about 5% of scenarios. The remaining 5% are special cases, such as certain journals having unique font requirements for figure captions, which must be fixed manually, but that's rare.
Third, when AI assists in writing paragraphs, it can cite articles stored in your literature library. This is its biggest difference from ChatGPT and Claude. Citations generated by ChatGPT are often fabricated. When I submitted my first draft, my advisor grilled me: "Are you sure this Journal of Applied Physics article was published in 2026? I can't find it anywhere." Being exposed on the spot was extremely embarrassing. WorkBuddy at least guarantees the citation sources are real because it only cites literature you've put in the library.
Honestly, for this point alone, I can accept its shortcomings elsewhere. Citation fabrication is an academic red line; cross it and you're dead.
The 40% That Gives Me Headaches
But the problems lie in the parts WorkBuddy cannot solve.
I made a table comparing WorkBuddy, ChatGPT, and Zhangqiao Research AI on several key dimensions (based on my actual testing of an 8,000-word Computer Science paper):
| Dimension | WorkBuddy | ChatGPT-4o | Zhangqiao Research AI |
|---|---|---|---|
| Connection from Lit Mgmt to Writing | Excellent | No such function | Good |
| Initial Draft Generation Speed | ~15 mins | 15-25 mins | ~10 mins |
| Initial Draft Plagiarism Rate | ~15%-20% | 25%-35% | 7.2% |
| AIGC Rate Control | ~30%-40% | 95%-100% | 9.5% |
| Format Compliance | Excellent | Needs manual adjustment | Good |
| Logical Coherence | Medium | Excellent | Medium |
| Innovative Viewpoints | Weak | Strong | Weak |
I found WorkBuddy's biggest weakness is that it has "no opinions of its own."
Its AI is better at integrating existing information than generating new ideas. If you ask it about "the development trajectory of this theory in field X," it can list comparisons of viewpoints from three papers clearly. But if you ask, "Based on these three papers, can I propose a different hypothesis?" it basically gets stuck or generates fluff like "Overall, further research is still needed in this field."
While ChatGPT talks nonsense regarding literature citations, it is indeed much stronger at sparking ideas, providing framework structures, and helping clarify logical chains. My current workflow is: Use ChatGPT to brainstorm frameworks and arguments, then use WorkBuddy to fill in specific content and handle formatting.
Why Does That Industry News Matter to Me?
Seeing that roundup article, honestly, I have mixed feelings. Its evaluation data is detailed, but for a third-year master's student, it fails to answer the core question: How exactly do I use these tools so that my advisor believes the paper was written by me?
ChatGPT-generated drafts have an AIGC rate of 95%-100%; dare you submit that directly? Zhangqiao Research AI controls plagiarism and AIGC rates very well, but its generated content has "slightly weaker generalizability outside specialized fields"—in plain terms, it's a bit templated and easily spotted by teachers as "written by AI."
WorkBuddy's AIGC rate is around 30%-40%, much better than ChatGPT, but higher than you might imagine. I tried using WorkBuddy to generate a discussion section, then modified it myself. Finally, when tested with a plagiarism checker, the AIGC rate was about 12%. I can accept this number, but the premise is that I spent a lot of time restructuring sentences, adjusting logical order, and adding my own analysis.
So, the real issue isn't whether the tool is good, but whether you are willing to spend time transforming AI-generated content into your own work.
My Real Advice
If you are still struggling with which tool to use, my advice is:
Break your paper into three parts and handle them with different tools.
1. Framework and Logic (20% of workload): Use ChatGPT or Claude to help clarify thoughts. Seek inspiration, not content.
2. Literature Citation and Formatting (30% of workload): Use WorkBuddy. It does this best and won't make mistakes.
3. Content Filling and Revision (50% of workload): This part must be done by you. AI can help write the initial draft, but you must do secondary processing—change sentences, add examples, adjust tone, check for logical gaps.
I've seen too many people, including myself, initially receiving an AI-generated draft, thinking "Wow, this is great," and submitting it directly. The result was being shut down by a single sentence from the advisor...