An Oncoming STORM for AI Academic Writing | Stanford STORM

“Synthesis of Topic Outlines through Retrieval and Multi-perspective Question Asking” is a bit of a mouthful, but STORM is a Stanford AI research project that is a “writing system focusing on the pre-writing stage to generate long, grounded, Wikipedia-like article for a given topic from scratch.” In a single prompt, that is only 20 words or less.

Empty Calorie or Academic-ish writing?

Much like I alluded to two years ago with the idea that AI could help remix or generate Open Educational Resources (OERs), STORM makes me wonder if we will see AI find a place in creating OER? The idea of an AI combined with previously published OERs and connected to Creative Commons and Public Domain resources seems prime for innovation. STORM come close, as it leans toward a more structured way of enhancing and creating Wikipedia articles, while possibly providing a launchpad for aspiring academic writers as well.

In widely used AI tools like ChatGPT, long-form writing is often built through iterative prompting, a back-and-forth that refines responses. This alone doesn’t typically make the most accurate results, and often writers might leverage retrieval-augmented generation (RAG), where external materials ground the AI’s output. Add in tweaks like “no yapping” to cut fluff, and the result is something that resembles an academic writing, but not necessarily is academic writing. As many technologists, instructional designers, or educators will point out, these AI-generated responses often feel like what I like to refer to as empty calorie writing. Writing that while it is technically correct, sometimes well-structured, it often still lacks a lot of the breath and depth that makes human academic writing. Things like DeepSeek, as I’ve mentioned recently, shakes this up slightly with their R1 (or Reason button in ChatGPT) to have the AI show their work or show how their thinking-aloud on the prompt it was given. But again it is still often reads flat or full of empty calorie writing.

“STORM models the pre-writing stage by (1) Discovering diverse perspectives in researching the given topic.

(2) Simulating conversations where writers carrying different perspectives pose questions to a topic expert grounded on trusted Internet sources.

3) Curating the collected information to create an outline.”

https://storm-project.stanford.edu/research/storm/

Write me an ‘article’ from a 20 word prompt or less?

STORM integrates a multiple AI agent approach. I centers this automated slef-itterating conversation into a structured outline that then becomes a Wikipedia-style format article without any back-and-forth with the user. What this means is that the AI engages in self-dialogue, using specialized “editor” agents to simulate the reasoning process, through its own think aloud similar to DeepSeek R1 or Reason in ChatGPT. But then it follows a “Perspective-Guided Question Asking” phase, mining diverse viewpoints from relevant Wikipedia articles and more academic style resources, then conducts internal follow-up conversations using a retrieval-augmented question-answering process. In short it researches, asks itself what it just researched, then poses why it is important and then synthesizes. This part removes the writer altogether from correcting or asking more questions as most are used to in something like ChatGPT. STORM calls this method it’s “BrainSTORMing Process,” which allows users to trace the AI’s thought process to see it’s rational and process to create its output.

It is clear that Stanford is trying to design this project for complex, multi-faceted topics and support the Wikipedia community. But in the hands of a time crunched undergrad, or a graduate student not knowing what to research; it could provide some structure as a jumping off point for their writing, STORM appears to be a compelling project, and could devlop into an interesting academic writing tool complete with citations, references, and weblinks with its writing. Providing another avenue for a strong foundation for someone to start academic writing; or a wrong-turn of misuse in academic dishonesty. Although it’s references that I’ve seen it use in my test prompts are much more academic in background, I haven’t seen it (nor do I know if it can) pull from actual academic journal or peer-reviewed articles. Which again in the hands of someone who is trying to learn and exercise academic writing is counter productive.

The resulting output, however, is pretty compelling and much closer to what academic writing is than what I see typically out of a pure ChatGPT output. It generates the generated article in-browser, or can be exported as a multipage PDF with a linked table of contents. Again, a system like this could be the making of a practical tool for preliminary research and pre-writing (if cited and permitted by the guidelines of an academic institution’s academic honesty policy), but I wouldn’t risk a final dissertation grade on its output.

Sample “Article”

Based on the 20 word prompt, you can be the judge of STORM’s output. I purposely chose the topic of neurological connections and the juxtaposition of pop culture references to see what sources it pulled from in its “brainstorming.”

CO-STORM

The other half of their platform is CO-STORM, which looks to be a more roundtable conversation model of STORM. However, at the time of writing, that part of the system is currently down.