When a student is starting out, or heading towards their doctoral defense, knowledge alone is not enough anymore. A breakthrough in artificial intelligence, advanced computer graphics, or natural language processing truly fulfills its potential when it’s communicated clearly—first to peers, then to funders, employers, policymakers, and the public. Here’s how USC Institute for Creative Technologies developed an AI-Integrated Communications Framework that lifted LinkedIn engagement by 51% to bring students’ backstories to the world.
Byline: S.C. Stuart, COMMS / & Writer, ICT
Founded in 1999 as a Department of Defense University Affiliated Research Center, sponsored by the US Army, ICT brings together researchers in artificial intelligence, graphics, geospatial sciences, human performance, learning sciences, medical VR, and narrative. Around 177 people work here at any given time, including faculty, postdoctoral fellows, and students.
Each year, ICT welcomes a rotating cohort of undergraduates, graduate researchers, doctoral candidates, and interns. Alongside their scientific training, we now encourage students to collaborate on first-person essays about their projects, methods, and experiences. This process allows them to craft backstories that are both professional and accessible, translating complex ideas for broader audiences while maintaining scientific rigor. Importantly, these articles are published under their own bylines – not the institute’s.
And we use AI to do this – but more on that in a moment (or scroll down to: “The AI-Integrated Communications Framework” if you need to skip ahead).
ICT: A Hub of Interdisciplinary Research
When I joined the USC Institute for Creative Technologies (ICT) in March 2023, I recognized a need to build a structured approach to communications and storytelling. Research at ICT was already groundbreaking, but the language and dissemination of that work were often confined to specialist audiences.
My background as a science and technology journalist trained me to translate complex R&D into compelling narratives – and I wanted to bring this methodology to ICT. I first designed a strategy to streamline our project sheets into a better structural flow (Background, Objectives, Results, Next Steps).
Then I worked with Dr. Randall W. Hill, Jr, our Executive Director, to recount ICT’s backstory [How We Built the Holodeck], summarize 25 years of research [25 Years of Wargaming at ICT], and reiterate ICT’s Commitment to National Defense to our Army sponsors. In collaboration with Anthony DeCapite, former United States Marine Corps Videographer (Director) and David Cobbins, former United States Army Non-Commissioned Officer and Combat Medic (Assistant Director), we shot and released a “trailer” / Overview video.
Dr. Hill pointed out, “In today’s competitive landscape, students need a public narrative of their own making—a professional backstory they can carry forward. A project expressed in plain language has a better chance of attracting interest and funding. If they can explain their research with clarity, they’ll stand out to employers, reviewers, and collaborators.”
To this end, over the past year, ICT has published more than forty essays spanning students from short-term undergraduate research experiences to PhD candidates preparing for international conferences.
Tianyi (Mavis) Zhan, PhD student in computer science, described retrieval-augmented generation for conversational agents—systems that respond not only with linguistic fluency but with sensitivity to human traits and emotions. Jing (Carl) Yang, Vision and Graphics Lab, wrote about light transport and material representation, explaining how his research brings new precision to digital environments, while Natali Chavez, visiting scholar from Greece, reflected on making her first film in Los Angeles using AI.
When PhD candidate Ala N. Tak’s paper “Mechanistic Interpretability of Emotion Inference in Large Language Models” was accepted to ACL 2025 in Vienna, we worked together on an article which explored how LLMs understand emotions.
Previously, we’d publish short news items on ICT’s site whenever a researcher was heading to a conference. But writing an article, under the student’s byline, proved much more compelling.
For example, when PhD student Bin Han’s paper “Can LLMs Generate Behaviors for Embodied Virtual Agents Based on Personality Prompting?” was accepted to IVA 2025 in Berlin, we published an eye-catching essay with the title “Crafting Personalities with Code” – and received several requests for the pre-press publication on LinkedIn almost immediately.
The AI-Integrated Communications Framework
How do we do this? After all, few students have the time—or training—to write anything but their academic research assignments.
The solution lies in a custom AI-integrated framework, which I started developing earlier this year, to produce first drafts (text and visuals) which are checked for grammar, spelling, tested embedded links – ready for the student to edit, refine and review. This strategy has nearly doubled our communications output in twelve months, while maintaining quality standards.
After an initial chat on Slack, I send the student ten questions in a Google doc – these are designed to hone down their backstory and research into digestible sections. I ask them to write no more than 1 – 2 lines per question, and encourage them to use the microphone option on Google docs to have personable, not purely-academic responses.
While they’re doing that, I grab the links to their publications (including pre-press), conference abstracts, lab notes, github, LinkedIn profile, and any other relevant information.
Once the student’s answers are in, I pour in all the data I’ve found – but don’t organize it (because AI is good at doing that, so why waste the time?) into the working doc.
In my office at ICT, I have two big screens, and several browser tabs open with several AIs, ready to be put through their paces. I start with a version of ChatGPT which I’ve been training since February 2025, using linguistic programming (i.e. talking naturally back and forth using the vocalization option, rather than barking individual prompts for tasks).
Before I share the Google doc of notes, I give a brief biographical sketch of the student, (reminding the LLM to avoid any inherent bias), and pointing out key phrases in their responses to tailor the writing style. I’ve trained the LLM as a rookie reporter, mirroring the drills I survived while training on a national newspaper back in London, England: how to write catchy headlines, keep the reader engaged, and don’t bury the lede.
It produces the first draft in about 2.5 seconds. The speed is startling, even after months of doing this. We then work on it for a few minutes until I switch to another browser tab and ask another LLM (Claude.ai) to refine it. This is essentially hacking an ersatz Generative Adversarial Network (GAN) by pitching two AIs against each other. Sometimes I add Notebook LLM into the mix if the student has several publications so we can create a knowledge base to draw on and query.
Once the final draft is complete, I go back to the first LLM to create promos suitable for LinkedIn, X, and Instagram, including standard USC hashtags and limited but effective Gen Z-style use of emoticons.
On a 3rd browser tab I open up Adobe Firefly, type the article’s title into the generative AI field and expand on it to describe what we need. Then I paste the final image into the draft, so the student has a compelling visual to review along with the text.
Sometimes the student’s essay is more of a personal account, rather than focusing on a specific research paper. Then we’ll take their standard headshot (or shoot a new one) and use Photoshop’s GenAI functionality to expand the background to 1920 x 1080 pixels for the web.
Ethics Check
A reasonable question arises: is this cheating?
IMHO, no, that’s not what we’re doing here. We’re using AI to generate a first draft of a professional personal statement, after ingesting the student’s own responses, publications, lab notes and other materials. The trained LLMs know how to craft a strong message, within pre-set restraints, delivering an output which the student can refine.
Essentially, the multiple AIs have been tasked with building a highly-structured container of content, to our requirements. AI is a tool to translate deeply intellectual findings to a suitably wide audience.
Once published, these articles serve as foundations for media training. The students and I practice going over these talking points in my office, so they can go and pitch their work effectively to funders, employers, and journalists in the future.
These essays are demonstrating how systematic AI integration can deliver significant operational value while ICT cultivates researchers who can communicate with confidence and clarity.
Internships and Storytelling: Engaging Students at All Levels
This summer we extended this essay process from our graduate students to our intern cohort. Between 2005 and 2025, the institute has hosted 660 interns, including 116 from Minority-Serving Institutions (MSI), along with cadets from West Point, the Air Force Academy, and ROTC programs.
Even though they’re at the very beginning of their careers, the results were engaging, thoughtful – and deeply-rooted in academic research: Kaelyn Ellison explored generative AI for human-computer interaction; Viviana York automated scenario-based learning experiences using AI; Camerin Lee chronicled a multimodal capture system for studying human-human interaction; Ella Park developed a multi-agent framework for generating high-quality multiple-choice questions; Oscar Arenas evaluated LLM-generated Spanish translations of science texts for K–12 students, and Joshua Shay Kricheli explored collaborative multi-agent AI systems.
Raising Visibility, Measurable Results
The essays are amplifying ICT’s visibility. Over the past year, LinkedIn engagement rose by 51 percent—a reflection of how student-shared backstories extend the reach of research beyond the lab. Visibility at conferences has similarly increased: students arrive at international meetings such as ACL, EMAS, and IVA, with their articles already published on ICT’s site and social media channels.
The cumulative effect is a pipeline from lab to industry and academia. Students arrive focused on research; they leave with deeper expertise, public-facing articles, and confidence from media training. For ICT, this strengthens recruitment, alumni relations, and its reputation for innovation.
The streamlined production process has enabled ICT to increase content output dramatically while maintaining editorial quality – a framework that could be adapted across research operations.
Without my AI assistants, this process would be a resource-intensive manual process (reading, digesting, abstracting and taking highlights from research publications, drafting, spell-checking, grammar review, tone tweaking, editing to final draft.) Using AIs has created multi-department output from a single (human-in-the-loop) creative.
Through this communications strategy, every student—intern, graduate, or doctoral—has the opportunity to carry forward both knowledge and storytelling skills. By embedding storytelling into research training, students learn to articulate their work clearly, engage broader audiences, and build professional backstories that support long-term career success.
The essays are tangible outcomes of this strategy, demonstrating how ICT cultivates researchers who can communicate with confidence and clarity. They reflect not only the excellence of ICT’s students but the institute’s commitment to preparing the next generation of researchers to be both innovators and effective communicators, capable of sharing their ideas with the world.
//