Stacy L ChinFounder, KeepYourEquity · I get the government to write you checks

Weak Proposals & Federal Trouble

24 Feb 2026 · 1 min

Watch on YouTube

1:21

In this monologue, Stacy Chin reflects on her decade‑long experience writing SBIR and other grant proposals and how AI tools like ChatGPT, Claude, and Gemini have changed her workflow. She notes that while AI can generate drafts quickly, the output often lacks the technical depth and rigor needed to win competitive awards. Through trial and error she developed a formula that balances speed with the precision reviewers expect, concluding that AI should be treated as a co‑pilot that extends, not replaces, expertise. She warns that misuse of AI can produce weak proposals and trigger serious federal repercussions. Specifically, she cites the NIH’s new policy that rejects applications substantially developed by AI and may withdraw awards if post‑award AI misuse is detected, illustrating the risk with a case of a PI who submitted over 80 clearly AI‑generated applications in a single year.

Chapters

When I first started using these tools, I love how quickly they can churn out drafts. If you took a deep dive at the text, I realized the content wasn't at the level of rigor that I wanted it to have. In other words, it didn't have the technical depth I knew would make a difference between an average SBIR and a winning one, especially after working in the space for over a decade. As I was experimenting with more tools like ChatGPT, Claude, and Gemini, I realized that wasn't the case. Yes, it took a lot of trial and error to figure out a formula to learn what worked what didn't work, and how to strike that balance between speed and the precision that reviewers are looking for. And then it hit me. AI will not replace expertise, but rather it would extend it. But that's only if you use it the right way. AI rather won't be your competitor, but rather it should be your co-pilot. But with every great tool also comes with its dark side. Use AI the wrong way and it can backfire really bad, leading to weak proposals, messy outputs, and even some cases trouble with the feds. And we don't want that. For example, NIH recently put out a policy policy. They will not accept any applications substantially developed by AI, and if they detect post-award misuse, they may withdraw your application completely. That means hours and weeks and months of wasted time on your end. In fact, they even called out a PI who submitted over 80 applications in a single year, all clearly AI-generated. And now that's wild.