AI’s Answer to Bad FOIA Requests
Irvin McCullough
Three years ago, a friend told me artificial intelligence could write. Curious, I asked a ChatGPT predecessor to draft a FOIA request. The result confirmed my suspicion: interesting technology, terrible requests.
When OpenAI released ChatGPT later that year, I tried again. This time, it was notably better, though still far from something I’d submit. Then came GPT-4. Suddenly, things changed, and with some careful prompting, a polished, submission-ready request appeared. It felt almost magical—yet the drafts still needed my edits and expertise to reach their full potential.
But who says AI should conjure perfection on its own? What if, instead, it collaborated with humans, evaluating and refining requests until the pair summoned something truly excellent?
Good FOIA requests benefit everyone. Behind every request are government employees tasked with deciphering unclear language, navigating disorganized databases, and making quick judgement calls. The most valuable advice I received from a FOIA mentor was simple: remember there’s a human on the other side of your request.
I asked over a hundred FOIA officers what distinguishes a good request from a bad one. Their responses centered around three main issues with weak requests: they’re unclear, they cast too wide a net, or they target documents exempt from disclosure. Bad requests create a lose-lose situation—requesters don’t get the information, officers struggle to deliver, and agencies drown in ballooning backlogs.
My co-founder and I created FOIA Friend to address exactly these problems. FOIA Friend combines AI’s analytical strength with human judgment, using a smaller, specialized model trained to evaluate request quality across three dimensions: specificity, scope, and FOIA compliance. It leverages more powerful large language models, similar to OpenAI’s GPT series, to suggest precise improvements that increase a request’s quality. Users incorporate these refinements directly, drafting clear, focused requests to submit through our platform.
Suppose someone asks the Department of Defense for “all documents relating to media visits to the Guantanamo Bay detention facility.” FOIA Friend would rate that request low on specificity and scope, and suggest targeted improvements, like narrowing it down to visitor logs, policy guidelines, or press briefing transcripts from specific years. Small changes like these significantly boost the request’s quality, making it easier for FOIA officers to find the right documents and fulfill the request quickly. That means less burden for their office and shorter waiting times for the requester.
Frequent FOIA filers might use the tool to sharpen requests incrementally, adding specific offices or record custodians suggested by the larger model. But other beneficiaries include student journalists, who can use FOIA Friend as an educational aid to learn best practices. When I was in college, I often wondered whether the Department of Homeland Security was surveilling my classmates. Yet even if I’d decided to file a request, it wouldn’t have been very good; I simply didn’t know what a great FOIA request looked like, much less how to write one.
FOIA has long been dominated by those few journalists and advocates with the time and expertise to craft effective requests. While these folks are often willing to help others climb the learning curve, tools like FOIA Friend can democratize access to agency records, allowing young reporters and ordinary citizens to participate in government oversight with confidence.
Most importantly, these refinements could eliminate countless hours wasted on emails and phone calls clarifying vague requests—often weeks, months, or even years after submission. “Writing a request and responding to a request is a cooperative venture,” one officer told me, and cooperation takes time. Given that thousands of people process over a million requests each year across federal agencies, sharpening the requests at submission could free up hundreds of thousands of hours, resulting in fewer delays and a greater number of fulfilled requests.
The goal isn’t more requests—although more people should feel confident filing them—but better ones. Better requests mean fewer rounds of clarification, less wasted effort, and a government that responds faster and more effectively. FOIA embodies a promise: that the public has a right to public documents. Technology alone won’t fulfill that promise, and neither will human effort in isolation. The future of government transparency lies in this partnership between human curiosity and AI capabilities, each enhancing the other.
With that collaboration, we might finally make government information truly accessible to all. Not because AI made it possible, but because AI helped us make it happen. Tools like FOIA Friend will empower all of us to demand more, ask better, and hold our government accountable like never before.
Irvin McCullough is the co-founder of FOIA Friend and a consultant who works with whistleblowers.