On October 28, I joined Curt Wood, senior fellow with the Center for Digital Government, for a live webinar hosted by GovTech and Naviant: “Public Records Requests Are Surging: Is Your Agency Ready?”
We spent an hour on what is actually driving the current surge in public records requests, where agencies lose the most time responding to them, and what a modern, defensible records program looks like. If you missed it, here is the recap, including what attendees told us in four live polls.
Highlights
- Volume and complexity are both up. Requests now come through portals, email, phone, and walk ins, and people are asking for email threads, text messages, and body camera footage, not just a single file.
- Intake and delivery are the light parts of a request. The real hours go to searching across systems, applying redactions, legal review, and preparing for appeal.
- AI is being used to audit records requests on the requester’s side, which is driving more appeals and burning more time.
- A mature records program has a single system of record, defined roles, written SOPs, documented exemption standards, measured service levels, and a clear audit trail, with or without new technology.
- Accessibility (WCAG 2.1) has two deadlines coming: April 2027 for agencies over 50,000 constituents, April 2028 for everyone else. It belongs in the process from day one, not bolted on at the end.
- Human in the loop stays the rule for any agentic or AI tool we use in this process. Technology gets you to a decision faster; a person still makes the call.
Why Public Records Requests Are Getting Harder, Not Easier
Three forces behind the pressure agencies are feeling right now: volume, complexity, and AI.
Volume
Requests arrive through online portals, email, phone calls, and walk ins, and people are asking for more than a single file.
Complexity
Agencies have to search through email threads, text messages, body camera footage, and records scattered across a shared drive to fulfill a request. In addition, not all records are digitized. And if they are digitized, it does not automatically make it easier to find because a file can be electronic and still not searchable. The hours build up in these complex steps: finding the information across a bunch of different systems, reviewing those records against exemptions, and the redaction process.
AI
With AI platforms being widely available to the public, AI is increasing the pressure to deliver accurate and timely requests. Agencies are reporting being bombarded with lengthy AI generated submissions overwhelming the agency making it nearly impossible to deliver within a reasonable time. At the other end of the process, agencies have noticed fulfilled requests being appealed with unnatural speed. This suggests people are using AI to verify their request was fulfilled accurately.
The combination of these challenges plus the public’s expectation of immediate responses (like we’ve all grown used to in everyday life) has made Public Records Request feel like an impossible feat.
And why does it feel so impossible?
Our live poll offers one more reason. We asked attendees which options best describe their agency’s public records process today, and they could choose more than one:
71% said their process is mostly manual: email, spreadsheets, and shared drives
48% have workflow tools for intake, routing, and deadline tracking
16% use software to extract, classify, or redact content in documents
3% use AI-assisted search across their records and systems
Look at where the tools sit. Workflow tools cover intake, routing, and deadlines, the lighter parts of a request. Only 16% have software that helps with redaction, and only 3% have AI-assisted search. Those are the two steps where the hours build up. And because attendees could choose more than one answer, some who have workflow tools also said their process is still mostly manual. It doesn’t have to stay that way. Below is guidance on where to start with your process to make it more manageable.
Where the Hours Actually Go
I walked through the life cycle of a request. Intake and delivery, the first and last steps, are usually light. The middle is where the hours build: searching across systems, applying redactions, running the request past legal review, and preparing for a possible appeal. One complex request can take ten times longer than a simple one.
My recommendation for any agency leader: track the process for a few weeks, even roughly.
- How long does each step take?
- How many rounds of clarification happen before the work can even start?
- Which requests are coming close to, or missing, the statutory deadline?
- And where does work actually get stuck?
We asked attendees the same question in a live poll: Where does your agency experience the greatest public records management challenge? Here are their answers, starting with the most common challenge:
Limited staffing, disconnected systems and manual workflows (42%)
Finding and collecting responsive records (28%)
Reviewing records and accurately redacting sensitive information (17%)
Managing request intake, tracking and deadlines (14%)
The results back up what we see in the lifecycle. Limited staffing, disconnected systems, and manual workflows are the conditions that make the complex middle of a request so heavy. Volume gets attention, but complexity is where the workload piles up. That does not mean the front end can be ignored. When AI-generated submissions flood in, intake is where the pressure shows up first.
Take the Pressure off Intake with Self-Service and AI Supported Triage
One audience member asked what an agency can do when it is flooded with requests, including AI-generated submissions. My answer was to start at intake. That is where an agency can keep routine work off staff desks and make sure every request that reaches a person is ready to work.
The first lever is self-service. When requesters can find status and information on their own, the cost to the agency stays low. Once that self-service breaks down, people start calling and emailing instead, which increases the cost of handling the request, and if they give up and show up in person, that cost can far surpass the original self-service cost.
The second lever is an AI research assistant that triages and preps the work for staff. The assistant can review an incoming request, flag when the ask is unclear, and draft clarifying questions. That draft goes to a person at the agency for approval before anything goes back to the requester. The tool prepares the groundwork, but a person still decides.
What a Mature Public Records Program Actually Looks Like
When asked what a modern program should look like, I said a mature one does not depend on whether a specific person is at their desk that day. It does not live in someone’s head, inbox, or personal spreadsheet. Instead, I described a program built on a few concrete pieces:
- A single system of record for every request
- Defined roles for who handles intake, searching, review, and final approval
- Written standard operating procedures (SOPs) for the most common request types
- Documented exemption review standards, so judgment calls are consistent
- Measured service levels: time to acknowledge, time to close, deadline compliance, backlog age, and appeal rate
- A defined audit trail showing what was searched, who was involved, what was withheld, and why
None of that requires technology on its own. Technology makes a mature process easier to sustain, but the process has to exist first.
So what keeps agencies from getting there?
We asked attendees: What is the biggest barrier to improving your records process? Here are their answers, starting with the most common:
Staff time to lead a project (39%)
Records spread across too many systems (35%)
Budget (22%)
Not a leadership priority right now (4%)
Concerns about AI accuracy and governance (0%)
Staff time and scattered records topped the list, the same conditions that make the middle of a request so heavy. That is why defined roles and a single system of record sit at the core of a mature program. Budget came in third at 22%, and no one picked concerns about AI accuracy and governance as their biggest barrier. Governance still matters, though, and so does one requirement with a firm deadline attached: accessibility.
Accessibility Belongs at the Start of the Process, Not the End
I pointed to two approaching deadlines tied to WCAG 2.1 (Web Content Accessibility Guidelines, the standard for making digital content usable with a screen reader and keyboard): agencies serving more than 50,000 constituents must comply by April 2027, and agencies at or below that threshold have until April 2028. If a constituent cannot request or read a public record, it has not really been made available to the public, and the penalties for missing the standard are significant.
Treating accessibility as a website checkbox addressed at the end of a project is the old pattern. It belongs in the requirements from day one, because it touches the request form, the status updates a requester receives, the documents produced, and the redaction method used. Add it after the fact, and an agency ends up remediating every response by hand, which is exactly the manual work a modern process is meant to remove.
Redaction and Accessibility Have to Work Together
Redaction remains one of the most labor-intensive parts of the process. AI tools can now review large record sets quickly, flag likely redaction areas for a person to confirm, and work with handwritten documents in ways I would not have believed possible ten years ago.
I tied redaction directly back to accessibility: when a section of a record is redacted, tagging it with the reason, citing the relevant statutory exemption, for example, keeps that information available to someone using a screen reader. The audit trail then works for every reader.
I used the term agentic technology, meaning automation that can carry out a multi-step task rather than just generate a one-time answer, to describe some of these newer tools, and I drew a clear line around how it should be used:
“Human in the loop is the most important thing anytime you’re using one of these automation technologies. These agents aren’t making decisions on their own, without a good governance process, which is the human that is doing the review.”
Where to Start
Asked for closing advice, I pointed agency leaders back to the same starting point: look closely at the current process before buying anything. Track requests for a few weeks, note where clarification is needed, check deadline data for near misses, and find out where the hours are actually going.
From there, standardize. Route every request into the same process, regardless of whether it came in through a portal, an email, a phone call, or a walk in, so it can be tracked and reported on consistently. And check the baseline: is the underlying data secure, are permissions set up correctly, and can your systems actually be searched the way a modern tool needs them to be?
We closed with one more poll: Where would your agency most benefit from automation? Here are their answers, starting with the most common:
Search, classification, and document collection (56%)
Redaction, review and quality assurance (28%)
Reporting, audit trails and end-to-end workflow coordination (12%)
Intake, routing, and deadline notifications (4%)
Search won by a wide margin, which tracks where the hours go. But any tool is only as good as the records and permissions underneath it.
Audience Questions
Can AI tools access an agency’s sensitive data behind the scenes?
My answer started with a caveat: digitizing a record does not automatically make it searchable. A record that has not been OCR’d, meaning converted from an image into text a computer can search, is nearly as hard to use as paper. The real starting point is making sure security and permissions are configured correctly before any search tool touches the data, so nothing sensitive surfaces in results it should not.
How can an agency create an AI agent? Are there resources available?
The right approach varies by agency. Some already have a vetted tool running in one department, a desktop ChatGPT or Gemini account, for example, and come to us to build governance around it so the rest of the department can use it sustainably. Our approach is to consult directly with the agency’s own AI governance team so the right people are involved in that conversation from the start.
The Bottom Line
Technology does not replace a mature process, but makes the process easier to sustain. The agencies in the best position are the ones that understand their own data, their own statutes, and their own bottlenecks first, and then bring in automation to support that process rather than patch around a missing one.
Want to see the full conversation? The recording covers each of these topics in more detail. And if you are ready to look at where your own process gets stuck, our public sector team can help you figure out where to start.
