FOIA Requests vs Civil Discovery: Why the ESI Workflow Differs

One collection, two rulebooks: a records request answers to a statutory clock, a discovery request to the court's schedule.

A FOIA request and a discovery request can reach into the same mailbox, the same shared drive, the same weary custodian, and yet they ask for something different in kind. Under FOIA, as the Air Force Compliance Division's guide to the statute puts it, "any person has a right, enforceable in court, of access to federal agency records," limited only by nine exemptions and three special law enforcement record exclusions. Discovery offers no such open door. It is bounded instead by Rule 26(b) proportionality: the needs of the case, the amount in controversy, the parties' resources, the importance of the issues, and the importance of the discovery in resolving them.

The strange hinge between the two is Exemption (b)(5), which shields agency memorandums that "would not be available by law to a party ... in litigation with the agency," so that the records statute quietly borrows its measure from litigation itself. Perhaps that is why the workflows blur. They are not the same. Courts have accepted that an agency may route every request through full exemption review rather than a faster categorical channel, and that this choice, slow as it may feel to the requester, does not by itself violate FOIA. Which leaves whoever holds the collection with a question that discovery habits cannot settle alone: whose rules govern this set of documents, and whose clock is already running?

How is a FOIA request different from litigation discovery?

A FOIA request asks an agency for records it already holds under a public-access statute, while civil discovery compels a litigating party to produce relevant, proportional evidence under court rules.

Before any collection begins, I would put three questions to every demand that reaches a legal team:

1. Who is asking, and under what authority? Any member of the public under the statute, or an opposing party under the rules of procedure.
2. What does the demand reach? Records that exist today, or whatever material the claims and defenses make relevant.
3. Whose clock is running? A deadline fixed by statute, or a schedule negotiated between counsel and blessed by the court.

The answers rarely coincide, and that quiet divergence is where the trouble begins. In a 2024 memorandum opinion, Judge Beryl A. Howell of the U.S. District Court for the District of Columbia described a suit in which the Washington Lawyers' Committee alleged that the Bureau of Prisons had failed to provide records responsive to over forty separate FOIA requests, in violation of the statute's timeframes, and that the agency followed a "policy and practice" of failing to run a search reasonably calculated to uncover all records requested. Nothing in that complaint turned on relevance. Everything turned on time and on search.

The common assumption is that a records request is simply discovery without a judge in the room, and yet the two regimes reward almost opposite habits. Discovery fights are about how much. FOIA fights are about how fast and how thoroughly. A useful first question, then, is whether the records a request names already exist, because many agencies will not aggregate data into a new record to answer one.

What the two do share, strangely enough, is the need to prove afterward what was done. In our platform, originals are immutable and content-hashed, each action lands in an append-only audit trail, chain of custody is documented, and a fail-closed privilege gate sits in front of production, so the process holds up if opposing counsel challenges it. The same reasoning behind a privilege gate that keeps privileged ESI out of review applies when an agency must later account for every page it withheld.

Enterprise platforms supply that defensible spine of Bates numbering, privilege and chain of custody; newer AI entrants supply cited answers and chronologies. Relevant was built as the combination rather than one or the other, and in our experience the economics follow from it, with AI-assisted review running cents per document against the dollars per document, and roughly $19K per GB, of manual review. The research section below walks through these findings.

Which is why the place to begin is not with the differences at all, but with the collection steps the two workflows hold, almost unnoticed, in common.

Which ESI collection steps do FOIA and discovery share?

Both workflows find custodians, pull email and files, and must trace every answer or withholding back to the exact source document that supports it.

The shared mechanics, the part a dual-practice team can genuinely build once, come down to three movements:

• Locate the sources. Identify the people and systems that hold the records, from mailboxes to shared drives.
• Collect without altering. Preserve email, documents and spreadsheets in a form whose integrity can be shown later.
• Review and record. Search, code and log each decision so the process can be reconstructed if anyone asks.

None of this changes when the demand arrives on agency letterhead instead of a Rule 34 caption. The gaps that one-click cloud collection tools tend to miss follow a collection into either regime, quietly, as if the regime were beside the point.

And yet the setting in which these steps run could hardly be more different. The Justice Department's Office of Information Policy reports that FOIA requests received in fiscal year 2025 exceeded 1.7 million, a 13% increase over the year before, and that agencies processed a record 1.6 million, a 9% rise. Requests received have climbed every year since FY20. Output set a record. Intake still won.

That gap is the tension at the center of this comparison. A litigation team can sometimes absorb an over-broad collection, because the schedule bends and the cost is argued later. A records office cannot. Every extra custodian swept in on a discovery reflex becomes another set of pages someone must read against a statute, in a queue that is already longer than the year.

Here, I think, the habit worth importing from good litigation practice is not breadth but verification. In our platform every output links back to the exact exhibit it came from, so an attorney can verify it in one click before filing; that is the structural answer to the sanctions risk made famous by Mata v. Avianca, and it serves a withholding log just as well as a brief. Where the records include personal or sensitive material, Relevant can run inside the client's own AWS account under the client's own keys, a trust posture no consumer AI tool and almost no small-firm tool can match.

The proof, in the end, is less in description than in watching. In our experience the most persuasive demonstration is to bring a messy collection and a hard question and see the system read, code and cite that evidence directly, rather than read a case study about someone else's.

Which leaves the harder question of who, exactly, is left in the records office to do the reading.

What happens when the 20-working-day clock is missed?

A missed deadline drops the request into backlog, and with reviewers scarce, relief comes from narrower scope plus private review that runs single-tenant or inside your own AWS account.

A request becomes backlogged, in the federal sense, the moment an agency fails to answer within the statutory timeframe of 20 working days. Federal News Network reported in March 2026 that the Defense Department's FOIA backlog rose 42% to more than 30,000 cases by the end of fiscal 2025, as components reported a 37% loss or turnover in FOIA officers. At the Defense Information Systems Agency, FOIA operations rested with a single individual; at the Defense Technical Information Center, staff was "reduced to zero." HUD's office lost 40% of its staff while requests rose 80% since fiscal 2024.

The clock, it turns out, is less a deadline than a measure of who is left to read. Calendar the date it lapses on the day a request arrives. Then decide, well before that date, what a partial first release could cover.

There is a strange comfort in that reporting, though, because it names the remedy as clearly as the wound. Commerce's chief FOIA officer blamed "inefficient search and review processes" alongside staff loss, and Interior began using automation to streamline communications around eDiscovery searches and a document review platform for redaction. In 2024, a vendor executive selling AI FOIA tools argued that integrating AI could be a "game changer" for backlogs; the same 2026 reporting found that many agency AI efforts were early and had largely failed, so far, to make much of a dent. Tools alone, it seems, do not shorten a queue that broad requests keep lengthening.

For a team that answers records requests and also litigates, I would resolve the tension in three moves:

1. Scope before you collect. Name the record types, custodians and date range that answer the request, and resist the discovery reflex to sweep wider.
2. Review on the shared spine, log on the statute. Use one platform for collection and coding, but record each withholding page by page against the exemption that supports it.
3. Keep the data under your own control. Personal and sensitive records should not leave your environment to be reviewed.

On that third point, our processing runs single-tenant, or inside your own AWS account under your own keys, with no vendor retention and no model training, so nothing you feed it trains anyone's model or waives privilege. We built Relevant to be operable and affordable for a solo or small firm on a single messy matter, the segment that Relativity and Everlaw price out and cannot serve without an administrator; the trade-offs are laid out in what a small firm should pay per matter versus an annual platform.

The thinness reaches below the federal level too, for one small school district's sysadmin wrote in 2024 that it has no in-house FOIA officer and pays outside counsel to review each request and draft the response. The research section above walks through these findings.

What should a team that handles both FOIA and discovery do next?

Keep one collection spine, then split the work at scope: run each records request against its statute, exemptions and clock, and each production against proportionality and the court's schedule.

The conventional wisdom treats a records request as discovery with a different caption, and I suspect the next two years will be unkind to that habit. Demand is still outrunning the people who review it; even in a record year, requests arrived faster than agencies could close them. A discovery-shaped search, wide and cautious and patient, was never designed to finish inside a statutory window while the queue keeps lengthening beneath it.

What I would build instead is narrower, and in some sense stranger. Collect once. Then let the withholding decision follow the statute rather than the reflex. The Air Force Compliance Division's guide describes the (c)(1) exclusion, which lets a federal law enforcement agency, under specified circumstances, shield "the very existence of records of ongoing investigations or proceedings," a kind of silence that sits oddly beside the logging instincts of anyone trained on privilege logs. The same documents can live in two worlds at once.

The first answer is also rarely the last. In 2025, requesters were filing considerably more appeals than before, and agencies often released more information on appeal, so whatever exemption log the first review produces should be written to be reread by someone who fully expects to win part of it back.