AI use case

AI for freedom of information request processing

AI that helps a public body handle freedom of information and open government requests: logging and clarifying requests, spotting duplicates, searching and deduplicating the records in scope, proposing redactions with the exemption that applies, and drafting the response letter, with an FOI officer deciding what is released.

By Len Debets · Last verified 26 September 2026 · 4 public deployments

EUR 90,000 to EUR 1.5 million
Indicative value per year
A government department that receives 5,000 information requests a year. Worked example, see how it is calculated.

What problem does it solve?

Freedom of information laws give everyone the right to ask for government records, and request volumes keep rising. Each request means finding every relevant record across email, file shares and case systems, removing duplicates, reading everything, redacting personal data and exempt material, and explaining the decision within a statutory deadline. Large requests can involve very large document sets, and similar requests can reach several offices of the same government body.

The work is mostly manual and legal in nature, so backlogs grow and deadlines are missed, which undermines the transparency the law is meant to deliver. Errors cut both ways: over redaction withholds information the public is entitled to, and under redaction leaks personal data.

  • US federal agencies received a record 1,707,197 FOIA requests in fiscal year 2025, 13.7% more than the year before, and ended the year with 339,671 backlogged requests, a 27% increase.2025 Annual FOIA Report Summary (2026)

How does it work?

  1. Log and clarify. Incoming requests are read, logged with key fields and compared with open and past requests, so similar requests are grouped and answered consistently. Unclear requests get a drafted clarification question.
  2. Collect and cull. Records gathered under the search plan are made searchable (including text recognition for scans), deduplicated and grouped by topic so reviewers see what is relevant first.
  3. Propose redactions. Named entity recognition and trained models mark personal data and other candidate redactions, each with the proposed exemption code.
  4. Review and decide. FOI officers and lawyers accept, change or reject every proposed redaction and decide what is released, with a second reviewer for doubtful cases.
  5. Draft the response. The response letter, including the exemptions relied on and appeal rights, is drafted from the decision record for the officer to finalise.
Audience
Employee facing
Autonomy
Copilot
Adoption
Early adopters
Channels
Internal tools, Email

What is it worth?

Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.

No public deployment has disclosed a measurable outcome yet.

Value drivers: Employee productivity, Speed and cycle time, Compliance quality, Inclusion and access.

Indicative value

A government department that receives 5,000 information requests a year

EUR 90,000 to EUR 1.5 million

FOI officer time released per year

How this is calculated

Formula: requests * hoursPerRequest * savedShare * hourlyCost. The low scenario uses every low input, the high scenario every high input.

InputLowHighBasis
Requests per year requests, requests per year3,0007,000Editorial assumption. Replace with your own request log.
Staff hours per request today hoursPerRequest, hours per request512Editorial assumption covering search, review, redaction and response. Large requests take far longer.
Share of those hours saved savedShare, fraction of hours0.150.3Editorial assumption. No agency on this page publishes a measured saving; legal review remains human work.
Fully loaded cost of an FOI officer's hour hourlyCost, EUR per hour4060Editorial assumption. Replace with your own staff cost.

What it leaves out: Time released, not cash saved. It leaves out licence and assurance costs, the value of meeting statutory deadlines, and the cost of a redaction error, which can be far larger than the saving.

Who already uses it?

4 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.

U.S. Department of Justice

United States · Government and public sector · 2025

ProductionGrade B

The Department of Justice reports a department wide entry for FOIA production tools, deployed in January 2025, that add AI to its FOIAXpress request processing: classifying documents, identifying sensitive or confidential information for redaction and removing duplicate documents. The outputs are classifications, recommendations and predictions, and the department rates the use as not high impact because it is not the principal basis for decisions with legal or significant effect. The expected benefits are faster processing, fewer human errors and more accurate, compliant processing; no measured outcome is published.

No outcome disclosed.

U.S. Food and Drug Administration, Center for Drug Evaluation and Research

United States · Government and public sector · 2025

ProductionGrade B

FDA's Center for Drug Evaluation and Research uses the FOIA Redaction (FRED) tool, a generative AI system built by FDA and contractor teams, to help FOIA staff redact records more efficiently and consistently, because redaction is time consuming and FOIA backlogs build up. The tool returns a PDF with boxes around the text it recommends redacting, each with a comment giving the redaction code. Its data are completed FDA Form 483 inspection records in their original and staff redacted versions, and every output needs human review and approval. Listed as deployed since May 2025; no outcome figures are published.

No outcome disclosed.

U.S. Department of the Interior, Office of the Solicitor

United States · Government and public sector · 2023

ProductionGrade B

The Department of the Interior's Office of the Solicitor lists four deployed tools, developed in house, that group incoming FOIA requests: two clustering tools (one embedding based, one density based on term frequency, run in its document review platform), a semantic similarity score and a lexical similarity tool. They identify requests that ask for the same or similar records, including similar requests sent to several offices, so that the work can be coordinated and responses kept uniform instead of duplicated. The inventory gives operational dates of August and November 2023; no outcome figures are published.

No outcome disclosed.

Provincie Noord-Holland

Netherlands · Government and public sector · 2021

ProductionGrade B

The Province of North Holland uses ZyLAB to handle large requests under the Dutch Open Government Act (Woo). After staff draw up a search plan and collect the potentially relevant documents, the platform makes the set searchable, including text recognition for scanned documents, helps staff judge relevance and, on instruction, produces a trial redacted version: generic rules recognise items such as phone and citizen service numbers, while names need their own individual rules, and staff switch the rules on themselves. Staff check every document to avoid too much or too little redaction, and what is released is agreed between the responsible staff and lawyers. In use since October 2021.

No outcome disclosed.

How do you implement it?

A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.

Data you need

  • Request log with past requests, decisions and exemptions applied
  • Access to the record stores searched for requests (email, file shares, case systems)
  • Written redaction rules per exemption, with examples

Systems to integrate

  • FOI case management or request tracking system
  • eDiscovery or document review platform
  • Email and records management systems
  • Public disclosure log or reading room for published responses

Complexity: Medium

Document review and redaction platforms are established products (the Province of North Holland has used one since 2021); the effort is in connecting to the places records live, tuning redaction to the agency's exemptions and building a review workflow that lawyers trust.

  1. 1

    Start with deduplication and grouping

    Grouping similar requests and exact duplicate documents is a lower risk place to start, because no redaction or release decision is automated. Check near duplicate removal with care, since versions that differ can both be responsive. The Department of the Interior's Office of the Solicitor has used request similarity and clustering tools since 2023 to coordinate answers to similar requests.

  2. 2

    Introduce proposed redactions with full review

    Let the tool propose redactions with the redaction code for each, as FDA's FRED tool does, while officers still review and approve every proposal. Measure how often proposals are changed.

  3. 3

    Tune rules to your exemptions

    Configure generic patterns (phone numbers, national identifiers) and individual rules for names, as the Province of North Holland does.

  4. 4

    Draft the response letter last

    Once decisions are recorded per document, generate the response letter from them, so the letter matches what was actually decided.

Guardrails

  • Every redaction and release decision is taken by an FOI officer, never by the tool
  • Second review for documents where the officer is in doubt
  • Redaction burned into the released file, with the original kept securely
  • Personal data in request logs and prompts masked and retained only as the law allows
  • Search plan documented, so the scope of records is defensible

KPIs to instrument

  • Median days from request to response and share within the statutory deadline
  • Backlog of open requests
  • Share of proposed redactions changed by reviewers
  • Redaction errors found after release
  • Appeals upheld against over redaction

Human in the loop

FOI officers and lawyers own the search plan, review every proposed redaction and decide what is released. A second reviewer or team lead checks doubtful documents. Requesters keep their complaint and appeal rights.

Common failure modes

Missed personal data
A name in an image, a signature or an unusual format is not recognised and is released. Reviewers must check every page, and scanned material needs extra care.
Over redaction by default
Accepting every proposal withholds information that should be public. Track changed proposals and appeals.
Incomplete search
AI speeds up review but does not fix a search that missed a record store. Keep the search plan explicit.
Redaction that can be undone
Visual boxes over text that can still be copied from the file. Use tools that remove the underlying text.

What are the risks and rules?

EU AI Act

Minimal risk

Tools that support staff in searching, deduplicating and proposing redactions are not listed in Annex III (point 5(a) covers eligibility for public assistance benefits and services, not access to documents), and every release decision stays with an officer. A public facing request assistant that talks to requesters would carry the Article 50(1) transparency duty.

Guidance

Controls to put in place

  • Documented redaction rules per exemption and a review workflow with sign off
  • Audit trail of proposed and final redactions per document
  • Quality sampling of released documents for missed personal data
  • Access controls on the record sets gathered for each request
  • Register or inventory entry for each AI tool used

Frequently asked questions

Can AI redact documents for FOI requests?
It can propose redactions. FDA's FRED tool marks the text it recommends redacting with a redaction code, and the Province of North Holland uses general rules that recognise phone and citizen service numbers, while names need their own individual rules. In both cases staff review the proposals and decide; accountability stays with the officer.
Where does AI help most in FOI work?
In the volume steps: grouping similar requests, deduplicating and sorting records, and proposing redactions of personal data. Judgment on exemptions and the public interest remains human work.
Is AI for FOI processing high risk under the EU AI Act?
No. Handling requests for access to documents is not listed in Annex III, and good practice keeps every release decision with an officer. If you add a chatbot that talks to requesters, it must tell them they are dealing with AI (Article 50).

How to cite this page

Blits.ai AI Use Case Library, "AI for freedom of information request processing", last verified 26 September 2026, https://www.blits.ai/ai-use-cases/freedom-of-information-request-processing. Licensed under CC BY 4.0. Method: how we verify use cases.

Changelog
  • 27 September 2026: First published

Related use cases

Cross industryBanking

AI for inbound correspondence triage and routing

AI that sorts inbound correspondence before anyone answers it: it takes every inbound letter, email, upload and secure message into one intake, identifies what it is, extracts the key fields, links it to the right customer and account, sets priority and routes it to the right team or workflow, replacing the manual sorting desk.

Deployments
6 public, best grade B
Reported accuracy
91%
Travelers, vendor claim
Government and public sector

AI drafting copilot for civil servants for correspondence, briefings and ministerial replies

A generative AI assistant that drafts replies to correspondence from the public and elected representatives, briefings, submissions and summaries for civil servants, grounded in the department's approved lines, policy documents and case data, with the official editing and approving every word before it is sent or cleared.

Deployments
5 public, best grade B
Autonomy
Copilot
Government and public sector

AI for court and case file summarization

AI that condenses court filings, case files, evidence recordings and earlier decisions into structured summaries, chronologies and draft case reports with references to the source pages, so that judges, prosecutors, tribunal staff and government lawyers find what matters faster, while the person responsible reads the underlying material and makes every legal judgment.

Deployments
4 public, best grade B
Autonomy
Copilot
Cross industryBanking

AI enterprise knowledge search for employees

An assistant that lets any employee ask a question in plain language and get a synthesized answer from the organization's own policies, procedures, product manuals and research, with citations to the source documents and only from documents the employee is allowed to see.

Deployments
4 public, best grade B
Autonomy
Assist
Cross industryGovernment and public sector

AI document intelligence for unstructured forms and documents

AI that takes documents in any format, such as scanned forms, PDFs, photos, emails and handwritten notes, splits and classifies them, extracts the required fields with a confidence score, validates them against business rules and source systems, and sends only the uncertain cases to a person before the data enters the downstream process.

Deployments
5 public, best grade B
Reported accuracy
at least 90%
Ancine, vendor claim