Policy series · Article 8 of 8
From Government Websites to AI Citation Infrastructure: A Public Policy Discussion
Government publishing policy has changed each time a new kind of reader arrived. This closing article traces that history, asks whether AI systems are the next reader, and sets out a staged way to decide.
A short history of the baseline
- 2018. The 21st Century IDEA Act. Public websites and digital services become expected to be accessible, searchable, secure and usable on mobile devices.
- January 14, 2019. The OPEN Government Data Act. Enacted as Title II of the Foundations for Evidence-Based Policymaking Act, it makes public data assets machine-readable and requires a data inventory.
- September 22, 2023. OMB M-23-22. Agencies are asked for structured, metadata-rich, machine-readable content, crawl access and, where interest is high, bulk data or an API.
- November 2024. NIST AI 100-4. A federal assessment of provenance techniques for digital content, with their limits.
- April 3, 2025. OMB M-25-21 and M-25-22. Governance and procurement rules for the federal government’s own use of AI.
- December 2025 and June 2026. The REAL Act and the AI Labeling Act. Proposals on labeling AI-generated content.
- June to October 2026. Public reports. CRS testimony on AI summaries, the ISD study of chatbot answers on voting, and the New York Attorney General’s warning.
The pattern
Each step responded to a new reader. People needed accessible pages. Developers and researchers needed reusable data. Mobile users needed small screens. Each time, policy added something to the baseline that agencies were expected to provide.
AI systems are now an intermediary between agencies and the public. Whether that warrants a further addition is the policy question. It may. It may also be that existing practice is enough, or that the right response is narrower than a general standard. The evidence to settle that does not exist yet.
Infrastructure, not only a website
A website is something people visit. Infrastructure is something other systems rely on. If AI systems come to depend on official sources, the question becomes what those systems rely on: pages they must interpret, or records that state their own authority and time and can be checked. An AI Citation Registry and an AI-readable feed are one answer. Others are possible, and any workable policy should be open to them.
Five questions for policymakers
- Is the problem measured? What share of AI answers about an agency’s notices are wrong, stale or misattributed, and in which categories?
- Which categories come first? Emergency, election and public-health information are the likeliest.
- Which fields are required? Authority, jurisdiction, source, time, integrity check and status.
- Who vouches for an authority? What is the rule for establishing that a publisher is who it says it is?
- Who verifies and who measures? A standards body, an oversight office or the agencies themselves.
A staged path
| Stage | What happens | Main trade-off |
|---|---|---|
| 1. Measure | Agencies and oversight bodies test how AI systems represent official notices. | Methods vary; needs a shared approach. |
| 2. Pilot | A few agencies publish records and compare results over time. | Limited reach; results need honest reporting. |
| 3. Guidance | GSA, NIST or OMB issue recommended practice drawn from the pilots. | Voluntary; uptake uncertain. |
| 4. Standard | A recognized body specifies fields and verification. | Must stay vendor-neutral and maintained. |
| 5. Statute or mandate | Congress or a state legislature requires conformance for defined categories. | Hard to reverse; most useful only after earlier stages. |
On the evidence reviewed for this series, measurement and pilots are the stages the record supports today. Later stages depend on what they find.
What this series has covered
Each article took one question: the overall gap in current policy, publishing standards, provenance standards, procurement, agency policy, accountability and the role of Congress. The common finding is modest: the public record documents a problem, and there is no independent evaluation yet of any remedy. Careful measurement is the sensible next step for everyone involved.
Sources
- 21st Century Integrated Digital Experience Act, Public Law 115-336
- Foundations for Evidence-Based Policymaking Act of 2018, Public Law 115-435 (Title II, OPEN Government Data Act)
- OMB Memorandum M-23-22, Delivering a Digital-First Public Experience (September 22, 2023)
- NIST AI 100-4, Reducing Risks Posed by Synthetic Content: An Overview of Technical Approaches to Digital Content Transparency (November 2024)
- OMB Memorandum M-25-21, Accelerating Federal Use of AI through Innovation, Governance, and Public Trust (April 3, 2025)
- Hunton Andrews Kurth, OMB issues revised policies on AI use and procurement by federal agencies (summary of M-25-21 and M-25-22)
- Rep. Foster and Rep. Sessions, press release introducing the Responsible and Ethical AI Labeling (REAL) Act (December 10, 2025)
- S. 4915, AI Labeling Act of 2026, 119th Congress (introduced June 24, 2026)
- FedScoop, CRS tried to use AI but less than 3% of results met their standards (June 25, 2026 House Administration Committee hearing)
- The Hill, reporting the Institute for Strategic Dialogue study of chatbot answers to voting questions (September 4, 2026)
- New York Attorney General, warning about AI-generated election information (October 14, 2026)
Sources were read in the original or, where noted, through a published summary. Statuses and dates are as of October 18, 2026.