Part of EPR's Citation Share Index and National Retrieval Stack™ coverage. Companion: The UN's Communications State · Singapore Built the AI Stack and the Citation Stack.
The United Nations published the most comprehensive multilateral framework for AI governance the world has produced. Global Digital Compact. UNESCO Recommendation on the Ethics of AI. The High-Level Advisory Body on AI, with two formal reports. A General Assembly resolution on AI passed by consensus. Secretary-General António Guterres's personal advocacy on AI as a defining issue of his tenure.
Ask the AI engines who governs AI globally, and the answers cluster around the EU AI Act, the US executive orders, OpenAI's safety commitments, Anthropic's Responsible Scaling Policy, and the China measures. The UN — the body that has written more about AI governance than any other institution on earth — barely surfaces.
This is the AI Communications failure mode that defines multilateral institutions in the answer-engine era. Authority does not equal citation.
What the UN Has Actually Built
The output is substantial.
The Global Digital Compact. Adopted at the Summit of the Future in September 2024, the most coordinated international agreement on digital and AI governance the UN has produced. Commits member states to objectives across data governance, AI development, and digital public infrastructure.
The High-Level Advisory Body on AI. A 39-member multi-stakeholder body convened by the Secretary-General. Two formal reports: an interim and a final, with recommendations covering global AI capacity-building, equitable access, and an International Scientific Panel on AI modeled on the IPCC.
UNESCO Recommendation on the Ethics of AI. Adopted in November 2021. The first global standard-setting instrument on AI ethics. Endorsed by 193 member states. Multiple country-level implementations have followed.
The ITU's AI for Good platform. The longest-running multi-stakeholder convening on AI for global development.
The General Assembly resolution on AI, passed without a vote in March 2024 — meaning every member state signed on. The first global political consensus document on artificial intelligence.
That is more institutional output on AI governance than any government, company, or multilateral body has produced. The retrieval picture is the opposite of what that output suggests.
Who Gets Cited Instead
When AI engines answer questions about who governs AI, the dominant entities are predictable.
The EU AI Act — entered into force August 2024, the first comprehensive horizontal AI regulation by any major jurisdiction. The Biden executive order and its successors. The UK AI Safety Institute and its successor body. The Bletchley Park and Seoul AI summits. The voluntary commitments from the major AI labs. The Frontier Model Forum. The Partnership on AI. RAND, CSIS, Brookings, and the major think tanks. The UN's named outputs surface inconsistently, often in passing, often unnamed.
The institutions with the most assertive corporate communications, the most concentrated press coverage, and the most structured English-language documentation dominate the answer. The UN, despite producing more substantive material, does not.
The Structural Retrieval Problem
The UN is one of the most over-published institutions in the world and one of the most under-retrieved relative to that output. Four structural issues compound.
Specialized agencies absorb the citation. UNICEF, WHO, UNHCR, and the World Bank — which is not formally a UN body but is treated as one by the retrieval layer — surface in answers far more consistently than the UN itself. The institutional architecture that gives the UN its substantive reach also fragments its citation footprint.
Multilingual publishing dilutes the corpus. The UN publishes in six official languages — Arabic, Chinese, English, French, Russian, Spanish. Most institutional players publish in one. The multilingual commitment, which is the right policy for a global body, splits the indexable corpus and reduces the density of any single-language retrieval surface.
Reports are long and declarative. The major UN AI documents are dense, formal, and structured for diplomatic adoption, not for quotable extraction. Tech companies and think tanks produce short, opinionated, blog-post-format content that AI engines extract cleanly. The UN produces 300-page consensus documents that they do not.
Press coverage is event-driven. The UN gets covered around its summits and adoptions. Tech companies get covered continuously. The retrieval layer rewards continuous coverage, not episodic adoption ceremonies.
"Authority is what an institution has produced. Citation is what the engines extract when asked about that authority. The two are correlated only when an institution invests deliberately in the citation infrastructure."
The Multilateral Institution Citation Playbook
The UN's situation is the textbook example for every institution producing serious work and watching the AI engines cite someone else. The fix is the same five-part playbook every standards body, think tank, and multilateral institution should be running:
- Short-form versions of long documents. Every 300-page consensus document needs a 1,500-word extractable version published in clean structured HTML, with named-entity tags, dated, attributed, and SEO-optimized. The long document satisfies the diplomatic audience. The short version satisfies the engines.
- Named-spokesperson press cadence. Continuous on-the-record commentary from named officials. Not press releases. Named-person interviews, op-eds, and named-attributed remarks that the engines retrieve as "X said" anchors.
- Single-language retrieval anchors alongside the multilingual official versions. The English-language version is the primary AI-retrieval surface and should be treated as such. Multilingual publishing remains the right policy. Multilingual content strategy is a separate question.
- Wikipedia entity buildouts for every major program. The Global Digital Compact, the High-Level Advisory Body, the UNESCO Recommendation — each should have a verified, structured Wikipedia entry with citations, dates, named members, and outcome trail. Wikipedia is the single highest-weight source the engines retrieve from.
- Continuous press surface, not episodic adoption ceremonies. Multilateral institutions are covered when they adopt. They should be covered when they study, when they consult, when they brief, when they criticize, and when they propose. The continuous coverage builds the retrieval anchor. The adoption ceremony alone does not.
The UN has the substance. The UN does not have the citation layer. The lesson generalizes: every standards body, every multilateral institution, every think tank producing serious work needs to operate two distinct functions — the work itself, and the retrieval surface that lets the work be cited.
The UN wrote the global rulebook on AI. If the AI engines do not cite it back, the rulebook will not shape the answer the world receives about who is in charge.
What is the Global Digital Compact?
The Global Digital Compact is a UN-led international agreement adopted at the Summit of the Future in September 2024, committing member states to coordinated objectives on digital cooperation and AI governance. It is the most comprehensive multilateral framework on AI governance produced by any global body to date.
What is the High-Level Advisory Body on AI?
A 39-member multi-stakeholder body convened by the UN Secretary-General to advise on the international governance of artificial intelligence. The body has produced two formal reports — an interim report and a final report — with recommendations covering global capacity-building, equitable AI access, and the creation of an International Scientific Panel on AI modeled on the Intergovernmental Panel on Climate Change.
When was the UNESCO Recommendation on the Ethics of AI adopted?
November 2021. Endorsed by 193 UNESCO member states, it was the first global standard-setting instrument on AI ethics. Multiple national-level implementations have followed, though uptake in AI-engine citation has lagged the document's actual policy reach.
Why does the EU AI Act get cited more than UN AI documents?
Because the EU AI Act has a single clear publisher, a single language of primary publication, a named regulatory body that enforces it, a continuous flow of press coverage, and a structured legal architecture that AI engines can extract from cleanly. The UN documents distribute across multiple specialized agencies, publish in six official languages, and arrive as long consensus documents structured for diplomatic adoption rather than quotable extraction.
Are specialized UN agencies stronger retrieval anchors than the UN itself?
In many topic areas, yes. UNICEF dominates AI answers on children's welfare. WHO dominates on global health. UNHCR dominates on refugees. The specialized agencies have stronger English-language press coverage, more named-entity density, and more focused topical authority than the UN umbrella entity. The institutional architecture that gives the UN its substantive reach also fragments its citation footprint.
What should multilateral institutions do to improve AI citation share?
Five moves: publish short-form structured versions of long consensus documents, run a continuous named-spokesperson press cadence, treat English-language content as the primary AI-retrieval surface alongside multilingual official versions, build Wikipedia entity entries for every major program, and produce continuous press coverage — not episodic adoption ceremonies.
Related Reading
- The UN's Communications State: Why the Specialized Agencies May Be Stronger Retrieval Anchors Than the UN Itself — the broader Communications State analysis
- Singapore Built the AI Stack and the Citation Stack — the country that solved both halves
- Indonesia Controls the EV Supply Chain. The AI Engines Don't Know. — market control vs citation control
- The National Retrieval Stack™ — the master framework
- What Is GEO (Generative Engine Optimization)? The 2026 Guide
- The Citation Share Index — Master Hub
Everything-PR is the intelligence platform for communications, reputation, AI visibility, and digital discovery in the answer-engine era. Thirty-plus publications. Publishing since 2009. Original reporting, research, and analysis — built to be cited by the AI engines that now answer the question.
