AI, Public Diplomacy and the Hanover Institute: What Foreign Ministries Should Learn
How generative engine optimisation, source provenance and AI-mediated information environments are reshaping narrative resilience for ministries of foreign affairs
In August 2026, investigative reporting drew attention to the Hanover Institute for Public Policy, a newly created website publishing an unusually large volume of policy-style material on politically sensitive issues. Subsequent reporting and disclosure records linked the operation to a government-backed communications campaign and raised a more unusual concern: that the content had been designed not only for human audiences, but also for visibility within AI-mediated information systems.
U.S. Department of Justice records identify Piro, Inc. as registered under the Foreign Agents Registration Act for work involving Havas Media Germany GmbH on behalf of the Israel Government Advertising Agency, LaPam. The Hanover Institute’s current funding disclosure describes the same relationship. Piro separately markets an “AI Story Optimization” service designed to increase the visibility and citation of content within AI-generated answers.
The case is important, but not primarily because of the controversy surrounding the actors involved. It makes visible a broader mechanism that foreign ministries should examine carefully. Generative AI systems increasingly sit between information sources and the people who consume them. As that intermediary role grows, efforts to shape the information environment may increasingly target the machines that retrieve, rank and synthesise information as well as the people ultimately reading it.
It would be premature to infer from one documented case how widespread such practices are. It would be equally unwise to assume that the case is necessarily unique. This article therefore treats Hanover as a case study rather than a generalisation: a lens through which to examine an emerging institutional question for diplomacy. What happens when the information environment of public diplomacy becomes machine-mediated? What should foreign ministries learn from an information environment in which machines increasingly mediate access to diplomatic knowledge?
Assessing AI-era influence operations requires separating intent, mechanism and measurable effect. Evidence of an attempt to influence AI-visible information environments does not by itself demonstrate that a particular model was successfully altered.
That distinction is particularly important here. An AI system citing a source does not prove that the source entered the model’s training data, changed its underlying parameters, determined the answer, or produced a durable change in future outputs. It establishes, at most, that the source became available within a particular information or retrieval pathway under particular conditions.
The underlying vulnerability also did not begin with Hanover. In July 2026, Demos published research examining how generative engine optimisation, or GEO, could be combined with geopolitical information operations. Its report documented Russian FIMI material appearing in AI-generated responses and also described an anonymised case in which a country’s Ministry of Foreign Affairs had reportedly contracted communications firms whose work included efforts to improve the visibility of narratives and affect GPT-generated framing in the United States.
Other research points in a more cautious direction. A peer-reviewed 2025 audit published by the Harvard Kennedy School Misinformation Review found little evidence that the Kremlin-linked material it tested had systematically “groomed” the AI systems examined. Only 5% of responses supported the tested disinformation claims, and the researchers argued that some problematic references were better explained by gaps in credible information — data voids — than by demonstrated manipulation. The studies use different methods and are not directly comparable, but together they reinforce a basic methodological requirement: the existence of an influence attempt, the technical possibility of influence and evidence that influence succeeded are three different propositions.
The significance of Hanover may therefore lie partly in its visibility. Similar attempts would not necessarily be easy to recognise if they resembled ordinary research, commentary, public affairs or institutional publishing. Hanover gives observers an unusually visible case through which to examine a mechanism that is likely to become increasingly important as AI systems take on a larger role in information discovery and synthesis.
1. Observation: a new strategic surface is becoming visible
The Hanover case demonstrates something more specific than the familiar observation that governments conduct online influence or public-diplomacy campaigns.
It shows an attempt to treat the source environment available to AI systems as a strategic surface.
According to The Guardian’s analysis, the Hanover Institute published 124 reports totalling more than 560,000 words between 6 and 14 August 2026. The scale itself does not demonstrate influence. It does illustrate how rapidly a substantial-looking corpus of policy-style material can now be created, published and made discoverable online.
Piro’s own description of its AI Story Optimization service is useful in understanding the logic. The company says it maps the information surfaces AI engines use, identifies relevant queries, creates structured and sourced content, deploys material on owned or third-party properties and repeatedly measures the resulting citation landscape. These are commercial claims about Piro’s services, not independent evidence that the techniques always work or that AI systems evaluate credibility in the way the company describes. They nevertheless demonstrate that influencing AI-mediated discovery is already being treated as a communications objective rather than a hypothetical future possibility.
The relevant question for foreign ministries is therefore not whether this particular campaign succeeded in altering any specific AI answer.
It is what the attempt itself reveals about the evolving information environment.
Traditional public diplomacy has generally assumed that communication ultimately reaches human audiences through journalism, broadcasting, search engines, social platforms, diplomatic engagement or direct institutional communication.
Generative systems introduce another intermediary.
A person researching a disputed foreign-policy issue may increasingly encounter neither the original ministry statement, nor the newspaper investigation, nor the academic paper. The immediate interface may instead be an AI-generated synthesis assembled from previously learned information, live search, retrieved documents or a combination of these.
Diplomats.Digital argued earlier this year in AI and Foreign Affairs: Beyond the Hype that the deeper impact of AI is not simply that institutions can produce more content, but that the diplomatic environment itself is becoming increasingly machine-mediated. Hanover helps make one consequence of that shift concrete: if machines become important intermediaries of information, influencing the information available to those machines becomes strategically relevant too.
The object of influence can move one layer upstream.
2. Mechanism: how could machine-mediated influence work?
The mechanism needs to be described carefully because technically different processes are often collapsed into the broad claim that someone is “influencing an LLM”.
They are not equivalent.
Training data is not search indexing. Search indexing is not retrieval-augmented generation. Retrieval is not an AI-generated search summary.
An attempt to influence one does not automatically influence the others.
A simplified pathway relevant to cases such as Hanover looks like this:
content creation → indexing and discovery → retrieval → synthesis → user answer
At the first stage, material is placed into the public information environment. Search engines, crawlers, indexes or other discovery systems may subsequently make it available. When a user asks an AI-enabled service a question, a search- or retrieval-enabled system may retrieve some of those sources. A model may then synthesise information from that material into an answer, sometimes displaying citations and sometimes leaving much of the underlying information lineage invisible.
This is fundamentally different from demonstrating that information has entered a model’s underlying training corpus or changed the model itself.
Contemporary AI services use different architectures. Some answers rely heavily on pretrained knowledge. Others combine models with live web search, retrieval systems, proprietary indexes or specialist databases. The same system may retrieve different sources depending on the wording and language of the query, location, time, configuration or subsequent changes to its infrastructure.
Actors can therefore attempt to increase the probability that particular sources become discoverable, retrievable or influential within AI-mediated information systems without having any access to the model’s weights or training process.
This is one reason the concept of generative engine optimisation needs careful treatment.
GEO itself is not inherently deceptive. Organisations have legitimate reasons to make accurate information easier for AI systems to discover and understand, just as they have long structured information for search engines and human readers.
The institutional concern begins elsewhere: when provenance, independence or authority is obscured in ways that make coordinated material appear to carry evidentiary weight that it does not possess.
That distinction matters.
Improving the discoverability of authoritative information is not the same as manufacturing the appearance of independent authority.
Nor should we assume that there is a single formula for persuading AI systems that a source is credible. Retrieval and ranking systems depend on multiple signals, many of them proprietary and system-specific. Commercial claims that particular actors can reliably “engineer” authority within AI systems should therefore be treated as claims to be tested, not descriptions of settled technical behaviour.
What has changed is less the existence of information optimisation than the form of mediation.
A conventional search engine generally displays multiple competing results and leaves the user to choose among them. Generative systems can collapse information from several sources into one fluent answer. When this happens, distinctions between original reporting, derivative commentary, independent evidence and coordinated material can become much less apparent.
Research into search-enabled LLMs has already identified substantial gaps between the sources systems consume and the sources ultimately made visible to users. That means provenance is not only a question of whether a citation exists; it is increasingly a question of whether the information lineage behind a synthesis can be reconstructed at all.
For diplomacy, that creates a new institutional problem.
3. Institutional risk: from apparent authority to analytical exposure
Hanover does not establish that AI-mediated influence is widespread or uniformly effective.
It does reveal several risks that foreign ministries should be capable of recognising.
The declining cost of apparent authority
Generative AI has reduced the cost of reproducing many of the textual and visual signals traditionally associated with institutional expertise.
Long-form reports, references, statistical tables, methodological language, professional design and repeated publication can now be produced far more quickly than in the past.
None of those characteristics establishes authority.
The problem is not that AI systems necessarily accept every professional-looking website as credible. The problem is that signals associated with expertise are increasingly inexpensive to reproduce, increasing the burden on both humans and machines to establish where information actually comes from.
For foreign ministries, institutional appearance can no longer serve as an adequate proxy for institutional standing.
Apparent source plurality without independent provenance
The second risk is subtler.
Diplomatic analysis depends heavily on triangulation. If several genuinely independent sources reach similar conclusions, confidence may reasonably increase.
But twenty pages supporting the same proposition do not represent twenty independent sources if they ultimately originate from the same organisation, funding relationship, dataset, communications campaign or generated-content pipeline.
The result can be apparent evidentiary breadth without corresponding independence of provenance.
This matters particularly for OSINT and rapid crisis analysis.
Analysts increasingly need to ask not only how many sources support this claim? but also:
How many genuinely independent provenance chains sit behind those sources?
AI-mediated representation risk
Foreign ministries have developed considerable capability for monitoring traditional media, social platforms, disinformation campaigns, public sentiment and reputational risk.
They may have far less visibility into how their country, policies or actions are represented across AI systems.
Those representations are not necessarily stable. The same question can produce different answers across systems, languages, jurisdictions and time periods. The sources used to construct those answers can also change.
Research has already shown that generative AI can produce systematic frames and evaluations of countries, while the growing use of answer engines means those representations increasingly form part of the environment through which users encounter foreign states.
A ministry can therefore monitor the visible information environment closely while knowing comparatively little about the machine-mediated representation derived from it.
This does not mean governments should attempt to control AI-generated answers. In open information environments, they cannot and should not.
It means those answers are becoming part of the environment in which diplomacy is interpreted.
Internal analytical exposure
There is an equally important risk in the opposite direction.
The AI systems used by citizens, journalists and researchers are increasingly entering professional workflows as well. Diplomats and analysts may use them for discovery, translation, summarisation, horizon scanning, document comparison or preliminary research.
The source environment can therefore affect diplomacy in two directions.
It may shape how external audiences understand a state.
It may also shape how diplomatic institutions understand the external environment.
Research on AI and public diplomacy has already raised concerns about the authoritative form of generated answers and the difficulty generative systems have in preserving uncertainty and competing interpretations. In diplomacy, where context and uncertainty often matter as much as factual retrieval, this becomes particularly significant.
An AI-generated synthesis can be a useful interface to information.
It should not automatically become evidence.
For high-consequence foreign-policy work, the ability to inspect sources, preserve uncertainty and reconstruct the analytical path is part of retaining institutional judgment. This is also consistent with the wider governance problem examined in DD’s AI, Platforms, and Retained Authority analysis: the value of AI use depends partly on whether the institution can reconstruct what happened, understand dependencies and retain meaningful human authority over consequential work.
4. Mitigation: what foreign ministries can do
The institutional response should not be to reproduce questionable tactics more effectively.
A competition in synthetic volume would weaken rather than strengthen the information environment on which governments themselves increasingly depend. It would also erode the distinction between legitimate public diplomacy and manufactured authority.
The more durable response is institutional capability.
Build AI narrative observability
Foreign ministries should begin developing a systematic understanding of how strategically important national positions, policies, crises and disputes are represented across major AI systems.
This should not become a dashboard for obsessively tracking individual chatbot answers. Outputs are variable and the systems themselves evolve quickly.
A more useful function would examine recurring patterns over time:
Which sources appear repeatedly?
Are important authoritative sources absent?
Do representations change materially across languages or jurisdictions?
Are unexpected domains entering the source environment?
Do apparently independent sources trace back to common origins?
The purpose is not control.
It is visibility.
Strengthen provenance analysis
When unusual source clusters or narratives emerge, ministries should be able to examine authorship, ownership, funding disclosure, publication timing, citation relationships and other indicators of common provenance.
This does not require treating unfamiliar sources as hostile.
It requires distinguishing between the content of a claim and the independence of the evidence supporting it.
In AI-assisted analysis, provenance should increasingly become part of ordinary source evaluation rather than a specialist activity invoked only after suspected disinformation has been identified.
Make authoritative information machine-legible
Foreign ministries also have a responsibility on the supply side of the information environment.
Official information should increasingly be:
attributable, timestamped, source-linked, structured, machine-readable, multilingual where relevant, historically retrievable, and clearly updated when official positions change.
None of these measures guarantees that an AI system will retrieve or cite official material.
Nor should official sources receive automatic epistemic priority merely because they are governmental. Governments, like all institutions, make claims that should remain open to scrutiny.
Their value lies in allowing users and systems to identify what the institution itself officially said, when it said it, on whose authority, and whether the position later changed.
That is information infrastructure.
The strategic objective should be:
Machine legibility without manufactured authority.
Protect internal AI-assisted analysis
Ministries should also distinguish exploratory AI use from evidentiary analysis.
AI can be extremely useful for widening the field of view, discovering material, comparing documents and reducing information-processing burdens.
But where generated outputs contribute to consequential assessments, analysts should be able to inspect underlying sources, identify important provenance relationships, verify material claims and record uncertainty.
The appropriate safeguard is not universal prohibition.
It is proportional human review.
Connect the functions institutionally
Finally, this should not become the responsibility of a communications team alone.
AI-mediated information risk crosses public diplomacy, strategic communications, regional and policy desks, open-source analysis, information integrity, crisis functions, technology governance and institutional AI use.
The precise organisational structure will differ between ministries.
The requirement is reliable connection between the functions.
This is consistent with DD’s Narrative Resilience Framework, which treats narrative pressure as an institutional judgment and coordination problem rather than a question of communication volume alone. The emerging AI provenance layer should be incorporated into that wider institutional cycle rather than developed as another isolated technology function.
5. Strategic lesson: narrative resilience now has a provenance layer
Hanover should not be treated as proof of a global pattern.
But neither should foreign ministries design their capabilities on the assumption that it is an isolated anomaly.
The economics of the information environment have changed. Policy-style content can be created and published at extraordinary scale. Search and retrieval systems can expose that material to generative models. AI systems increasingly mediate access to knowledge. And the provenance of the sources contributing to a generated answer may remain largely invisible to the person reading it.
The incentives created by this environment are not specific to one government, one conflict or one geography.
Future cases may involve states. They may involve corporations, political organisations, advocacy groups, commercial interests or actors whose identities are considerably more difficult to establish. Some activities may be intentionally deceptive. Others may be ordinary public relations, legitimate public diplomacy or routine attempts to improve the discoverability of authoritative information.
That is why the analytical task requires greater discrimination rather than broader suspicion.
Intent, mechanism and effect must remain separate.
An actor may attempt to influence AI-generated answers without succeeding.
A source may appear in a generated response without determining its conclusion.
A retrieval system may surface problematic material because of deliberate optimisation, or because credible sources are absent from a particular information space.
And a strategically relevant vulnerability can exist even when measurable downstream effects remain uncertain.
For diplomacy, the institutional challenge is therefore broader than counter-disinformation.
Foreign ministries need greater visibility into the information environments from which machines construct representations of countries, policies and crises. They need better ways to distinguish apparent source plurality from independent evidence. They need official information architectures that make authentic provenance easier to recognise. And they need internal practices that preserve human judgment when AI-generated synthesis enters diplomatic analysis.
This builds on a wider shift already visible across the foreign ministry. As DD’s The Future Ministry of Foreign Affairs argues, AI, narrative pressure, crisis velocity and digital dependence increasingly intersect as institutional capability questions rather than isolated technology or communications issues.
The Hanover case gives that shift a particularly concrete expression.
Narrative resilience in the AI era must therefore extend beyond monitoring narratives to understanding their machine-mediated provenance.
The question for foreign ministries is no longer only:
Who is saying what, to whom, and through which platform?
Increasingly, they must also ask:
From which sources are machines assembling the answer?
Sources / Further Reading
Primary documentation
U.S. Department of Justice — Foreign Agents Registration Act, Piro, Inc., Registration No. 7732.
The public record identifies Piro, Inc. and its foreign-principal relationship with Havas Media Germany GmbH on behalf of the Israel Government Advertising Agency, LaPam, and contains associated registration and informational-material filings.
U.S. DOJ FARA record — Piro, Inc.
The Hanover Institute for Public Policy — Funding disclosure.
The Institute’s current self-description identifies Piro, Havas Media Germany and LaPam and explains its account of the funding and editorial relationship. This should be treated as the organisation’s own disclosure rather than independent verification of its claimed editorial arrangements.
Piro, Inc. — AI Story Optimization.
Piro’s description of how it approaches AI search visibility, source mapping, content deployment and citation monitoring. Its effectiveness claims are commercial claims and should be read as such.
Reporting on the Hanover case
Nick Cleveland-Stout, Responsible Statecraft, 17 August 2026 — “Israel creates fake think tank in likely attempt to dupe AI chatbots.”
Early reporting on the Hanover site, disclosure trail, publication pattern and apparent AI orientation.
Responsible Statecraft investigation
Jason Wilson, The Guardian, 26 August 2026 — “Fake US thinktank set up and funded by Israel sought to game AI for propaganda.”
Detailed subsequent investigation examining publishing volume, technical characteristics and the relationship between the site and the wider communications operation.
Research and wider context
Carl Miller, Hannah Perry and Flynn Devine, Demos, 2026 — GEO for Geopolitics: What happens when AI and information warfare collide.
Research into GEO, RAG poisoning and geopolitical information operations, including a Russian FIMI case study and anonymised examples of state-level GEO activity.
Maxim Alyukov, Mykola Makhortykh, Alexandr Voronovici and Maryna Sydorova, HKS Misinformation Review, 2025 — “LLMs grooming or data voids?”
A useful empirical counterpoint showing why problematic AI references should not automatically be treated as evidence of successful manipulation.
Harvard Kennedy School Misinformation Review study
Luigi Di Martino and Heather Ford, Place Branding and Public Diplomacy, 2024 — “Navigating uncertainty: public diplomacy vs. AI.”
An early public-diplomacy analysis of generative AI, uncertainty, synthesis and the epistemic characteristics of AI-generated answers.
Navigating uncertainty: public diplomacy vs. AI
Ilan Manor and Elad Segev, Policy & Internet, 2025 — “What ChatGPT ‘Thinks’ About Your Country? Sentiments and Frames of AI Geographies.”
Empirical research into how generative AI constructs differing representations of countries and regions.
What ChatGPT Thinks About Your Country
Jason Miklian, 2026 — “How Artificial Intelligence LLM Engines Shape the Global Conflict Information Environment.”
A large-scale study of AI answers across 28 conflicts examining retrievable information environments, GEO exposure and source selection.
Global Conflict Information Environment study
Ilan Strauss et al., Data & Policy, 2026 — “The attribution crisis in LLM search results.”
Research examining the gap between information retrieved by search-enabled LLM systems and the sources ultimately attributed to users.
The attribution crisis in LLM search results
Online sources last reviewed 31 August 2026. Descriptions of the Hanover Institute and Piro’s services reflect publicly accessible pages at the time of review. Subsequent changes to those pages should not be read back into the record described here.
For private briefings or institutional inquiries, contact Diplomats.Digital.