This chapter examines how artificial intelligence (AI) is transforming public communication, focusing on productivity, responsiveness, and accountability. It highlights patterns in the adoption and uses of AI for communication across OECD countries. Through survey data, it offers a stocktaking of the degree of scaling and the resources deployed to support teams in using this technology effectively and responsibly. The chapter discusses both opportunities and risks, and the need for guardrails to maintain safety and public trust.
Reforming Government Communication for the Age of Algorithms
Abstract
AI is becoming a new driver of change for public communication. The mainstreaming of generative AI is reshaping how people search for information, how organisations produce content, and how governments can communicate with citizens. In the 2025 OECD Survey of Government Communicators, 35% of respondents expect AI to significantly enable teams to produce more and better communication, and 23% more believe it even will open up entirely new possibilities for more effective communication.
Most communication teams are still at an early stage of AI adoption. While expectations are high, 46% of institutions remain in an exploratory phase, and use is often driven by individual experimentation rather than organisational strategy. This limits the ability of teams to scale AI safely and consistently.
AI is currently being used more to improve existing tasks than to transform communication. Communication teams are mainly using AI for drafting, translation, content adaptation, brainstorming and improving access to public information and services. These uses can ease constraints on time, skills and resources, but remain far from the more strategic applications AI could enable.
AI can strengthen responsiveness and interaction with citizens. Methods such as sentiment analysis, audience segmentation, automated monitoring can be enhanced by AI; chatbots and virtual assistants can similarly help governments better understand public concerns, identify emerging issues, tailor information to different needs and improve access to services.
Scaling AI in public communication depends on trust and safeguards. Because AI introduces risks for transparency, accountability, data quality and over-reliance on automated systems, institutions will need clear guidance, human oversight, testing, transparency and stronger organisational capacity to use AI safely and maintain public trust.
The popularisation of generative AI chatbots like Chat GPT since 2022 has brought the power of Artificial Intelligence (AI)1 directly into the hands of people and organisations all over the world. While AI has been advancing for years in several domains, this moment was a tipping point that made it an immediate new reality for everyday citizens.
For governments, this breakthrough means adjusting to economies and societies that are rapidly transforming how they operate and connect. It also means vast opportunities to reshape the very machinery of government. The OECD has captured these early shifts in its report on Governing with Artificial Intelligence: The state of play and way forward in core government functions (OECD, 2025[1]), offering a broad look at how AI is transforming government.
The report shows how AI has the potential to reshape the public sector by enhancing operational efficiency and maximising value for taxpayers; enabling more tailored, accessible, and inclusive public services; strengthening accountability and anomaly detection; advancing communication and engagement with citizens; and unlocking opportunities for collaboration with external stakeholders. AI is also helping organisations work more efficiently, by improving decision making and supporting civil servants in their day-to-day work and continuous learning.
However, AI adoption also introduces significant risks. Challenges related to data quality, limited transparency, and over-reliance on automated systems can undermine trust and weaken accountability (OECD, 2025[1]). Amid low levels of public trust in institutions (OECD, 2026[2]), governments face a growing responsibility to ensure that the use of AI is trustworthy and is accompanied by appropriate governance arrangements, safeguards and oversight. OECD 2025 Trust survey results show that 35% of respondents are very sceptical about the potential of AI use by government institutions, with zero expected positive expectations.
To make sense of the different emerging applications of AI in government, the OECD (2024[3]) classifies them according to three main impacts. While these concern the government in general, they align broadly with the same opportunities encountered in the communication function:
Productivity: Achieving greater efficiency in internal operations and improving the design, decision making, and delivery of policies and services. This can involve leveraging predictive AI tools for smarter planning, automating workflows, and enabling public servants to dedicate more time to essential, high-impact work rather than routine tasks.
Responsiveness: Making public policies and services more adaptable to the changing needs of citizens and specific groups by refining how they are designed and delivered. This is also achieved by enhancing opportunities for civic participation and providing more tailored public services in a proactive manner.
Accountability: Strengthening oversight and transparency can increase public satisfaction and reinforce the image of government as competent, fair, and responsive. These improvements help build trust in the government’s ability to innovate and drive positive change.
The field of government communication shares much of the same opportunities for efficiency and innovation from AI as other public sector functions. AI can revolutionize how governments communicate by enabling real-time, personalised engagement and improving accessibility. For instance, chatbots can provide instant responses, while predictive analytics can tailor messages to citizen needs.
At the same time, government communication is also highly exposed to the effects that AI is having on the information ecosystem more broadly. For one, certain groups seem to be increasingly substituting web search and navigation with queries to generative AI chatbots to get information. For example, 7% of respondents across 48 countries surveyed by the Reuters Institute for the Study of Journalism (Newman et al., 2025[4]) are using generative AI chatbots routinely for news. For people under 25, the share is more than double and likely to grow rapidly. This trend could upend how governments get information to citizens and interact with them. It also raises concerns that inadequately trained AI models may reduce the visibility of public-sector content or, in some cases, exclude it altogether.
The impact of AI in government communication will thus rest on how well governments prepare for and adapt to an information ecosystem transformed by AI, as much as on how they adopt it in their working methods and use it to innovate. As a matter of sound governance, effective stewardship of AI requires a sufficient level of institutional understanding and capacity. Strengthening this understanding is essential to enabling informed oversight, proportionate regulatory approaches and the responsible, effective and trustworthy use across public administrations.
The introduction of popular generative AI technologies has come with considerable hype, making it difficult to distinguish genuine advancements from overstated claims. For communicators, this has been accompanied by substantial fears that the technology could flood the information space with false or misleading information and sow confusion and distrust (Lovari and De Rosa, 2025[5]).
Nonetheless, most senior government communicators surveyed by the OECD share positive expectations overall for how AI will transform government communication. Figure 3.1 shows that more than one third of respondents expect AI to “significantly enable teams to produce more and better communications.” One in 5 even foresee AI “drastically open[ing] up new possibilities for more effective communication.” Conversely, those who expect it to have a moderate impact on productivity are in the minority.
Despite these expectations, AI adoption is still in its infancy in most teams, often limited to individual experimentation, as discussed below. Similarly, survey responses concerning the use of AI tend to confirm that productivity improvements are the main focus of current applications of AI by communication teams. More innovative, outside-the-box approaches are still some way off for most. This leaves sizeable opportunities for larger and more strategic uptake.
To what extent do you expect AI to transform the field of public communication?
Note: Data shown above refers to 57 responses to the relevant question in the OECD Survey of Government Communicators. Full answer options: “Drastically open up new possibilities for more effective communication”, “Significantly enable teams to produce more and better communications”, “Mostly increase efficiency by freeing up staff time and lowering costs of producing content”, “Moderately improve some processes and automate junior tasks”, “Other” and 2 options which were not selected: “No significant transformation”, “We will not see the real impact of AI for the field for several years”.
Source: OECD 2025 Survey of Government Communicators.
3.1.1. Responsible AI for boosting efficiency and effectiveness in communication
According to the 2025 OECD Survey of Government Communicators, the most common uses of AI across the 57 institutions surveyed fall in the productivity category identified by the OECD (2024[3]). This is the area where AI can deliver rapid gains for efficiency and quality of government communication (Yun, Yun and Xue, 2024[6]). Constraints on time, staff, budgets and other resources are a common problem across communication teams. Such constraints were among the most commonly cited key challenges voiced by government communicators in the 2025 survey, reaffirming a persistent problem already seen in the earlier 2020 survey (OECD, 2021[7]).
There is growing evidence that AI can alleviate these constraints. Studies from related fields have estimated considerable gains in staff time derived from the automation of simple tasks and acceleration of processes with AI. In the United Kingdom, a review by the Government Digital Service estimated that AI could save civil servants two weeks a year (Government Digital Service, 2024[8]). A global survey of marketing and communications decision makers found 85% claim they are currently gaining back as much as five weeks a year thanks to AI (Canva, 2025[9]).
AI use to automate routine communication tasks, speed up time-intensive content creation and data analysis can free up staff to focus on higher-value activities. This is reflected in the prevalent use of AI to generate or edit text-based content among OECD Survey respondents (and their peers in industry (Canva, 2025[9])), along with some communication teams also using it for audiovisual media (Figure 3.2).
For which of the following tasks do communicators in your institution commonly use AI?
Note: Data shown above refers to 56 responses to the relevant question in the OECD Survey of Government Communicators. Multiple options possible. Full answer options: “Translating content into other languages and/or making it accessible to audiences with hearing or visual impairments.”, “Generating text-based content, such as press releases or web content.”, “Operating custom chatbots for the public (e.g. on government websites).”
Source: OECD 2025 Survey of Government Communicators.
Besides these routine tasks, AI is also proving to be a valuable tool to support reasoning, planning and decision making. Over half of government communicators surveyed are embracing AI for brainstorming, and one in five are using it to develop communication plans. Generative AI tools are especially useful to help teams quickly generate fresh ideas and come up with creative solutions to problems or more tailored messaging.
The area in which AI offers the greatest productivity potential is in the realm of creative outputs (Canva, 2025[9]). Generative AI is proving increasingly sophisticated in enhancing and tailoring messaging, support branding, and creating compelling content in different formats. As online content continues to shift away from text and towards images, audio and video (see Chapter 2), generative AI can help communicators keep up with the production of greater volumes of quality graphics and videos.
3.1.2. Artificial Intelligence as a lever for responsive, relatable and interactive communication
OECD analysis in the field of public communication has long emphasised the importance for public institutions to shift their communication efforts towards approaches that prioritise listening to citizens and giving them a say in decisions (OECD, 2021[7]; OECD, 2023[10]). Combined with clear communication of reforms, these elements have a positive association with trust in the national government (OECD, 2026[2]). Two-way communication that enables a feedback loop between policymakers and citizens has thus been heralded as a priority for reforming the function and its core purpose. The findings of the 2025 Survey of Government Communicators echo this, with about 2 in 5 respondents claiming that more citizen-centric communication is one of the top three priorities for reforming the function.
AI can help make communication more responsive, interactive and accessible. As seen in Chapter 2, communicators already focus on analysing public sentiment and segmenting audiences to target them with more relevant content. However, by incorporating ever-advancing AI into the process, these tasks can be performed with greater precision and fewer resources.
Firstly, research, analysis and insights about what people think and feel, how they talk about policy issues,2 are the foundation to any effort to communicate better with them. AI advances are making tasks of sentiment analysis more precise and granular than ever, for example by distinguishing between humour and sarcasm to determine the tone of a social media post or news articles.3 Analysing media and information channels is one of the more common applications of AI, with 36% of communicators surveyed by the OECD highlighting it as one of their practices in Figure 3.2, and 21% claim to use AI in data collection and analysis.
On the other hand, communicators increasingly claim to have too much data and struggle to know how to make sense of it, which AI can help do (OECD, 2025[11]). Some institutions are building customised tools to bring real-time, precise insights to decision makers in communication and beyond (Box 3.1). Proprietary and commercial tools are also being developed that allow customisable forecasting and early-warning, for instance in the event of crises.
Defining relevant audience segments based on complex profiles that blend values and beliefs with demographic characteristics is also increasingly effective thanks to AI. Machine learning has become a better judge of audiences’ profiles, reduces human bias and can achieve more granular segmentation (Youyou, Kosinski and Stillwell, 2015[12]).
All these insights from organisational listening, along with specific audience segments, allow communication teams to address the information and engagement needs of all citizens in a more targeted way. Content personalisation, for instance, is a promising application of AI used by 12% of communication teams. This entails enhanced capacity to micro-target audience segments with many variants of highly tailored content designed to resonate with them (Deloitte Digital, 2025[13]).
France’s AI data and analysis tool
The Government Information Service has applied AI in the development of a new custom tool to track and analyse different sources of data across information channels. On top of a range of Softwares as a Service (SaaS) already in use by teams and individuals, this tool scrapes sources including social media posts, press articles, and broadcast transcripts. The model is then trained to analyse this diverse data set, drawing connections between individual data points and identifying primary and secondary themes that emerge from public discourse across all sources. This AI-driven process helps communicators identify trends and rapidly generate reports, reducing efforts in tracking and reconciling multiple sources of monitoring and analytics. Beyond saving time, this tool has been contributing to more insightful decisions and effective communication.
DG Communication’s analytics dashboard and AI efforts at the European Commission
The European Commission’s Directorate-General for Communication (DG COMM) has established a specialist team that helps facilitate artificial intelligence to provide communicators with deeper and more actionable insights. This work begins with content analytics. Rather than manually tagging thousands of photos, videos, and web pages, AI now performs this task automatically. It also supports translation and consolidates performance data from multiple platforms into a single overview. As a result, communication teams can quickly understand what works and what does not, an analysis that previously required hours of manual effort. The team has also worked on internal chat bot that allows workers to quickly search for information and respond to questions and interactions of citizens online.
However, AI solutions are not the only focus in DG COMM. The team is strengthening its analytical foundations. A key element of this effort is the development of country dashboards. They provide two complementary types of reporting. The country intelligence reports, which support national-level communication planning, while regional reports allow strategies to be adapted to local contexts. The indicators used include trust drivers and economic risk factors, enabling communicators to base their decisions on evidence rather than intuition.
Together, these initiatives form a coherent system. AI accelerates and simplifies content processing, while the dashboards organise, contextualise, and visualise the broader analytical picture. The underlying principle is clear: structured data and robust analytics are essential prerequisites for advanced, predictive AI. By combining automation with strong data tools, DG COMM is making communication planning faster, more systematic, and firmly grounded in evidence.
The use of BIG KINDS (Korean Integrated News Database System) by the Korean government
The Korean government uses BIG KINDS is an AI-based news big data platform operated by the Korea Press Foundation to better align communication with public demand.
BIG KINDS collects and stores a vast archive of Korean news articles, more than 113 million items spanning decades, and provides tools to search, analyse, and visualise this content through a web-based interface. The platform draws from 104 different media outlets, including national, regional, economic, and broadcast sources.
Using advanced AI, BIG KINDS organises thousands of daily news articles into topics and themes. It groups related stories, identifies issues gaining attention, and provides AI-generated summaries that condense extensive coverage into clear, readable formats. Users can also compare how different media outlets report on the same issue, revealing differences in framing and editorial focus. Beyond topic-based analysis, the platform enables users to track trends over time, supporting longitudinal studies of media attention. Visualisation tools show which issues dominate coverage and how narratives evolve, offering a structured view of the news landscape.
BIG KINDS is free and open to anyone who registers online. Users can search full text, download analysis data, set up email alerts, and even interact with a natural-language AI chatbot for conversational queries. The platform also allows document uploads for custom analysis and highlights emerging topics through its “Today’s Issues” feature.
The platform helps public communicators quickly understand what the public encounters in the media and adjust messaging. Combined with Korea’s real-time public opinion analysis system (tracking mentions and sentiment across social media and online communities), BIGKINDS supports data-driven communication.
Sources: Written information provided by the French Government Information Service to the OECD Secretariat, March 2025; interview with DG COMM officials, October 2025; Written information provided by the Korean Public Communication Office within the Ministry of Culture, Sports and Tourism to the OECD Secretariat, November 2025; https://www.kinds.or.kr/.
It goes along with improvements in the accessibility of content, which 42% of teams are using AI for. This includes notably captioning, translation, and other ways to make communication suited to those with disabilities or linguistic barriers (Chun and Noveck, 2025[14]).Such improvements stand to dramatically alleviate the constrains governments face to ensure their communications and channels are accessible to all (Cazenave and Bellantoni, 2022[15]).
Translating listening and insights into action involves testing and refining messages, content, and channels to ensure the effectiveness of the communication strategy (Macnamara, 2022[16]; OECD, 2025[11]). On this dimension too, AI-powered solutions show potentially groundbreaking advances. The use of “synthetic audiences” or “silicon sampling” mean using AI-generated publics that simulate real people. These are created using existing datasets or large language models (LLMs) to mirror the characteristics, attitudes, and behaviours of actual humans. Some proprietary tools claim to produce results from synthetic audiences that match population surveys almost exactly, while methods that rely on common LLMs alone tend to underrepresent certain niche views or minority groups (Wihbey and D’Alonzo, 2025[17]).
Synthetic audiences are being used by some communication teams to analyse opinion and test communication strategies, messages, or content before launching them at scale. Such testing and iteration can make the difference for communication’s impact, yet it is likely to fall prey to the above-mentioned resource constraints. An OECD review of French government communication departments, for instance, found that this type of testing happens in only a fraction of campaigns (OECD, 2025[11]).
While the literature tends to acknowledge not to treat synthetic audiences as a substitute for listening to real people (Wihbey and D’Alonzo, 2025[17]), the low barrier to using some of these technologies means that communicators can obtain a measure of evidence even in those situations where no testing would otherwise be possible. At a minimum, use of synthetic audiences can improve survey design and refine the questions to ask humans to maximise the value of social research. In due course, technologies may also improve and come increasingly closer to reliable and representative mirrors of the public.
While these technologies can contribute to the goal of more responsive, relevant and relatable communication, it is important that they remain a supplement to greater efforts by institutions to connect with real citizens. The resources and time that AI-based insights can save then ought to be invested in opportunities for higher quality contact with communities.
The proliferation of virtual assistants to interact directly with citizens
The idea of enabling an ongoing, interactive dialogue that mirrors individual conversations gained some traction when social media first took off as a means for individuals and organisations to interact directly. However, generative AI has become the real gamechanger to get closer to this type of conversational communication. LLMs and small language models (SLMs) have ushered a new generation of custom chatbots that governments can deploy as a more effective interface for citizens to find information, access services, give input or feedback, and more.
Chatbots have become a common feature of public institutions websites for some time, notably increasing during the COVID-19 pandemic as a way to address people’s questions and navigate evolving guidance and health measures (Alfonsi et al., 2022[18]). Until recently, however, many chatbots have been simplistic and rules-based, unable to determine the direction of a conversation or mimic human speech (Syvänen and Valentini, 2020[19]). Today, they can provide more personalised experiences and cater to individual needs, even accessing live services data and users’ government accounts information (OECD, 2025[1]). By using retrieval augmented generation (RAG), which leverages a trusted database alongside LLMs, institutions can train the bots based on their own data and information, reducing hallucinations and misleading answers (Lewis, 2021[20]).
Moving beyond fixed bot scripts to deliver tailored information and recommendations (including options for plain or sign language) can dramatically increase people’s satisfaction with the information and day-to-day interaction with government. This in turn can be linked with the drivers of trust in institutions (OECD, 2026[2]). The OECD report on Governing with AI (2025[1]) includes numerous examples of how this technology is deployed in OECD countries (Box 3.2).
Chatbots are also important sources of data and insights of what citizens need and where they face issues. This includes questions that citizens may feel uncomfortable asking a human (Syvänen and Valentini, 2020[19]). For example, a virtual recruiter for the British Navy is asked questions that are not typically posed to human recruiters, such as “what is it like to be a Muslim in the Navy?” (Baugh, 2023[21]). Chatbots can also identify recurring issues as they develop, revealing trends that otherwise would have only come to light through the co-ordination of service agents over a longer period of time (Vassilakopoulou, 2023[22]; Larsen and Følstad, 2024[23]). This feedback loop can be particularly valuable during crisis communication situations, where chatbots can provide immediate responses to queries (Cheng, 2024[24]).
Notwithstanding their potential, it is important to note that AI chatbots do still require human oversight and maintenance. AI‑powered chatbots can be susceptible to manipulation, resulting in the reproduction of falsehoods, the creation of fabricated news‑like content, or the unintentional dissemination of incorrect information through hallucinations. These problems and limitations with chatbots can in turn cause citizens frustrations and harm their trust, for instance with incorrect or misleading answers (Larsen and Følstad, 2024[23]). In this respect, in France the government’s generative AI system Albert was deliberately deployed not for direct interaction with citizens but as an assistance tool for frontline public servants when responding to citizens’ requests. Public servants remain fully responsible for formulating and delivering responses building on the information generated by AI (Government of France, 2025[25]).
Portugal’s virtual assistant
In 2025, the Administrative Modernisation Agency (AMA) in Portugal introduced an expanded version of its existing chatbot into a comprehensive virtual assistant, integrated across over 5 000 pages of its government websites. Combining ChatGPT and a dedicated knowledge base, the assistant delivers information on more than 2 300 public services in 12 different languages, ensuring that language is no longer a barrier for its diverse population. The virtual assistant is slated to add new interactive support features, such as process tracking and appointment scheduling.
mAigov assistant in Greece
Greece’s mAigov is an AI-driven assistant featured on the gov.gr portal, designed to guide users through more than 1 300 digital services and over 3 200 administrative processes by responding to queries in natural Greek, whether typed or spoken. Integrating advanced reasoning abilities, the chatbot can preserve conversational context, ensuring an enhanced user experience. Supporting 25 languages, it serves a wide range of users, including Greeks living abroad. Planned enhancements for mAigov include issuing certificates and booking appointments for citizens, which will further simplify access to public services.
Bürokratt system of virtual assistants in Estonia
Bürokratt is Estonia’s government LLM-based digital assistant, which operates as a co-operative network of personalised AI agents that seamlessly interact, execute tasks, and deliver human-centered, secure, and intelligent services. It enables citizens to access information and services through natural language interactions. Bürokratt aims to make engaging with the state as simple as chatting with a friend.
A chatbot to navigate environmental data and research in France
SofIA is an example of a specialised chatbot designed by the French Ministry for Ecological Transition and powered by Mistral’s LLM to streamline access to the vast scientific and technical resources held by the Ministry. Thanks to a shared, specialised database, the chatbot allows users to rapidly find information. Environmental considerations have been central to SofIA’s development, with the platform displaying the carbon emissions generated by each search to users.
Note: Adapted from OECD (2025) Governing with AI.
3.1.3. Looking ahead: Strengthening generative engine optimisation in public communication
The emerging applications of AI discussed above speak to the enormous opportunities to address longstanding challenges to public communication. However, the use of AI in public communication is introducing some new challenges of its own. The popularity of generative AI tools is changing how people use the internet (Honan, 2025[30]). AI chatbots are rapidly beginning to replace search engines and web browsing with ready summaries and answers to user queries. Further, generative AI summaries have been incorporated into traditional search engines, meaning many users may not navigate past this information. Early research has shown that users offered an AI search summary click through to links far less than those who do not receive an AI summary (Chapekis and Lieb, 2025[31]).
This development poses a major challenge for third-party information sites, including news providers and government websites, which must adapt to the evolving digital landscape to remain effective and to reach citizens with trustworthy information. The effects on how people consume news are already being felt by media organisations that are encountering a decline in site visits (Simonetti and Blunt, 2025[32]; Savage, 2025[33]). The drop in visits from search results for British tabloid The Mail Online, for example, was reported to be as much as 56% and 48% for the desktop and mobile sites respectively (Tolbitt, 2025[34]).
Public communicators must thus grapple with information about their institutions, policies and programmes being relayed in answers crafted by popular LLMs. This carries risks of policy and government information being distorted, blended with less reliable sources or simply not recalled by AI overviews.
To overcome these challenges, public communication teams could explore approaches to shape how generative AI tools access and present government information. Like Search Engine Optimisation (SEO) became a regular feature of digital communication two decades ago, Generative Engine Optimisation (GEO) is becoming established as an answer to the ongoing transformation in people’s digital and information-seeking habits. This field is seeing considerable experimentation, but early research suggests that GEO can boost visibility for an organisation’s information by up to 40% in AI chatbot and search responses (Aggarwal et al., 2024[35]).
The digital marketing field is providing rapidly-evolving guidance and practices for GEO. For example, generative engines appear to be more reactive to content that includes citations, quotations from relevant sources, and statistics. Building mentions of the entity or brand across the web privileging authoritative sources likewise helps LLMs select it in their answers. Stylistic changes such as structuring content in formats appropriate for natural language queries (such as writing frequently asked questions or FAQs) also appeared to boost the performance of sources in generative engines. Specific modifications to webpage codes can similarly improve the page’s visibility in relevant generative AI answers (Kumar and Lakkaraju, 2024[36]).
As AI continues to transform how people get their news and find information, it is paramount that government communication departments can rapidly adapt and also exploit possible opportunities to improve how they reach citizens with relevant and trustworthy information in a new information ecosystem.
Building on the emerging applications of AI in government, the near-term priority for institutions is to put in place programmes to scale AI adoption systematically across teams, ensuring that its benefits are realised organisation-wide while introducing the necessary guardrails that can ensure AI in government is reliable and trustworthy. Almost all OECD countries are now working with strategies and agendas that establish high-level visions for the use of AI in government (OECD, 2025[1]). The United Kingdom’s Government Communications has developed a strategic approach to responsible AI adoption, presented in Annex A. Adopting AI was the second most-selected reform priority for communication by government communicators in the OECD survey, with 39% of respondents listing it among their top three.
Despite the clear imperative for scaling trustworthy AI in public communication, the reality on the ground reveals that adoption across governments remains variable and, in many cases, exploratory (OECD, 2025[1]). In this respect, public sector communicators tend to lag behind their private sector peers, who often enjoy more enabling environments and pressure to adapt rapidly to industry innovation (Macnamara, 2025[37]; Zerfass, Hagelstein and Tench, 2020[38]). Shifting towards systematic use, with full integration of AI in relevant processes will require considerable investment and upskilling, as discussed further in the next chapter.
The 2025 OECD Survey of Government Communicators finds that when it comes to the level of adoption of AI by communication teams, only a handful of respondents claim that “AI tools are embedded in the teams’ working methods […] and are routinely used” or that they are being increasingly integrated (Figure 3.3). Conversely, 46% of respondents indicate that AI use remains exploratory in their teams, meaning their teams are “testing some AI tools to decide which ones to integrate more regularly in their work and for which purposes.”
However, as many as 35% of the teams in the OECD survey seem to rely mainly on individuals’ own initiative and experimentation. In many workplaces, experimentation with AI tools is being driven from the bottom up before being integrated into corporate training and guidance (OECD, 2025[1]). This informal experimentation is providing valuable lessons but can often occur outside of formal approval processes or in the absence of guidelines that mitigate risks of AI use. These findings highlight that for over one third of institutions, there is either a gap in, or low adherence to, organisation-wide initiatives and efforts to promote AI adoption in relevant processes. The same finding was echoed in interviews conducted by the OECD Secretariat with members of the Public Communication Network.
To what extent has your institution integrated AI in its communications?
Note: Data shown above refers to 57 responses to the relevant question in the OECD Survey of Government Communicators.
Source: OECD 2025 Survey of Government Communicators.
Fostering the right enabling environment for AI testing and scaling, including dedicating adequate resources to this priority objective, is essential to reap the benefits while mitigating risks. Some institutions, like the European Commission’s Directorate-General for Communication (DG COMM) have tasked specialists with spearheading initiatives to support wider AI adoption, including by developing bespoke and proprietary tools suited to their organisations (Annex A). Oversight and safety considerations, including the protection of data, tend to play a large role in institutions’ choices for developing their own tools, in addition to meeting specific needs and uses.
3.2.1. Building strategies and resources to support effective and responsible uptake of AI in public communication
For most communication teams the development of tools to embed AI use in public communication are still lagging. Figure 3.4 shows which types of guidance, programmes, and resources have been introduced in respondents’ teams. The data reveals that a small minority of teams (14%) have most of these elements and resources in place (at least 5 of the listed options). This stands in contrast to a sizeable group (30%) who have only one of these elements in place or those who have no resources at all (16%), highlighting a large gap between frontrunners and laggards.
Whereas few teams benefit from full-fledged strategies for AI in communication, two in five now have access to a range of toolkits, manuals and guidelines on using AI, and more than half can benefit from dedicated training and learning opportunities. Consistently with the findings above, pilot programmes are common to close to a third of teams. Several teams can also rely on cross-government resources such as innovation labs and on strategies for AI in the public sector.
What kinds of resources are available to your team to support effective applications of AI to their work?
Note: Data shown above refers to 57 responses to the relevant question in the OECD Survey of Government Communicators. Multiple options possible. The question required that unless the answer specifies “in the public sector”, answers are intended to be specific to the communication function.
Source: OECD 2025 Survey of Government Communicators.
Of all the resources needed to enable trustworthy AI in government communication, clear guidance on ethics and risk mitigation are viewed as the most pressing priority and are considered essential for successfully scaling new approaches. Most respondents (93%) would like to see new or updated ethics and propriety guidance in relation to AI use, AI-generated content and AI-powered tools.
The OECD report on Governing with AI (2025[1]) notes that the technology has been found to carry a wide range of risks affecting “trust, fairness, privacy, safety and accountability, among others”. For public communication, there are particular concerns around the accuracy and trustworthiness of content when AI is involved in its generation. While LLMs can create text that is understandable and compelling, at present, their outputs do not consistently meet standards of factual accuracy or reliability without human verification. Similar limitations apply to audio-visual content formats.
Human oversight is thus widely emphasised as the necessary condition for using AI to produce content at the time of writing. In the field of journalism, for example, research shows that audiences are more likely to trust news content where AI is used in “back-end” functions, such as editing and translation, rather than in generating whole articles or presenting audio/video broadcasts (Arguedas, 2024[39]). Similarly, people tend to be more comfortable with AI-driven content personalisation that can help improve relevance and comprehension, provided editorial control and transparency rest with humans.
To this end, the Reuters Institute for the Study of Journalism recommends the use of clear labels that specify both where and how AI was used, as well as indicating with what oversight (Arguedas, 2024[39]; Newman et al., 2025[4]). It is possible to envision a future where a majority of content is generated in part, if not entirely, by AI and where labelling or verification of human oversight (such as the equivalent of a CAPTCHA system for content generation) could become necessary.
The prospect of greater personalisation of communications mentioned above is one of the areas that is also most prone to risks. Without transparency and oversight, this can contribute to information silos, where people receive only the messages relevant to them. There is the possibility that increased personalisation can, thus, lead to increased fragmentation, aggravating existing trends seen on social media platforms. Responsible use of personalisation, instead, likely involves conscious efforts to resonate with different audience segments while ultimately conveying the same underlying messages and information.
AI-based personalisation is also closely linked to persuasion and, potentially, manipulation. These risks are not limited only to malign actors: institutions have long been dealing with ethical questions on the grey areas and boundaries of persuasion, reputation management, and “spin” (OECD, 2021[7]; OECD, 2023[10]). Researchers have predicted the growing effectiveness of generative AI at personalised persuasion due to AI’s ability to analyse data, deliver more relevant messages and avoid personal biases (Matz et al., 2024[40]). Across four studies, personalised messages crafted by ChatGPT exhibited more influence than non-personalised messages. While the dangers posed by the misuse of AI for persuasion are significant, it is important to define the boundaries for responsible use of these tools to support desirable, public-interest goals.
Finally, when it comes to mitigating the risk of bias in AI-generated content, whether arising from the ways in which algorithms are trained or from human interpretation of AI outputs, the picture is less straightforward. In this respect, human verification and oversight could serve to ensure AI outputs are not inadvertently discriminating against, misrepresenting, or excluding some groups that are underrepresented in their training data.
While AI continues to develop, public communicators should ensure they do not perceive AI as an errorless tool capable of solving every problem (Narayanan and Kapoor, 2024[41]). As highlighted in the OECD report on Governing with AI (OECD, 2025[1]), AI should always be evaluated against other alternative solutions. If progressed, it should then be tested and implemented on a case-by-case basis, weighing up both the risks and benefits.
The emphasis on risks can lead to over-cautious responses in public sector institutions, which already tend to have a risk-averse culture. In this sense, the risk of inaction highlighted by the OECD Governing with AI report should not be discounted. The advantages of measured adoption now far outweigh those of hesitation. As such, countries should progress work rolling out high-return, low-risk AI across their institutions (OECD, 2025[1]). The OECD Framework for Trustworthy AI in Government can support governments in navigating the challenges and opportunities to harness AI (Box 3.3).
The OECD Framework for Trustworthy AI in Government is structured around three core pillars: enablers, guardrails, and engagement. This framework is designed to help governments make the most of AI’s opportunities while proactively addressing its risks and the challenges of implementation.
Firstly, enablers involve putting in place essential governance structures and processes, recognising data as the cornerstone of AI, developing robust digital infrastructure, nurturing digital skills and talent, making targeted investments, leveraging public procurement effectively, and broadening AI’s benefits through strategic partnerships. Secondly, guardrails refer to both binding and non-binding policy instruments, procedures that promote transparency, and mechanisms to ensure accountability. Finally, stakeholder engagement encompasses a range of activities, such as convening citizen assemblies, involving civil servants, engaging end users in the development of AI systems, and fostering cross-border collaboration.
Source: OECD (2025[1]), Governing with Artificial Intelligence: The State of Play and Way Forward in Core Government Functions, https://doi.org/10.1787/795de142-en.
References
[35] Aggarwal et al. (2024), “GEO: Generative Engine Optimization”, arXiv, https://arxiv.org/pdf/2311.09735.
[27] AI empowering Estonia (n.d.), Virtual assistant Bürokratt, Ministry of Justice and Digital Affairs, Estonia, https://www.kratid.ee/en/burokratt (accessed on 15 July 2026).
[18] Alfonsi, C. et al. (2022), “Public communication trends after COVID-19: Innovative practices across the OECD and in four Southeast Asian countries”, OECD Working Papers on Public Governance, No. 55, OECD Publishing, Paris, https://doi.org/10.1787/cb4de393-en.
[39] Arguedas, D. (2024), Public attitudes towards the use of AI in journalism, Reuters Institute for the Study of Journalism, University of Oxford, https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2024/public-attitudes-towards-use-ai-and-journalism.
[21] Baugh, S. (2023), “AI and innovation in communications [Speech]”, LinkedIn, https://www.linkedin.com/posts/simon-baugh-8108b214_ai-and-innovation-in-communications-speech-activity-7111266717653159936-4JA7/ (accessed on 1 December 2025).
[9] Canva (2025), The State of Marketing and AI, https://www.canva.com/marketing-ai-report/.
[15] Cazenave, E. and A. Bellantoni (2022), “Accessible and inclusive public communication: Panorama of practices from OECD countries”, OECD Working Papers on Public Governance, No. 54, OECD Publishing, Paris, https://doi.org/10.1787/222b62d9-en.
[31] Chapekis, A. and A. Lieb (2025), “Google users are less likely to click on links when an AI summary appears in the results”, Pew Research Center, https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/.
[24] Cheng, Y. (2024), Crisis Communication in the Age of AI: Navigating Opportunities, Challenges, and Future Horizons, Routledge, https://doi.org/10.4324/9781032619187-9.
[14] Chun, S. and B. Noveck (2025), “Introduction to the special issue on ChatGPT and other generative AI commentaries: AI augmentation in government”, DGOV: Digital Government: Research and Practice, Vol. 6/2, pp. 1-8, https://doi.org/10.1145/3735513.
[13] Deloitte Digital (2025), Artificial Intelligence and the Future of Public Engagement, https://www.deloitte.com/content/dam/assets-zone3/us/en/docs/industries/life-sciences-health-care/2024/amc-ai-the-future-of-public-engagement.pdf.
[26] European Union (n.d.), Digital Skills and Jobs Platform, https://digital-skills-jobs.europa.eu/en/search.
[8] Government Digital Service (2024), M365 Copilot experiment: Findings report, https://assets.publishing.service.gov.uk/media/683db42bd23a62e5d32680d0/M365_Copilot_Experiment_Findings_Report.pdf.
[25] Government of France (2025), IA : connaissez-vous Albert ?, https://www.info.gouv.fr/actualite/ia-connaissez-vous-albert.
[29] Government of Portugal (2025), Assistente virtual do gov.pt facilita a interação em 12 idiomas, Digital.gov.pt, https://digital.gov.pt/noticias/assistente-virtual-do-gov-pt-facilita-a-interacao-em-12-idiomas.
[30] Honan, M. (2025), “AI means the end of internet search as we’ve known it”, MIT Technology Review, https://www.technologyreview.com/2025/01/06/1108679/ai-generative-search-internet-breakthroughs/.
[36] Kumar, A. and H. Lakkaraju (2024), “Manipulating Large Language Models to Increase Product Visibility”, arXiv, https://arxiv.org/pdf/2404.07981.
[23] Larsen, A. and A. Følstad (2024), “The impact of chatbots on public service provision: A qualitative interview study with citizens and public service providers”, Government Information Quarterly, Vol. 41/2, https://www.sciencedirect.com/science/article/pii/S0740624X24000194?ref=pdf_download&fr=RR-2&rr=9631fcc65e5f9e21.
[20] Lewis, P. (2021), “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks”, arXiv, https://arxiv.org/abs/2005.11401.
[5] Lovari, A. and F. De Rosa (2025), “Exploring the Challenges of Generative AI on Public Sector Communication in Europe”, Media and Communication, Vol. 13/1, pp. 1-15, https://www.cogitatiopress.com/mediaandcommunication/article/view/9644.
[37] Macnamara, J. (2025), Public Communication and Public Policy: Reviving Democracy by Recalibrating Public Communication, Wiley Blackwell.
[16] Macnamara, J. (2022), “Measurement, evaluation + learning (MEL): New approaches for insights, outcomes, impact”, in The Routledge Companion to Public Relations, Routledge, London, https://doi.org/10.4324/9781003131700-21.
[40] Matz, S. et al. (2024), “The potential of generative AI for personalized persuasion at scale”, Scientific Reports, Vol. 14/Article 4692, https://www.nature.com/articles/s41598-024-53755-0.
[28] Ministère de la Transition écologique, France (2025), Quand l’intelligence collective rencontre l’intelligence artificielle : l’exemple de la refonte du système d’information documentaire, Innovons pour les transformations, https://www.innovation-transformations.ecologie.gouv.fr/projets/quand-lintelligence-collective-rencontre-lintelligence-artificielle-lexemple-de-la-refonte-du-systeme-dinformation-documentaire.
[41] Narayanan, A. and S. Kapoor (2024), AI Snake Oil: What Artificial Intelligence Can Do, What It Can’t, and How to Tell the Difference, Princeton University Press, https://press.princeton.edu/books/hardcover/9780691249131/ai-snake-oil.
[4] Newman, N. et al. (2025), Digital News Report 2025, Reuters Institute for the Study of Journalism, University of Oxford, https://doi.org/10.60625/risj-8qqf-jt36.
[2] OECD (2026), OECD Survey on Drivers of Trust in Public Institutions 2026 Results: Navigating Rising Expectations and New Horizons, OECD Publishing, Paris, https://doi.org/10.1787/9eb63fec-en.
[1] OECD (2025), Governing with Artificial Intelligence: The State of Play and Way Forward in Core Government Functions, OECD Publishing, Paris, https://doi.org/10.1787/795de142-en.
[11] OECD (2025), Public Communication Scan of France: Evaluating public communication performance to strengthen public trust, OECD Public Governance Reviews, OECD Publishing, Paris, https://doi.org/10.1787/7450361a-en.
[43] OECD (2024), “Explanatory memorandum on the updated OECD definition of an AI system”, OECD Artificial Intelligence Papers, No. 8, OECD Publishing, Paris, https://doi.org/10.1787/623da898-en.
[3] OECD (2024), “Governing with Artificial Intelligence: Are governments ready?”, OECD Artificial Intelligence Papers, No. 20, OECD Publishing, Paris, https://doi.org/10.1787/26324bc2-en.
[10] OECD (2023), Public Communication Scan of the United Kingdom: Using Public Communication to Strengthen Democracy and Public Trust, OECD Public Governance Reviews, OECD Publishing, Paris, https://doi.org/10.1787/bc4a57b3-en.
[7] OECD (2021), OECD Report on Public Communication: The Global Context and the Way Forward, OECD Publishing, Paris, https://doi.org/10.1787/22f8031c-en.
[42] OECD.AI (2024), OECD AI Policy Observatory, https://oecd.ai (accessed on 1 June 2026).
[33] Savage, M. (2025), AI summaries cause “devastating” drop in audiences, online news media told, https://www.theguardian.com/technology/2025/jul/24/ai-summaries-causing-devastating-drop-in-online-news-audiences.
[32] Simonetti, I. and K. Blunt (2025), News Sites Are Getting Crushed by Google’s New AI Tools, https://www.wsj.com/tech/ai/google-ai-news-publishers-7e687141.
[19] Syvänen, S. and C. Valentini (2020), “Conversational agents in online organization–stakeholder interactions: A state-of-the-art analysis and implications for further research”, Journal of Communication Management, Vol. 24/2, pp. 339–362, https://doi.org/10.1108/jcom-11-2019-0145.
[34] Tolbitt, C. (2025), “Google AI Overviews leads to dramatic reduction in clickthroughs for Mail Online”, PressGazette, https://pressgazette.co.uk/publishers/digital-journalism/google-ai-overviews-leads-to-dramatic-reduction-in-click-throughs-fo.
[22] Vassilakopoulou, P. (2023), “Developing human/AI interactions for chat‑based customer services: lessons learned from the Norwegian government”, European Journal of Information Systems, Vol. 32/1, pp. 10-22, https://doi.org/10.1080/0960085X.2022.2096490.
[17] Wihbey, J. and S. D’Alonzo (2025), AI Simulations of Audience Attitudes and Policy Preferences: “Silicon Sampling” Guidance for Communications Practitioners, Elsevier BV, https://doi.org/10.2139/ssrn.5533958.
[12] Youyou, W., M. Kosinski and D. Stillwell (2015), “Computer-based personality judgments are more accurate than those made by humans”, PNAS, Vol. 112/4, https://doi.org/10.1073/pnas.1418680112.
[6] Yun, L., S. Yun and H. Xue (2024), “Improving citizen‑government interactions with generative artificial intelligence: Novel human‑computer interaction strategies for policy understanding through large language models”, PLOS ONE, Vol. 19/12, https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0311410.
[38] Zerfass, A., J. Hagelstein and R. Tench (2020), “Artificial intelligence in communication management: a cross-national study on adoption and knowledge, impact, challenges and risks”, Journal of Communication Management, Vol. 24/4, https://doi.org/10.1108/JCOM-10-2019-0137.
← 1. The OECD defines AI as a “machine-based system that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments.” AI systems vary widely in complexity and purpose. See: (OECD.AI, 2024[42]; OECD, 2024[43]).
← 2. Listening by organisations is intended as the practice of gathering and analysing structured and unstructured feedback from the public via a range of open communication channels, call centres, complaints, social research or participation methods, and interactions with services and frontline staff, compatibly with data protection and privacy regulations. It does not refer to snooping or tracking individuals and their communications.
← 3. Demonstration of commercial AI tools during the workshop “Echoes of EuroPCom” organised by the European Committee of the Regions, 20 November 2025.