Much of the current discussion has focused on the obvious applications. Campaigns are experimenting with AI-generated speeches, fundraising emails, advertising copy, policy summaries and volunteer communications. Vendors promise faster content production, lower costs and greater efficiency. Those developments matter, but they are ultimately extensions of work that campaigns were already doing.



The more significant transformation is taking place elsewhere.

Large language models (LLMs) are beginning to change how political information reaches voters. Increasingly, citizens are not searching for campaign websites, comparing manifestos or reading several news articles before forming an opinion. They are asking ChatGPT, Gemini, Claude or Perplexity to explain candidates, summarise policies or compare parties on their behalf. AI is gradually becoming another intermediary between campaigns and the electorate.



That shift raises a question which very few campaigns are asking: who inside the campaign is responsible for making sure AI understands us correctly?

I suspect that, over the coming years, answering that question will become as important as hiring a digital director was twenty years ago. The person responsible may have a different title, but for now, I would describe the role as the Campaign AI Engineer.

The title is deliberately provocative because the responsibility barely exists today. Yet if we look carefully at how campaigns are already adapting to LLMs, the market appears to be assembling precisely this function, long before it has settled on a name.



Political campaigns are entering the age of machine interpretation



For most of the internet era, digital strategy revolved around one objective: making information easier to find.

Campaigns invested heavily in search engine optimisation, social media distribution and digital advertising because success depended on visibility. The strategic challenge was straightforward. If voters searching for a candidate or policy encountered your information before your opponent’s, you gained an advantage. Digital campaigning was largely a competition for attention.



LLMs fundamentally change that logic.

When someone asks Google about a candidate, they receive a list of websites and decide which sources to trust. When they ask ChatGPT or Gemini the same question, they increasingly receive a single, synthesised explanation. Rather than directing users towards information, these systems attempt to interpret it on their behalf. The voter no longer evaluates ten sources individually. They evaluate AI’s reconstruction of those sources.



That may appear to be a subtle distinction, but it changes the optimisation problem facing campaigns. Search engines rewarded visibility. LLMs reward coherence.

An AI system does not simply retrieve a campaign website and repeat what it says. It attempts to reconstruct an understanding of the campaign by drawing together information from every publicly available source it considers relevant: manifestos, interviews, parliamentary speeches, podcasts, local newspaper articles, Wikipedia pages, government databases, campaign websites and countless other fragments scattered across the internet.

Every public appearance becomes another piece of evidence. Every policy document becomes another signal. Every interview contributes to the model’s understanding of who the candidate is, what they stand for and how confidently that conclusion can be presented to a voter.

This means that campaigns are increasingly competing on something they have rarely considered before. They are no longer competing simply to be found but to be understood.

The distinction matters because understanding is cumulative. A campaign with a coherent, well-structured and consistent public information ecosystem is far easier for AI to interpret than one whose policies are scattered across PDFs, interviews, social media posts and outdated webpages. As more voters begin relying on AI to navigate political information, that difference becomes strategically significant.



Campaigns don’t have an AI problem. They have an information architecture problem.



This, in my view, is where much of the current debate has gone wrong.

When ChatGPT misrepresents a candidate or omits an important policy position, campaigns often conclude that AI itself is unreliable. We all know LLMs still hallucinate and make factual mistakes. But many of the problems campaigns attribute to AI originate long before a voter ever opens a chatbot.



Political campaigns have never really treated information as infrastructure.

Policy positions appear in speeches, interviews, press releases, downloadable manifestos and social media posts, often using different language in each. Candidate biographies vary slightly from platform to platform. Important media appearances remain available only as videos with no searchable transcript. Local reporting contains valuable information that never appears on official campaign websites. Flagship policies may be announced repeatedly, yet never collected in one authoritative location.

Human beings are remarkably good at connecting those pieces together. We instinctively understand that two slightly different descriptions of the same policy probably refer to the same idea. We can infer priorities from context, recognise contradictions and judge which source is most authoritative.



LLMs cannot make those assumptions with the same confidence.

They reconstruct campaigns from the information available to them. If that information is fragmented, inconsistent or poorly organised, the reconstruction is likely to be fragmented, inconsistent or incomplete as well.

The problem isn’t AI but information architecture. The term has traditionally belonged to web development, where it refers to the organisation and structure of information. Political campaigns should begin thinking about it much more broadly. 

Information architecture encompasses every piece of publicly available knowledge that contributes to how a campaign is interpreted. It asks whether policies are presented consistently across platforms, whether biographies reinforce one another rather than diverge, whether speeches and interviews are searchable, whether authoritative sources are clearly identifiable and whether a language model can reconstruct the campaign’s central priorities without having to infer them from dozens of disconnected fragments.



This is not simply a technical exercise. It is rapidly becoming a communications discipline in its own right.

Campaigns have spent years investing in better messaging. Increasingly, they will also need to invest in better knowledge systems. Those are related, but they are not the same thing. A compelling message is of limited value if the systems increasingly responsible for explaining politics cannot accurately interpret it.



The market is already building the Campaign AI Engineer



If the idea of a Campaign AI Engineer sounds speculative, it is worth looking at where campaigns are already spending money. The strongest evidence that this role is emerging comes from the growing ecosystem of products and services designed to solve exactly this problem.

Political technology companies largely focused on helping campaigns communicate with voters more efficiently. They built fundraising platforms, voter databases, canvassing software, volunteer management systems and digital advertising tools. A growing number of companies are now tackling a different challenge altogether: helping campaigns understand what AI understands about them.

Perhaps the clearest example is  CampSight , a platform launched by  Run for Something Action Fund . CampSight does not help campaigns generate AI content, nor does it promise better speeches or more persuasive fundraising emails. Instead, it continuously queries ChatGPT, Google AI and other language models in much the same way a voter would. It shows campaigns how they are described, which sources language models rely upon, where factual gaps exist and which parts of a candidate’s platform are missing altogether. Campaigns can then improve the underlying information available online and measure whether those interventions change how AI systems subsequently describe them.

The significance of CampSight lies less in the technology than in the assumption on which it is built. It treats AI-generated answers as something campaigns can monitor, audit and improve in exactly the same way they monitor polling, fundraising or media coverage. In other words, it assumes that a campaign’s representation inside LLMs has become another strategic asset that can be actively managed rather than passively accepted.

CampSight is far from an isolated example.  Caucus AI  has begun tracking what it calls a campaign’s presence across LLMS, measuring how candidates are represented when voters ask AI about the issues that matter to them. Rather than monitoring television coverage or social media engagement, these tools monitor something entirely new: what AI systems actually tell voters, which campaign messages consistently survive AI summarisation and how those answers evolve over time. If media monitoring helped campaigns understand how journalists portrayed them, AI monitoring is beginning to help campaigns understand how language models portray them.

The same shift is becoming visible beyond electoral politics. Public affairs firms started advising organisations to prepare their digital infrastructure not only for traditional search engines but also for AI crawlers. Their clients are increasingly asking a different set of questions than they were only a few years ago. It is no longer enough that Google can find a position paper or policy announcement. Organisations also want to know whether ChatGPT or Gemini can accurately interpret it, whether those systems cite authoritative sources and whether key positions survive AI summarisation without distortion.

The concern is not merely theoretical. In a February 2026 report prepared for PSG Consulting and Innovating for the Public Good, Dewey Square described what it calls an “inverted funnel” in AI training data. The researchers found that many high-factuality news organisations significantly restrict AI crawlers, while lower-factuality and more partisan websites remain far more accessible. 

The consequence is not simply that AI has access to less information; it is that AI may reconstruct political actors from a very different information ecosystem than the one human readers encounter. For campaigns, this introduces an entirely new strategic question. It is no longer sufficient to ask whether authoritative information exists online. Campaigns must also ask whether the systems increasingly interpreting politics can actually access, retrieve and rely upon that information in the first place.

Campaigns themselves are beginning to respond in much the same way. Earlier this year, The New York Times reported on Dustin Lloyd, a Democratic candidate for the Missouri House of Representatives, who discovered that ChatGPT barely mentioned one of the defining themes of his campaign: support for small businesses. Rather than dismissing this as another AI hallucination, Lloyd’s team expanded policy pages, clarified biographical information and published additional explanatory material that made the campaign easier for language models to interpret. When journalists repeated the same questions several weeks later, ChatGPT’s responses had improved accordingly.

What makes these examples so interesting is that none of them is really about AI in the way most campaign professionals think about it. None focuses on generating content, writing speeches or creating advertisements. Instead, they all address the same underlying problem from different directions: how campaigns are represented inside AI systems, how that representation can be measured and, crucially, how it can be improved.

Viewed individually, these developments appear relatively modest. Viewed together, however, they reveal something much more significant. The market has already begun assembling the responsibilities of a new campaign function.

Nobody is formally hiring a Campaign AI Engineer—at least not yet. But campaigns are already paying for someone to monitor AI outputs, analyse citations, identify weaknesses in machine understanding, improve the structure of campaign information and measure whether those interventions actually change what AI tells voters. The function already exists. The only thing still missing is a name.



What does a Campaign AI Engineer actually do?



One of the reasons this role remains difficult to recognise is that its responsibilities cut across departments that campaigns have traditionally kept separate.

Part of the job resembles communications. Part resembles digital strategy. Part overlaps with web development, policy research and data management. Yet none of those teams currently owns the broader question of how AI reconstructs the campaign.



A Campaign AI Engineer would.

Their responsibility would not be generating campaign content with ChatGPT. Ironically, that may become one of the least interesting uses of AI inside a campaign.



Instead, they would continuously evaluate how major language models describe the candidate, monitor which sources AI relies upon, identify factual inconsistencies and work with communications, policy and digital teams to improve the campaign’s overall information architecture. They would decide whether a flagship policy deserves its own dedicated landing page rather than remaining buried inside a 70-page manifesto. They would ensure that major speeches are transcribed, that interviews reinforce rather than contradict existing messaging and that authoritative information consistently outranks outdated reporting. They would test how different AI systems explain the campaign, compare those answers against campaign priorities and identify where misunderstandings originate.



In other words, their task would not be producing messages. It would be engineering understanding. That distinction is likely to become increasingly important as AI systems evolve from experimental tools into routine gateways through which citizens access political information.



Why this matters for European campaigns



While many of the earliest examples come from the United States, there are good reasons to believe European campaigns may ultimately benefit even more from this way of thinking.



European political environments are inherently more complex. Coalition politics, multilingual electorates, fragmented party systems and multiple layers of government make it significantly harder for both voters and AI systems to interpret political information accurately. Explaining the differences between two parties in a two-party presidential race is one challenge. Explaining the ideological distinctions between six coalition partners, regional parties and European parliamentary groups is another entirely.



The same complexity applies to public affairs. Increasingly, organisations are asking AI to explain legislative proposals, compare political stakeholders or summarise regulatory debates. If those systems become a routine starting point for understanding public policy, then the quality of the underlying information ecosystem becomes strategically important for advocacy organisations, businesses and governments as much as for political campaigns.



That makes information architecture more than a campaign concern. It is gradually becoming part of democratic infrastructure.



Campaign organisation, not content generation, will define the next phase of AI



Much of the discussion surrounding AI in politics has centred on productivity. How much faster can campaigns produce content? How many hours can they save? Which tasks can they automate?



Those are worthwhile questions, but they are unlikely to define the long-term impact of LLMs.

The more profound organisational change is that campaigns are beginning to communicate with a new audience altogether: AI itself. As AI increasingly mediates the relationship between campaigns and voters, ensuring that those systems understand a candidate accurately becomes a strategic capability rather than a technical afterthought.



Whether the title ultimately becomes Campaign AI Engineer, AI Information Architect or something else entirely is, in many respects, secondary. Every major technological shift has eventually produced specialists who did not previously exist. There is little reason to believe AI will be any different.



The campaigns that gain an advantage over the next decade are unlikely to be those that simply generate more content with AI than their opponents. They are more likely to be those that recognise a more fundamental shift is taking place: campaigns are no longer competing only to persuade voters directly. They are increasingly competing to ensure that the systems interpreting politics on behalf of those voters understand them better than they understand anyone else.