After social media, generative artificial intelligence (AI) is becoming an important source of political information. Increasing numbers of voters consult chatbots and large language models — such as ChatGPT or Gemini — for guidance on candidates and programmes. Caucus AI, cited by the New York Times, estimates that at least 16 million voters worldwide already obtain such information directly from language models or integrated AI search results.
What is answer-engine optimization (AEO) and how does it work?
Shifts in user behaviour have led many people to replace keyword search with conversational AI interactions. Modern AI assistants increasingly perform live web searches and incorporate recent online content into their responses. That development has given rise to answer-engine optimization (AEO): a set of practices and an emerging industry focused on deliberately shaping the web sources that AI systems rely on.
AEO builds on the idea that the more extensive and well-structured an actor’s digital presence is — official websites, Wikipedia entries, news sites, forums such as Reddit — the higher the chance those sources will be used by AI. However, which sources a particular chatbot chooses is not fully transparent and can vary between models.
An example of system vulnerability
Caucus AI’s experiment highlighted the dynamism and vulnerability of these systems: it took only 12 minutes for a change on Wikipedia to appear in the answers provided by chatbots. This illustrates how quickly source-level edits can affect AI-delivered information.
Cybersecurity, ethical and democratic risks
Legal and ethical framing of AEO can start from the principles developed for search engine optimization (SEO): it is legitimate for politicians or organisations to publish accurate, transparent and well-structured material so AI systems interpret their positions correctly. The problem arises when actors use “black hat” or deceptive techniques — creating fake or seemingly independent sources, coordinating edits of public content, or manufacturing artificial consensus. Such methods may not always be illegal, but they can violate service providers’ terms and become unlawful if they amount to deception or breaches of electoral rules.
The central risk is not only that manipulated information enters AI source pools, but that AI systems present it in a neutral, confident and coherent manner. Because AI responses often appear objective and polished, users may be more inclined to accept them even when they are based on incomplete or biased sources. In addition, language models can produce so-called hallucinations — false or inaccurate statements — which further threaten informed voter decision-making.
Responsibility and regulation
Determining legal responsibility for false chatbot assertions is a complex issue. Algorithms cannot be treated as independent legal actors; responsibility must be allocated among the human and organisational actors who built, operate, supplied data to, or deployed the system. If a chatbot’s error stems from a deceptive source, the publisher of that source may bear responsibility. If a service provider knows its system makes serious political errors but fails to implement adequate controls, the provider’s liability becomes relevant.
Regulatory frameworks currently differ significantly across regions (Europe, the USA, China) and are still evolving, while technology is developing faster than legislation can keep up. In future, liability assessments will likely focus on who created the faulty information, who had actual control over the system, and who could reasonably foresee the potential harm.
Competition, fairness and political effects
If AEO becomes an expensive, resource-intensive industry, better-funded political actors could gain an advantage in visibility within AI responses. Nevertheless, as in SEO, smaller or newer candidates are not necessarily excluded: by concentrating on niche topics, local issues, or credible subject-matter presence, they can still be effective.
Financial disparities will matter, but they will not be the sole determinant of favourable AI representation. It remains unclear to what extent credibility can offset larger campaign budgets.
Conclusion
Answer-engine optimization poses new challenges for electoral communication and democratic processes. Because AI assistants react quickly and are sensitive to source changes, digital presence, source management and transparency have become central issues. Clear regulation and rules on responsibility are necessary to reduce the risk that manipulated or inaccurate AI-provided information sways voters’ decisions.
