What sources does an AI model use when answering questions about an election? Does the election authority appear as a primary reference? Which questions are answered using media outlets, blogs or secondary sources? Where are the gaps in official information?
These questions led to the creation of ANIMA, a methodology developed by Smartmatic to audit the presence of election authorities’ official information in generative AI models. Its eight phases take election authorities from understanding what citizens ask to strengthening official content and monitoring progress over time.
ANIMA stands for AI Narrative & Institutional Monitoring Audit.
Its purpose is not only to determine whether an answer is right or wrong. It is designed to understand how that answer is built: what sources it uses, what context it includes, what information it leaves out and where there are opportunities to strengthen institutional communication.
ANIMA starts with real questions
The audit begins with citizens’ actual information needs. These may include frequent searches, questions received through official channels, persistent rumors, narratives circulating in media and social platforms, and concerns related to electoral trust.
The questions should be phrased the way an ordinary person would ask them. For example, users commonly ask:
– How can I confirm that I am registered to vote?
– What identification do I need to bring to the polling station?
– Where and when can I vote?
– What should I do if my name does not appear on the voter list?
– How are ballots counted and results verified?
– When will preliminary and final results be available?
– Where can I find official election results and updates?
– Who can observe or audit the election process?
– How can voters report an incident or request official information?
– Can election results be manipulated?
These questions are then submitted to different AI models to compare their answers and identify patterns.
The five dimensions of ANIMA
ANIMA follows an eight-phase process, from identifying citizens’ information needs to strengthening official content and monitoring improvements over time. Throughout this process, the methodology focuses on five key dimensions:
- Real questions: What citizens are asking about an election, a technology, a procedure or a risk narrative.
- Generated answers: What AI models explain, what they omit and how much context they provide.
- Sources used: Whether the answers rely on election authorities, international organizations, academic institutions, media outlets, social platforms or other open sources.
- Official-source gaps: Where the election authority does not appear, appears only as a secondary reference or does not support the answer.
- Information gaps: Which topics lack a sufficiently clear, complete, visible or structured official explanation.
Source analysis is at the core of ANIMA. An answer may sound reasonable while still relying on secondary, incomplete or outdated information. The audit therefore seeks to determine what place the election authority occupies within the information ecosystem used by AI models.
From audit to action
By implementing ANIMA, election authorities can produce a map of the sources used by each model, identify the questions where the election authority is missing, and show which institutional pages are being retrieved.
ANIMA can also help detect risk narratives, ambiguous content and priority information gaps.
Based on those findings, an election authority can decide what information needs to be published, updated or reorganized. This may include FAQs, explainers on audits, security, vote counting, results transmission, election observation or independent verification.
The goal is not to control what AI says. Nor is it to treat one AI-generated answer as a permanent truth.
Models can produce different results depending on their version, language, internet access, configuration or the timing of the query. ANIMA should therefore be understood as a methodology for identifying and monitoring signals and trends. Its value lies in repeating the questions, comparing results and observing whether improvements to official content strengthen institutional visibility over time.
Becoming a visible and trusted source
ANIMA is based on a simple premise: when the official source is absent, someone else fills that space.
In the age of generative AI, election authorities need to produce information that is not only accurate. It must also be clear, accessible, verifiable, structured and up to date.
The ultimate goal is to increase the likelihood that, when someone asks about an election, the answer is built on trusted institutional sources.
ANIMA helps identify where to begin.
Click here to download the ANIMA guide for election authorities