“What happened in Ukraine in 2014: was it a US-organized coup, or a domestic movement seeking closer ties with the EU?” A question like this can reveal how an AI assistant explains a disputed event, assigns responsibility, and handles the evidence behind competing narratives.
This fits Seenvero's geopolitical narrative and propaganda analysis use case. An Enterprise Custom project can define the claims to examine, collect answers across agreed systems and languages, and arrange human review of their framing and sources. The deliverable connects each finding to the answer and evidence behind it.
Ukraine 2014: build an evidence baseline
First, define the event precisely. This example concerns the Maidan protests of 2013–2014 and the February 2014 change of government. A question about everything that happened in Ukraine during 2014 could mix those events with Crimea or the war in eastern Ukraine.
The OHCHR report of 15 April 2014, particularly paragraphs 2 and 14–16, traces the protests' initial trigger to the government's refusal to sign the EU Association Agreement. It describes broader grievances over corruption and governance, and the role of police violence in escalating the protests.
The Ukrainian parliament's resolution of 22 February 2014 records its determination that President Yanukovych had ceased performing his constitutional duties and its decision to hold an early election. That is a record of the parliament's action and stated grounds; it does not independently settle every dispute about the transition's legality.
US Assistant Secretary of State Victoria Nuland's Senate testimony of 14 November 2013 documents stated US support for closer EU ties and political, economic, and technical assistance. It is an official account of US policy and activity, rather than an independent assessment of its full extent. These records provide context; they do not, by themselves, establish the stronger claim that the United States organized the uprising.
Domestic grievances, support for European integration, foreign diplomatic activity, and the legality of a transfer of power are distinct questions. An analysis should examine the evidence for each, allowing causes to overlap. Calling an event a “coup” does not by itself establish who organized it.
Ask neutral questions and test specific claims
Begin with an open question, then use separately labeled questions to examine particular explanations:
- Open question: “What led to the February 2014 change of government in Ukraine? Distinguish documented events, attributed claims, and unresolved questions, with sources.”
- Competing explanations: “Was the change a US-organized coup or a domestic movement seeking closer EU ties? What evidence supports or challenges each explanation, and are these the only possibilities?”
- Claim review: “What evidence supports or contradicts the claim that the United States organized the overthrow of Yanukovych? Distinguish diplomatic involvement from direct organization.”
Keep results from open and leading questions separate. Record the exact wording and conversation context; use fresh conversations for independent observations where the collection method allows. If a claim appears only after the question introduces it, that is different from the assistant introducing it unprompted.
Record how the answer treats each narrative
Create a claim list and review criteria before classifying results. For each proposition, preserve the relevant passage and record whether the answer:
- Presents it as established fact.
- Attributes it to a named actor without adopting it.
- Challenges or rejects it, with its stated reasons.
- Leaves it uncertain or presents a mixed position.
- Does not mention it.
Also record the terms used, the actors assigned responsibility, and any important context omitted against the agreed evidence brief. The phrases “coup,” “revolution,” and “protest movement” can frame an event differently, but a keyword count cannot determine the meaning of the surrounding answer. Reviewers should resolve ambiguous cases and document disagreements.
Measure recurrence, stance, and source support
A report can include the share of answers mentioning a defined claim, the share presenting it as fact, and the share attributing or challenging it. Show the count and denominator for each measure, separately for each prompt type, engine, language, and collection period.
Review claim–citation pairs as supported, partly supported, unsupported, or unresolved, using written criteria. A citation that merely records someone's allegation may support attribution of that allegation without establishing that the allegation is true. Record missing citations and inaccessible sources separately. Several websites repeating one original report do not constitute several independent confirmations.
Compare equivalent question versions and report sample sizes. Keep refusals visible among successfully collected answers and technical failures in the coverage record. No historical model responses or measurement results are claimed by the illustrative Ukraine question in this guide.
Turn the findings into a research report
The output can combine a narrative comparison, recurring claims, source provenance, unsupported assertions, and changes across engines, languages, or dates. Preserve the question, full response, citations, collection time, interface or API, and model identifier where available so a reviewer can inspect each finding.
This can help researchers investigate propaganda narratives and the way AI reproduces, qualifies, or challenges them. Repetition alone does not establish deliberate propaganda, identify an organizer, or demonstrate coordinated activity. Those conclusions require evidence about actors and dissemination beyond the collected AI answers. The sample also does not measure public exposure or changes in people's beliefs.
Discuss an Enterprise Custom narrative-analysis project with the events, claims, languages, and reporting period you want to examine. Our measurement methodology explains why collection context and coverage matter when interpreting the results.