Where election narratives travel
Voters talk about politics in the language they use at home. In much of Southern Africa that can mean several languages within one household, and a single post may switch between them mid-sentence. Local-language radio clips, voice notes forwarded in messaging groups and community pages carry claims to audiences that rarely read national English-language media.
Narratives also move between platforms. A claim may start in a closed messaging group, reappear as a short video on a public platform and then reach a wider audience through screenshots shared back into other groups. Each hop can involve translation, and each translation can change the claim slightly.
Monitoring built around English keywords misses most of this. It also misses the slang, transliteration and spelling variants that native speakers recognize immediately and that a keyword list does not.
Much of the content is not text at all. Voice notes, short videos and images with captions burned into them carry claims that text-based keyword monitoring cannot read, so transcription and extraction of on-screen text in the relevant languages belong in basic coverage rather than in a later phase.
What gets missed
The narratives that do the most harm during an election are often practical rather than ideological. One of the most common is also the easiest to overlook: a claim already corrected in English that keeps circulating in another language, untouched for the audiences most likely to act on it.
Narratives that target the mechanics of voting include:
- False information about polling dates, times, locations or identification requirements
- Claims about how ballots are marked, counted or destroyed, intended to discourage turnout
- Impersonation of the electoral management body, a party or a public official
- Framing that sets one community, region or language group against another
- Translated versions of claims already corrected in English, circulating again in a language the correction never reached
Signals of coordination
Content analysis tells analysts what is being said. Behavior analysis tells them whether the spread is organic. The two need to be read together, because genuine public concern and a coordinated campaign can carry the same message.
Coordination by itself is not the problem. Parties, civil society groups and electoral bodies all organize supporters to share messages, and they do so openly. The concern is coordination that hides who is behind it, such as a network of accounts presenting as unconnected ordinary citizens, or a single operator running many personas.
None of the signals below proves a campaign on its own. Together they support a hypothesis that an analyst tests against account histories, network structure and timing, and the finding should record which evidence supports it and how strong that evidence is.
Signals that point toward coordination include:
- Clusters of accounts created within a short period, with similar names, profile images or descriptions
- Identical or near-identical text published by different accounts within seconds or minutes of each other
- The same images, videos or links shared by accounts with no other visible connection
- One narrative appearing in several languages at the same moment, which points to a central source rather than independent translation
- Amplification that comes mostly from new or long-dormant accounts
Responding in the open
The response matters as much as the detection. A government or electoral body that answers a disinformation campaign with fake accounts, covert amplification or automated bots damages its own credibility and the public trust in the process it is trying to protect. Those methods should be ruled out entirely, whatever the pressure of the moment.
Effective responses are transparent and accountable. They are:
- Published from identifiable official accounts, under the name of the body responsible
- Approved by a human reviewer before release, with the approval recorded
- Written in the same languages, and posted on the same platforms, where the claim is circulating
- Checked by native speakers before release, since an awkward or inaccurate translation can end up reinforcing the claim it was meant to correct
- Specific and factual, pointing people to authoritative information on how, when and where to vote
- Aimed at false claims and deceptive behavior, never at opinions or legitimate criticism
Preparing before the campaign period
Most of this work cannot be set up in the final weeks before polling day. Language coverage, a baseline of normal conversation, approval chains and verified official accounts all need to be in place early, so that unusual activity stands out and a response can go out quickly.
Dark Pools Gov AI supports this preparation with social media intelligence across 9 platforms in 11+ languages, network analysis that exposes coordinating accounts, and counter-disinformation workflows that trace narratives to their source. Responses are published only after human approval, through official accounts, and never through covert bots or fake accounts.
Agreeing the protocol with electoral authorities in advance, including who approves which kinds of response and how quickly, makes it much easier to act in public and on the record when a damaging narrative starts to move.