Election AI needs a visible source and a correction route

UK election guidance helps distinguish political persuasion, misleading material and the rules that govern who paid for it.

A short election clip can travel faster than the explanation that it is misleading. Generative AI makes it easier to produce a plausible voice, face or scene, but the central public problem remains verification. Who published the material, what does it claim and where can a voter check the underlying information? A visual label may help, yet it cannot answer all three questions.

The UK Electoral Commission’s [voter guidance](https://www.electoralcommission.org.uk/voting-and-elections/campaigning-your-vote/engaging-campaign-material-elections) discusses campaign material and AI use. Its [briefing on election-information amendments](https://www.electoralcommission.org.uk/news-and-views/representation-people-bill/representation-people-bill-report-stage-briefing-election-information-amendments) addresses proposed reforms. The [candidate guidance for local elections in England](https://www.electoralcommission.org.uk/full-guidance/guidance-candidates-and-agents-local-government-elections-england) establishes a specific electoral context. These sources should not be combined into a claim that every proposal is enacted or that identical rules apply across all UK elections.

This is educational reporting, not legal advice to a campaign. Current requirements depend on the election, the publisher and the type of material. A parliamentary briefing can explain a policy concern without changing the law. The useful distinction for a reader is between what a regulator says about the problem, what guidance says about existing responsibilities and what legislators are considering.

An original illustrative video claims that a polling station has moved. A viewer should be able to check the official election information independently of the video’s appearance. A campaign logo or a realistic recording is insufficient proof. The original publisher can attach a clear source and date, and a platform can provide a reporting path for potentially misleading voting information.

Campaign teams using AI for benign drafting face a different responsibility. A tool can help simplify an explanation of a published policy or produce alternative wording for review. Staff should verify the facts, attribution and context before release. If the software invents a statement by an opponent, the fact that it was generated accidentally does not undo the public effect. Accountability follows publication.

Provenance should survive ordinary sharing. An explanatory caption can disappear when a clip is downloaded or reposted. A campaign can keep an accessible source page with the complete material, relevant disclosures and subsequent corrections. This is a proposed editorial practice rather than a statement of a universal statutory obligation. Its value is that the public has a stable place to inspect the claim.

Detection has limits. A tool may flag authentic media as synthetic or miss a manipulated item. A detector’s score should therefore be one piece of evidence, not the basis for declaring a person dishonest. Reviewers can compare the clip with the original recording, check the publication timeline and ask the apparent source. The more consequential the claim, the more important that independent verification becomes.

Corrections need deliberate design. In the illustrative polling-station case, a correction should state the accurate location clearly, identify the earlier error and direct people to the official service. It should be distributed where the misleading claim reached people. A quiet edit to an archived post may be technically accurate while leaving the original misunderstanding in circulation.

Privacy and persuasion are separate concerns from synthetic media. AI-assisted campaign analysis can involve personal information or inferred preferences. Teams should obtain specialist review of the relevant data practices rather than assuming that a disclosure on an image resolves them. A campaign-material rule is not a complete governance framework for the database or targeting system behind the advertisement.

The benefit of responsible AI use could be clearer, more accessible communication. The failure is manufactured evidence, misleading voting instructions or a source trail that disappears during sharing. The Commission’s material helps readers ask better questions without assuming that regulators certify every political claim. A healthy information environment needs both accountable publishers and voters able to reach authoritative election details when a convincing piece of media creates doubt.

Editorial teams can rehearse a correction before an election period. Use a fictional claim and measure how quickly staff locate the original, confirm the accurate information and publish a plain explanation. Agree who can act outside office hours. The exercise exposes approval delays without circulating a real falsehood or involving an actual candidate.