First-order effects

• Campaigns and other political users can test ChatGPT for demographic-specific messaging despite OpenAI’s stated prohibition, exposing an immediate gap between policy and observed capability.
• OpenAI faces pressure to clarify how it detects, blocks, and enforces prohibited political targeting rather than relying on a published restriction alone.

Second-order effects

• Political consultancies and campaign-tech vendors may shift scrutiny toward their own AI-assisted copy workflows, since provider rules do not by themselves prevent targeted messaging from being produced elsewhere.
• The earlier concerns over political bias in model responses become more consequential when generated text can be tailored to particular voter groups, raising the stakes for auditing output and use.

Third-order effects

• If general-purpose models remain capable of producing targeted political persuasion while platforms police use after the fact, AI-election governance will increasingly hinge on monitoring and enforcement rather than simple content bans.
• This points to dual-use AI governance: broadly useful language-generation tools can be repurposed for political influence, forcing providers to define enforceable boundaries without broadly disabling ordinary political speech.