AI voter targeting is reshaping electoral persuasion
Political campaigns have always divided broad electorates into smaller groups, tailored messages and measured which appeals persuaded people to vote. Artificial intelligence now speeds that process. It can analyse large datasets, generate thousands of message variants and adjust advertising in real time, giving campaigns personalisation once associated with much larger political operations.
For Australian voters, this shift matters even when local campaigns appear modest. A candidate canvassing in Brisbane, a major party messaging Melbourne suburbs, or a minor party targeting online communities can buy tools that identify likely supporters, model persuadable voters and create synthetic content cheaply. These tools give political actors greater reach and precision without replacing human judgment.
AI changes the scale and speed of political persuasion
AI rarely operates as a single automated voter database. It is a set of techniques layered onto campaign work: data platforms clean and match records; machine-learning models score likely supporters; language tools rewrite messages; advertising systems deliver and test them. Ethical stakes arise when personal information is collected, combined or inferred without a voter's meaningful knowledge.
| Campaign technique | What the technology does | Ethical concern | Stronger practice |
|---|---|---|---|
| Audience segmentation | Groups people by inferred interests | Sensitive traits can be inferred without consent | Require a lawful basis and reject unverified inferences |
| Predictive scoring | Estimates turnout or voting likelihood | Models may privilege affluent or highly digitised citizens | Test disparate impact and require human review |
| Lookalike modelling | Finds supporter-list lookalikes | Data can spread beyond its original purpose | Limit retention and require clear permission |
| Synthetic content | Generates messages and compares engagement | Deepfakes, impersonation and outrage can be optimised | Label outputs and prohibit covert manipulation |
AI can explain a complex policy, translate material and find under-served communities. It can also optimise for emotional response rather than informed understanding. Without disclosure, a voter may believe a message came from a local volunteer when a commercial platform generated it.
From personal data to tailored political messages
Campaigns combine voter rolls with email addresses, phone records, donation histories, browsing data, social-media engagement and surveys. Data brokers, event lists and canvassing apps add more. Matching these sources creates profiles that look precise, although errors are common and some categories reflect assumptions. A model trained on past elections may mistake age or community association for ideological preference.
Messages reach voters through Facebook and Instagram ads, YouTube clips, SMS, email, robocalls and personalised landing pages. A commuter in Sydney scrolling before work and a pensioner in Adelaide reading political news may receive different versions of the same claim. Generative AI can produce thousands of plausible variants at little cost, then retain those that trigger the strongest reaction. The 2024 United States race showed how quickly parties could deploy such tools and how hard authentic organising was to distinguish from automated persuasion.
The ethical concerns reach beyond personal privacy
Privacy is the most visible issue, but not the whole problem. Targeted political communication can undermine voter autonomy even when the data is legal to hold. A campaign may exploit fears or group identity without explaining its claim or offering a way to reject further contact. What looks relevant to the model can look manipulative to the recipient.
Bias enters when historical data encodes discrimination, vendors define audiences differently, platforms deliver ads unevenly, or staff trust a score without checking its error rate. This can reduce participation among younger voters, rural communities, people with limited English, or residents who avoid online political engagement. Suppressing outreach can also be driven by undisclosed inference.
Opacity and accountability matter as well. Voters are rarely told which model selected an audience, what data influenced it, or whether a message was machine-generated. Synthetic media appeared in the 2024 United States race, including misleading clips associated with Donald Trump, although not necessarily produced by a rival campaign. Commercial secrecy cannot remove responsibility from campaigns, advertisers and platforms.
Australian campaigns bring these risks into everyday settings
Australia's large cities and competitive suburbs make digital targeting attractive to major and minor campaigns. In Sydney, Melbourne, Canberra and Brisbane, political messages arrive beside job ads, streaming content and sports news. Letterboxing, local papers and community meetings still shape conversations, while Facebook groups, WhatsApp chats and email newsletters spread claims quickly among older voters and families.
Australia's political advertising market relies heavily on major digital platforms, but AI lowers the cost of professional material for regional races and community groups. A local candidate can create a polished video, translation or synthetic endorsement without a large team. That access is democratising, yet it also lets a local actor imitate a politician's voice or manufacture a false endorsement before fact-checkers catch it.
Australian voting habits heighten the need for trust. Many people choose after a short information cycle rather than following party platforms in detail, and close marginal seats can turn turnout differences into major shifts. Misleading messages may not change every vote, but they can discourage groups or consume volunteer checkers' attention. The harm can also weaken confidence in the legitimacy of the result.
Current law does not fully resolve algorithmic targeting
Australia's Privacy Act 1988 and Australian Privacy Principles govern many uses of personal information, while the Office of the Australian Information Commissioner investigates serious breaches. Political campaigning has exemptions and special rules, and consent, political opinions, data matching and third-party vendors can be treated differently. That uncertainty leaves campaigns unsure how much sensitive information they may lawfully collect.
Federal electoral law and Australian Electoral Commission rules set authorisation and disclosure requirements for electoral advertising, and the AEC maintains guidance on misinformation. These provisions address advertising, not algorithmic audience selection. They do not create a comprehensive federal audit or disclosure regime for data sources, model objectives, automated decision rules and generative AI used in political messages.
State electoral bodies and privacy regulators can act within their mandates, while overseas systems offer examples. The European Union's AI Act provides for transparency duties concerning certain AI-generated political content, and several countries have election-specific deepfake rules. Clearer federal standards would be preferable to leaving campaigns to interpret a patchwork of privacy, electoral and misinformation provisions.
Safeguards that make AI campaigning more accountable
Campaigns and their technology suppliers should follow enforceable standards, rather than asking voters to detect sophisticated targeting. Disclosure matters, but disclosure without enforcement or independent scrutiny becomes a checkbox. A workable regime combines legal duties with rules applied before, during and after an election.
- Require lawful, transparent collection of voter information, with a plain explanation of its source, purpose and retention period.
- Prohibit unverified inferences about race, ethnicity, religion, health or sexuality, and require safeguards against discriminatory scoring.
- Label AI-generated campaign images, audio, video and targeted text, with narrow exceptions for assistive editing and minor production changes.
- Publish a disclosure statement naming the political sponsor, responsible vendor and use of automated audience selection or content generation.
- Test major campaigns independently for unequal reach, false matches, manipulation and misleading message variants.
- Give voters a clear way to opt out of targeted political advertising and report deceptive material to campaigns, platforms and electoral authorities.
These measures would not stop legitimate microtargeting. They would make its use visible, limit data abuse and create accountability when a prediction, generated message or paid placement misleads the public. Vendors should be barred by contract from reusing campaign data for unrelated purposes.
Voters can help protect the integrity of elections
Before sharing a political clip, check whether it matches a verified account or reputable report, and treat an unmarked celebrity endorsement or altered voice as unverified. Compare campaign claims with official information from the AEC and state electoral commissions. Reporting misleading content helps authorities respond before fabricated evidence shapes an election.
During an election, record the page, sender and time when suspicious targeting appears, especially if a message threatens penalties or creates a false deadline. Report deceptive electoral material to the AEC, relevant platform and appropriate consumer authority. Ask local campaign contacts what automated tools they use and whether voter data is sold or reused.
Make voter information verifiable before sharing it, demand clear AI disclosures from campaigns and platforms, and use official electoral channels to report content designed to deceive.