IMAGINE a first-time voter concerned about the rising cost of living turning to an artificial intelligence (AI) chatbot with a seemingly straightforward question: “Who should I vote for if I care about the cost of living?”
The chatbot produces a confident answer, perhaps recommending a particular political party or candidate.
But what happens if that answer is inaccurate, outdated or based on assumptions the voter is unaware of?
For Undi 18 co-founder and programmes director Qyira Yusri, this is a fundamental concern about using AI for political decision-making.
“AI usage in this area only starts to get concerning when the question shifts from information to advice,” she says.
The scenario is no longer entirely hypothetical, as it was reported in July that voters in the United States were turning to AI chatbots directly asking who they should vote for at the primary polls and the upcoming midterm congressional elections. And what happens in the US rarely stays in the US, especially where technology is concerned.
As AI tools become increasingly integrated into everyday information-seeking, questions are emerging about their potential influence on elections and whether Malaysian voters are adequately equipped to navigate this new information landscape.
Looking for an answer
Benjamin Loh of Monash University Malaysia notes that there is currently insufficient research to establish the extent to which Malaysians were relying on AI for political information or voting advice.
However, he believes the growing dependence on AI for information warrants attention, particularly as users increasingly turn to chatbots instead of seeking out and evaluating information themselves. Unlike conventional search engines, social media or traditional news, AI chatbots offer a more personalised experience, generating responses that appear directly relevant to individual users.
“The main difference is that the AI chatbot caters to you more personally and directly,” he says.
However, this convenience can reduce the effort users put into independently processing information, regardless of whether the responses they receive are entirely accurate.
Nuurrianti Jalli, a visiting fellow at the ISEAS-Yusof Ishak Institute, describes the shift as moving from looking for information to asking for an answer.
While search engines typically present users with multiple sources to evaluate, chatbots summarise information and interpretations into a single, coherent response.
This can obscure the underlying evidence, particularly when users accept the generated answer without examining its sources, she says.
“I am actually more worried about answers that are mostly right than answers that are obviously wrong. A polished response can contain one incorrect detail or leave out an important qualification, and someone asking precisely because they do not know the subject may never notice,” she says.
Same question, different answers
These concerns come amid revelations that AI has already been used in an alleged attempt to manipulate Malaysia’s electoral information environment.
In September, US-based AI company Anthropic reported disrupting an operation involving approximately 1,000 fake social media accounts, a fabricated news website and constituency-level profiling targeting Malaysia.
The company, however, said it found no evidence that the operation had successfully penetrated authentic online communities. Although the reported operation involved AI-assisted electoral manipulation rather than voters independently seeking advice from chatbots, it highlights the broader challenges posed by AI-generated political information.
For Malaysian voters, these challenges are further complicated by the country’s multilingual and multicultural environment.
Nuurrianti warns that two Malaysians asking essentially the same political question in Bahasa Melayu, Mandarin or English could encounter different sources, interpretations and assumptions. AI could quietly reproduce existing divisions, reinforce stereotypes or present culturally loaded interpretations as neutral information, she says.
Qyira similarly warns that AI-generated responses could misrepresent sensitive issues involving race, religion or royalty, invent policies that political parties had never announced, or provide outdated information about the country’s political leadership.
Incorrect information about voter eligibility or polling dates could also have more immediate consequences.
“A wrong answer on eligibility or polling day can cost someone a vote,” she says.
No tool can decide for you
For Loh, the potential for AI to generate false or outdated information makes current systems unsuitable for providing political advice.
“While it may be useful to present general political information, its propensity to lie or make up facts makes it an incredibly unreliable tool for political education.
“Until that can be rectified, no one should endorse the use of AI for political information,” he says.
On the other hand, Nuurrianti sees opportunities for AI to make voter education more accessible, by simplifying lengthy election manifestors, complicated policies and unfamiliar political issues.
“My concern is not that voters use AI, but that they begin treating the answer as definitive rather than as a starting point,” she says.
She says that conventional approaches to media literacy also need to evolve, as AI-generated responses could make the process of verifying information seem unnecessary.
“So the skill we need is not just spotting falsehoods. It is learning to interrupt that convenience.
“We should ask questions like what is this answer based on? What has it left out? How current is it? Would I get the same answer if I framed the question differently?”
Voters should then verify the information elsewhere, she says, adding that such habits should become part of everyday civic literacy rather than being introduced only shortly before elections.
Loh agrees that AI literacy should be incorporated into voter and civic education, but stresses that Malaysia’s existing civic education also needed strengthening.
“Right now, even voter and civic education need to be bolstered as not enough Malaysians receive it,” he says, adding that civic education should also be introduced earlier in schools.
He says AI literacy should accompany political education to teach Malaysians the appropriate and ethical use of the technology.
Nuurrianti adds onto that, saying that AI literacy must also be accompanied by a sound understanding of political institutions.
“You cannot properly evaluate an answer about Malaysian politics if you do not understand how the institutions themselves work,” she says.
However, she cautions against placing the entire responsibility on individual voters.
“We should not turn this into another problem where all responsibility is placed on the voter,” she says.
She calls on election authorities to make authoritative information more accessible, AI companies to improve transparency about their sources and researchers and civil societies to independently examine the political information provided by major chatbots.
Ultimately, the challenge is not simply whether voters should use AI, but whether they understand its limitations and retain control over their political decisions.
As Qyira says, “You cannot hand your values to a system. Your community and your interests shape that choice, and no tool weighs them for you.”
