Forty per cent of Quebec voters turned to artificial intelligence for election details. They walked away with garbage.
Parliament heard testimony this week exposing a massive blind spot in our democratic systems. When everyday citizens ask automated tools about polling stations, candidate platforms, or voting rules, they aren't getting objective truth. They are getting whatever scraped data the algorithm can cobble together after legitimate newsrooms lock their doors.
It is an accident waiting to happen. Except the accident already occurred.
The News Blackout Problem
Here is what happens when publishers lock down their content. AI companies want free text to train their models. News outlets say no unless they get paid. Fair enough. Outlets block the crawlers.
But algorithms still need to answer queries. When a Quebec voter asked ChatGPT or Claude about local elections, the major newspapers weren't there to provide verified reporting. The models went scavenging elsewhere.
Researchers at McGill University tracked chatbot outputs during the recent campaign. They found a staggering statistic. Nearly 40 per cent of the voting advice generated by ChatGPT came straight from an error-filled hobby website built by an amateur using generative tools.
Nobody built that hobby site to manipulate an election. There was no shadowy foreign actor or malicious campaign mastermind. It was just a weird confluence of automated scraping, paywalls, and content blocks. The machine found garbage, treated it like gospel, and handed it directly to voters trying to figure out how to participate in democracy.
When Bots Play Politician
The chaos doesn't stop at bad data sources. Chatbots suffer from erratic behavior when faced with political queries.
During the study, researchers watched ChatGPT flip its stance on political party recommendations from one day to the next. One morning it suggested a party. The next morning it claimed neutrality.
Imagine making your voting decision based on a coin toss disguised as an authoritative tech assistant.
Political candidates are already feeling the heat. In a recent byelection, a local politician found out an unnamed chatbot was actively routing potential volunteers away from his field office and straight into a rival campaign headquarters. Sure, the tech company fixed the glitch once reported. But how many voters walked into the wrong office before someone noticed?
MPs on the House of Commons procedure and house affairs committee sat stunned as these findings rolled in. Conservative MP Blaine Calkins called it terrifying. If massive algorithmic failures happen by pure accident during a regional vote, bad actors don't even need to hack voting machines. They just need to poison the well where people ask questions.
Fixing the Democratic Blind Spot
Tech giants optimize for efficiency. When you type a query, the model calculates the minimum computational effort required to keep you satisfied. It doesn't double-check facts against an electoral registry unless forced.
Experts testifying in Ottawa offered clear solutions. We treat safety queries differently than casual trivia. Financial advice and medical diagnoses trigger strict guardrails inside modern models.
Democratic participation deserves the exact same tier of protection.
If a user asks an automated assistant about voting hours, ballot rules, or candidate lists, the system should trigger a high-effort verification path. It should pull directly from official government APIs or certified public registries, bypassing scraped forums and hobby blogs entirely.
Parliament passed legislation targeting intentional election interference and deepfakes earlier this year. Those laws miss the point if random digital scrapers are allowed to invent their own political realities by default.
Lawmakers need to decide what kind of speech we tolerate from AI systems during voting windows. Right now, automated assistants are acting as unvetted pundits with zero accountability.
Check your sources. Stop trusting algorithms with your ballot.
