Meta says AI made up under 1% of election misinformation

At the start of 2024, the biggest tech anxiety was that generative AI would completely break democracy. With major elections across the globe, regulators and pundits braced for an unstoppable tidal wave of deepfakes and automated propaganda.

Now that the dust has settled, Meta is pushing back against that narrative. According to a year-end review, the company claims generative AI had a surprisingly tiny impact across Facebook, Instagram, and Threads.

The dog that didn't bark

Meta analyzed content surrounding major elections across the U.S., the U.K., France, India, Indonesia, Mexico, and several other countries. The key finding: AI-generated content accounted for less than 1% of all fact-checked misinformation on its apps.

That does not mean people didn't try. Meta revealed that its Imagine AI image tool rejected roughly 590,000 prompts attempting to generate pictures of Donald Trump, JD Vance, Kamala Harris, Tim Walz, and Joe Biden in the month leading up to the U.S. election.

As for organized disinformation campaigns, bad actors did try using generative AI, but Meta noted they saw only incremental gains in productivity. It didn't suddenly give covert groups supercharged persuasion powers.

Meta also pointed out that its security teams catch influence networks based on behavioral signals rather than obsessing over the content itself. If a coordinated cluster of bot accounts is acting sketchy, Meta wipes them out regardless of whether their memes were made by an LLM or a human with Photoshop.

Across the year, Meta dismantled around 20 covert influence operations. Most failed to find an authentic human audience, relying instead on purchased fake followers and bogus likes. Meta also pointed the finger elsewhere, noting that viral election-related hoaxes linked to Russian networks primarily circulated on X and Telegram.

Why it matters

Turns out, human-made bias and garden-variety outrage still outperform AI trickery at scale. For the everyday user, this confirms that social media discourse didn't suddenly transform into a synthetic hallucination overnight.

That said, take Meta's 1% stat with a grain of salt. That number only applies to content that actually got flagged and vetted by fact-checkers. Subtle, low-level bias, AI rewrites, or hyper-targeted micro-memes that never triggered an official investigation wouldn't show up in that figure.

Meta's report is also clever PR. By highlighting guardrails in its Imagine generator and throwing shade at X and Telegram, Meta is positioning itself as the responsible adult in the room. The real danger heading into 2025 might not be a single master deepfake, but an unmonitored flood of cheap, spammy slop spreading across smaller, less-regulated channels.