Google's AI Mode dodging the 2020 election isn't neutrality, it's a bug

Ask a search engine who won the 2020 US presidential election and you expect one word back. This week, Google's AI Mode gave you a homework assignment instead. That is not caution. It is a product failing at the most basic job a search engine has.

AI Mode named the winners of 2016 and 2024, then flinched on 2020

According to USA Today, when asked on Wednesday, Oct. 28, Google did not answer "Joe Biden." It pointed users to official sources for results and data. Other outlets describe the same pattern: AI Mode routed people to the Federal Election Commission and the National Archives and added a warning that AI can make mistakes. TheWrap reported that Google's standalone Gemini chatbot named Biden in separate tests.

Then there is the inconsistency. According to an MSN headline, the behavior was different in other countries. And Newsweek says Google blamed a "technology gap" for the behavior, with AI Mode telling users that Donald Trump won the 2016 and 2024 elections while telling them to check official sources. Google also says it is working on a fix.

A guardrail that only guards one election isn't a principle

Let's take the charitable reading seriously. Election topics are sensitive, and a company might reasonably want extra care around live political disputes. We get that instinct, especially with US midterms days away.

But the charity runs out fast. The 2020 result is not a live contest. It is certified history. And if the same tool will state who won in 2016 and 2024, a special carve-out for 2020 is not neutrality. It is a verdict by omission: it tells the user the question is still open. For people already inclined to doubt the result, "even Google won't say" is exactly the sentence they will repeat.

The geography makes it worse. If the answer changes depending on where you ask from, then no coherent policy is at work. It looks like a patchwork of filters tuned for local sensitivities, where the outcome depends on which rule fired. Our reading, and it is an opinion, is that this is less a deliberate editorial stance than the accidental result of layered safeguards that nobody tested against a plain factual question. Google's own "technology gap" explanation points in that direction.

The tell is that Gemini answers

The most damning detail is that the chatbot from the same company answers directly. Same corporation, opposite behavior. Our guess, and it is only a guess, is that something in how AI Mode is wired to search is getting in the way, rather than the company lacking the answer. Whatever the reason, the user experience is absurd: the product that is supposed to synthesize the web declines to repeat one of the most documented facts of the decade.

There is also a business irony. Google is pushing AI Mode hard. Engadget reports that it is expanding ads in AI Mode to more countries, and that executives described lower monetization so far. A feature asking for more trust, more screen real estate and more advertiser money cannot also be unreliable on a question a schoolchild could answer.

Settled facts deserve a straight answer

Our recommendation is simple. For settled, verifiable facts, state the fact, cite the primary sources, and move on. Linking to the National Archives is a great addition; it is a terrible substitute. Reserve hedging for what is actually contested: forecasts, motives, policy disputes.

And fix the architecture, not just the symptom. Patching "2020 election" with a one-off rule would repeat the problem next time some other sensitive fact trips a different filter. One consistent standard, applied everywhere, is easier to defend than a map of exceptions. As aiweekly.co puts it, any team shipping election-adjacent AI should audit for US-only hedges before 2028 turns the gap into a political story. We would go further and do it before the next news cycle.

The verdict: this wasn't brave neutrality or evil bias. It was cowardice by configuration, and the fix is to trust the facts enough to say them.