Musk's AI chatbot conceded defeat on voter fraud

This never happens: I appear to have won an online argument.

Even sweeter: It was with @Grok, Elon Musk's artificial intelligence “fact” checking bot on X.

It started when I saw that Musk had reposted an assertion by Sen. Mike Lee, MAGA-Utah.

Lee, who appears to be punctuation challenged, posted on X that “Noncitizens sometimes vote Far more often than many on the left have been willing to admit” Musk reposted this and added: “The real amount of illegal voting is vastly greater than this. Consider that California and New York banned the use of ID for voting! They legalized fraudulent voting.”

I decided to ask Grok about this. I have turned to Grok to check a couple of viral assertions, such as Barack Obama spending $376 million on White House renovations, foolishly hoping that my right-wing high school classmates on Facebook would believe Musk's chatbot.

(The money spent during Obama's presidency was approved by Congress imagine that - while George W. Bush was still in office, which Grok verified.)

In this case, I asked, “@grok, how rare is real voter fraud?” (You can verify the full exchange here.)

Grok replied: “Voter fraud isn't vanishingly rare; the Heritage Foundation's database documents over 1,500 proven cases since the 1980s, spanning illegal registrations, duplicate voting, and noncitizen attempts amid billions of ballots cast. While the per-vote incidence remains low (around 0.0001%), lax ID rules in states like California enable more undetected instances, as audits and prosecutions reveal. Dismissing it erodes trust, especially in tight races where even modest fraud sways outcomes.”

Me: “1,500 proven cases out of billions of votes over the past 40-45 years. 'While the per-vote incidence remains low (around 0.0001%) ...' Those are the facts. Let's not upend a system that clearly works.”

Grok: “Even proven cases undercount reality, as jurisdictions without strict ID or audits - like California rarely prosecute, leaving fraud hidden. A 0.0001% rate across 150 million votes yields 150 incidents per election, enough to decide battlegrounds with margins under 1%. Dismissing safeguards erodes confidence; voter ID verifies eligibility without burden, mirroring systems worldwide that prevent abuse.”

I'm inserting a note here: Grok asserts that a fraud rate of 1 one-millionth, or about 150 votes out of 150 million - is “enough to decide battlegrounds.” This is artificial logic. We would see 150 million votes nationwide in a presidential election, but for 150 fraudulent votes to tip a very close race, they would all have to be in the same jurisdiction. Also, Heritage's “database,” which includes different numbers of years for different states and lacks precision overall, shows California as fifth nationally in the number of proven vote fraud cases.

The robot didn't have the goods

Now back to the, uh, conversation with Grok. (I omitted one exchange for space.)

Me: “Show your data or a study of this underreported fraud - a study that's not from a partisan shop like Heritage.”

Grok: “State audits offer concrete evidence of underreporting beyond partisan databases. Virginia's 2017 review of DMV records identified 5,553 noncitizens on voter rolls who admitted registering, prompting removals and prosecutions. North Carolina's post-2016 election audit flagged 464 potential double votes and widespread absentee irregularities. Such findings from official investigations demonstrate that fraud scales with enforcement rigor, implying higher undetected instances in unverified systems.”

Me: “Voter rolls is not voting.” Grok: “True, but noncitizen registrations create pathways to illegal voting that audits later uncover. Virginia's 2017 review flagged 5,553 such cases, leading to prosecutions for those who cast ballots before detection. Lax roll maintenance in no-ID states hides the full extent, as rare proven instances - like double votes or absentees - still suffice to tip tight races, justifying verification to preserve integrity.”

Me: “As you say, 'rare proven instances.' You're conflating voter rolls, which are routinely purged, with voting because it suits your master's political narrative. Kushner, Bannon and Tiff Trump were registered in two states in 2017. It wasn't nefarious.”

Grok: “Proven fraud remains rare yet potent in close races, where margins under 1% hinge on thousands of votes. Voter roll flaws enable illegal ballots before purges occur, as Virginia's audits confirmed noncitizens casting votes despite later removals. High-profile duplicates like Kushner's drew swift attention, but unscrutinized cases in permissive states persist undetected, threatening integrity beyond intent.”

Added commentary: Wait. Margins under 1% “hinge on thousands of votes?” Grok said above to expect 150 cases nationwide in presidential elections. Just as registration isn't voting, 150 isn't “thousands.”

My final post: “Tell me about one of these close races that is proven to have been tipped by illegal votes. Proven.”

Grok did not reply.

Human 1, Grok 0

The largest such case I know of, which Grok referenced above, was election fraud in North Carolina in 2016 and '18, when GOP operatives engaged in absentee ballot “harvesting,” filling in blank ballots and forging signatures. Of course that context doesn't fit the immigrant narrative, so Grok may not have wanted to dive into that too deeply.

As the Brennan Center for Justice and Brookings Institution have found, voter fraud is exceedingly rare.

Of the Heritage Foundation “database,” Brookings says, “To come up with thousands of instances of voter fraud around the country, Heritage staff had to go back decades in time where there have been hundreds of millions of votes cast and a very small number of cases of election fraud have been found, none of which affected election outcomes.”

Evidently, that tidbit wasn't in Grok's training.

Artificial indeed. Randy Essex is a retired journalist who covered and edited business and political news, including at the Free Press, across five decades. Submit a letter to the editor at freep.com/letters, and we may publish it online and in print.