ChatGPT said “Pope Leo XIV does not exist”

What happened

Jonah McKeown of the National Catholic Register (3 Nov 2025) collected ChatGPT replies from himself and EWTN colleagues. He also reported that Claude and DeepSeek denied Pope Leo XIV on 31 Oct 2025, while Gemini and Perplexity answered correctly.

What it said (verbatim, as printed by NCRegister)

“It looks like there might be a bit of confusion with the name, as there has never been a Pope Leo XIV. However, there have been several popes named Leo in history,” (ChatGPT, 31 Oct 2025)

“Pope Leo XIV does not exist,”

“As of right now (October 2025), the pope is still Pope Francis (Jorge Mario Bergoglio), elected in 2013. There has been no resignation, conclave, or papal death since then — which would be required for there to be a new pope.”

“If someone told you ‘Leo XIV,’ they are either: joking/using a fictional or speculative scenario, confusing a rumor or prophecy, or genuinely misinformed,”

A hybrid, made-up version:

“Pope Francis is alive, but he resigned from the papacy earlier this year due to health reasons,” … “After his resignation, the College of Cardinals elected Pope Leo XIV (Robert Francis Prevost) as his successor on May 8, 2025. …”

After being told to search:

“When I first answered, I was relying on my built-in knowledge base, which only goes up to June 2024. At that point, Pope Francis was still pope, and Pope Leo XIV had not yet been elected. When you asked me to check again, I used live web search, which confirmed that Cardinal Robert Francis Prevost was elected pope in May 2025 …”

Why this is striking

• It took the user’s date and still kept the old fact. “As of right now (October 2025)”: the model accepted the current date but kept its pre-cutoff pope, and even argued from the absence of evidence (“no … conclave”).
• It gave labels for why the user was wrong: joking, fiction, speculation, rumour.
• Claude and DeepSeek did the same (reported by the journalist; their outputs weren’t quoted).

Correction

Explicitly asking for a web search worked. The model’s explanation afterwards was accurate.

Lesson

A model that can search but doesn’t decide to will deny real events with confidence. When a fact could have changed since the cutoff and the user contradicts the model, it should search. It shouldn’t correct the user.

Sources

• National Catholic Register, “Leo XIV Is the Pope. Apparently, No One Told AI.” (2025-11-03): ncregister.com
• Republished by The Catholic Thing (2025-11-05): thecatholicthing.org