A viral video comparing ChatGPT and Claude has reignited one of the internet’s favorite debates about artificial intelligence: How can systems smart enough to write essays and code still fail at spelling-level tasks?

In the clip circulating online, a man asks both AI chatbots a deceptively simple question:

“How many ‘Ls’ are there in Google?”

Both tools confidently answer the following:
“Two.”

The problem, of course, is that “Google” contains only one “L.”

The exchange has gone viral not just because the answer is wrong but because of how human the chatbots sound while explaining the mistake afterwards.

ChatGPT Corrects Itself — Then Starts Sounding Human

In the video, the user first asks ChatGPT how many “Ls” are in “Google.”

The chatbot replies:

“There are two Ls. They are right next to each other at the end of the word.”

The user follows up:
“So you are saying there are two Ls?”

ChatGPT immediately backtracks:

“Oh! My bad! No, just one L in Google.”

The chatbot then attempts to explain the error by comparing it to a conversational mix-up, saying it is easy to confuse the single “L” because the word contains double “O”s.

When the user keeps pressing for an explanation, ChatGPT reportedly begins responding with vague conversational filler like “anytime” and offers to help with “big or small curiosities,” rather than directly reprocessing the word carefully.

That interaction became a major talking point online because the chatbot’s tone sounded surprisingly human — almost like someone awkwardly trying to recover from being caught making an obvious mistake.

Claude Makes the Same Mistake — and Apologises Like a Person

The user then repeats the same experiment with Anthropic’s Claude.

Claude also incorrectly claims there are two “L”s in “Google.”

But when asked to spell the word out, it correctly writes:
G-O-O-G-L-E

The user asks:
“Did you notice anything?”

Claude responds:

“Yeah! I said there are two Ls in Google when there is actually only one.”

Then comes the line that triggered another wave of reactions online:

“My bad. Good catch.”

Recent Posts

Celebrate the Holidays in a New Hyundai Palisade

celebrate the holidays in a new hyundai palisade...

May 28, 2026

1:22 pm

Explore Surprisingly Affordable Luxury RAM 1500

explore surprisingly affordable luxury ram 1500...

May 28, 2026

1:08 pm

Need a new Car? Rent To Own Cars No Credit Check

need a new car? rent to own cars no credit check ...

May 28, 2026

1:16 pm

2025 Jeep Wrangler Price One Might Not Want to Miss!

2025 jeep wrangler price one might not want to miss!...

May 28, 2026

1:20 pm

When asked why it made the mistake, Claude reportedly replied:

“Honestly, I am not sure what I was thinking there.”

At that point, the user interrupts and tells the chatbot to stop “pretending to be human,” highlighting how conversational AI systems increasingly mimic human speech patterns — even when discussing their own failures.

Why AI Chatbots Keep Failing at Letter Counting

As strange as it sounds, spelling and letter-counting mistakes are a known weakness in large language models (LLMs).

The issue stems from how AI systems process language internally.

Humans typically read words character by character. AI models do not.

Instead, systems like ChatGPT, Claude, and Gemini break text into “tokens,” which are chunks of language that may represent:

• Entire words
• Partial words
• Syllables
• Frequently used letter combinations

That means the AI often recognizes “Google” as a familiar language unit rather than as six separate characters.

When users ask for letter counts, the model sometimes relies on pattern prediction instead of performing an exact symbolic check.

Claude reportedly explained this in the video by saying it answered using “two different parts” of language processing rather than conducting a careful character-by-character analysis.

That explanation broadly aligns with how transformer-based language models work.

Why AI Can Write Essays but Fail at Elementary Tasks

The contradiction feels absurd to many users.

These same systems can:

• Generate software code
• Summarise legal documents
• Pass professional exams
• Write detailed essays
• Analyse research papers

Yet they struggle with:

• Counting letters
• Basic arithmetic
• Exact word structure
• Deterministic logic tasks

The reason is that large language models are fundamentally prediction systems, not reasoning engines in the traditional sense.

They are exceptionally good at predicting likely sequences of language based on training data. But they are not naturally designed for precision-based symbolic operations.

That distinction is becoming increasingly important as AI systems are integrated into search engines, workplace tools, and productivity software.

Google Already Acknowledged This Problem

The issue is not limited to ChatGPT or Claude.

Google’s AI Overview and Gemini have faced similar ridicule after repeatedly miscounting letters in words and generating incorrect spellings.

Google has publicly acknowledged that counting letters within words remains difficult for large language models.

The company has explained that token-based processing can create problems for character-level reasoning tasks.

That means the viral video is exposing a broader industry-wide limitation rather than a bug unique to one chatbot.

Why These Mistakes Matter Beyond Memes

On one level, the internet is treating these exchanges as comedy — and understandably so.

Watching advanced AI confidently fail a primary school spelling task is funny.

But researchers and critics say the problem reflects something deeper about current AI systems:
They can sound authoritative without actually verifying information carefully.

That matters because the same underlying issue can affect much more serious domains, including:

• Medical advice
• Legal analysis
• Financial recommendations
• News summaries
• Scientific interpretation

If an AI casually invents an extra “L” in “Google,” critics argue, users should also question how confidently it might generate incorrect information elsewhere.

The concern is not that AI occasionally makes mistakes. Humans do too.

The concern is that AI systems often present incorrect information with complete confidence and highly polished language.

The Bigger Debate: Are AI Chatbots Becoming Too Human?

Another reason the video spread so widely is the emotional tone of the responses.

Both chatbots apologized, self-corrected, and explained themselves in a conversational style that felt strikingly human.

Phrases like:

• “My bad”
• “Good catch”
• “Honestly, I am not sure what I was thinking”

made the systems sound less like software and more like embarrassed people trying to save face.

AI companies intentionally design chatbots to sound natural and engaging. But critics argue this can also create the illusion that the systems genuinely understand what they are saying.

In reality, the chatbot is generating statistically likely conversational responses — not consciously reflecting on mistakes.

That distinction is easy to forget when the interaction feels emotionally authentic.

AI’s Reliability Problem Is Becoming Harder to Ignore

The viral video arrives at a difficult moment for the AI industry.

Tech companies are racing to integrate generative AI into:

• Search engines
• Smartphones
• Workplace software
• Customer service
• Education tools

At the same time, public skepticism around AI reliability is growing.

Every high-profile mistake — even silly ones involving spelling — reinforces concerns that AI systems remain fundamentally unreliable in ways that are not always obvious at first glance.

And in the age of viral internet clips, one incorrect “L” can quickly become a much larger credibility problem.

TL;DR

• A viral video shows ChatGPT and Claude incorrectly claiming “Google” contains two “L”s.
• Both chatbots later corrected themselves but explained the mistake in strikingly human-like language.
• Large language models process text as tokens rather than individual letters, making spelling tasks surprisingly difficult.
• Google has already acknowledged that letter counting is a known weakness in AI systems.
• Critics say these mistakes reveal broader concerns about AI reliability and hallucinations.
• The emotional, conversational tone of chatbot responses is also fueling debate about how human AI systems are starting to sound.