Image Courtesy: X
Google’s latest AI search tools are once again drawing attention for making surprisingly basic mistakes, this time struggling to spell simple words correctly and even miscounting letters in the company’s own name.
Users recently discovered that Google’s AI Overview incorrectly claimed there are two “p” letters in “Google,” while also producing bizarre spellings for common words like “journalism” and even the surname of the current US president. The errors surfaced as Google continues expanding generative AI across Search despite earlier controversies involving inaccurate and misleading AI-generated answers, according to TechCrunch.
Google acknowledged the issue in a statement, saying that “counting within words has been a known challenge for LLMs” and that the company is working on fixes. But the problem goes deeper than simple typos.
Large language models, or LLMs, are not designed to understand language the way humans do. Instead of reading words letter by letter, these systems break text into chunks called tokens, which can represent full words, syllables, or fragments of language. That architecture allows AI systems to generate remarkably fluent responses, but it also creates blind spots in tasks humans consider trivial, such as counting letters or spelling words consistently.
Researchers say the issue is rooted in the core design of transformer-based AI systems, the same architecture powering tools from Google, OpenAI, Anthropic, and others. Because the models process numerical representations of language rather than actual characters, they can produce highly intelligent reasoning in one context while failing elementary spelling tasks in another.
The latest incident also revives concerns surrounding Google’s aggressive push to integrate AI directly into Search. Earlier AI Overview rollouts generated widespread criticism after the system surfaced satirical content as factual information and produced dangerous or nonsensical recommendations, including advising users to eat rocks or add glue to pizza recipes.
While spelling mistakes may appear harmless, critics argue they expose a larger issue with public trust in AI-generated information. As these systems become more deeply integrated into search engines, productivity tools, and business software, even small inaccuracies can undermine confidence in outputs that many users increasingly treat as authoritative.
For the AI industry, the situation highlights an uncomfortable reality: modern language models can solve advanced coding problems and accelerate scientific research, yet still fail at tasks most children learn in primary school. Researchers say that limitation may not disappear anytime soon because it stems from how today’s AI models fundamentally process language.
