Google AI overview can explain rocket science, but struggles to spell simple words

Google’s ambitious AI makeover for Search is once again attracting attention for all the wrong reasons. Only this time, the controversy is not about recommending glue on pizza or suggesting people eat rocks. Instead, Google’s AI Overview appears to be losing a battle against basic spelling.

Social media users this week began posting screenshots of Google’s AI-generated responses confidently giving incorrect answers to questions that most schoolchildren could solve without difficulty. Ask how many “p” letters appear in the word Google, and the AI reportedly says two. Ask how many “d” letters are in journalism, and the system somehow invents a new spelling entirely.

The bizarre responses have quickly become internet comedy material, but they have also reignited larger concerns about Google’s growing reliance on generative AI inside Search, particularly for tasks that were already simple and reliable before artificial intelligence entered the picture.

Google AI Overview has a spelling problem

Among the more widely shared examples online was Google’s AI Overview incorrectly spelling the word journalism as “j-o-u-r-n-a-d-i-s-m” while simultaneously claiming the word contains two “d” letters. In another case, it correctly identified that there is one “p” in the surname of the US president, only to spell it as “t-r-p-u-m”.

More from Tech

Google acknowledged the issue, saying that counting letters within words remains a known challenge for large language models, or LLMs, the technology powering modern AI chatbots and text generators.

The limitation is not exactly new. For years, users have tested AI systems with deceptively simple questions like how many “r” letters appear in strawberry, often producing wildly incorrect answers. Ironically, the same AI systems capable of writing software code or solving advanced scientific problems still struggle with tasks that rely on precise spelling and character recognition.

The reason lies in how these models process language. Unlike humans, AI does not “read” words letter by letter. Instead, most systems break text into tokens, which may represent syllables, fragments or entire words. The AI then converts those tokens into numerical patterns rather than understanding spelling in a traditional sense.

Google AI Overview has another issue

The spelling chaos arrives alongside another awkward AI Overview moment involving the word “disregard”. Users searching the term were greeted not with a straightforward dictionary definition, but with what looked suspiciously like an AI chatbot response.

Instead of explaining the meaning of the word, the AI displayed: “Understood. Let me know whenever you have a new prompt or question!”

And there is more. Google AI worked more like a conversational model with simple words like 'Forget', 'Ignore', 'Stop' and more.

The response quickly went viral, partly because it perfectly captured growing criticism surrounding Google’s redesigned Search experience. Many users argue that AI Overviews increasingly dominate the page while genuinely useful links are pushed further down.

For simple searches like definitions or spellings, critics say the AI layer often feels unnecessary and overcomplicated. What once took seconds through clean search results is now becoming a conversational AI interaction that not everyone asked for.

Google has already patched the “disregard” issue, but the latest spelling blunders suggest the company’s AI-first Search future may still need a spellcheck of its own.