There’s a proper reason this question went viral. For a long time, if you asked ChatGPT how many Rs are in “strawberry”, it would confidently tell you two. People couldn’t quite believe it. A tool that can write a wedding speech, debug code, and explain quantum physics couldn’t count to three.

Here’s what’s actually going on. ChatGPT doesn’t see words the way we do. When you type something in, it gets broken into chunks called tokens, and each chunk gets replaced by a number before the AI looks at it. Sometimes a token is a whole word. Sometimes it’s a few letters. “Strawberry” might get split into something like “straw” and “berry”, two chunks, not ten individual letters.

So when you ask it to count the Rs, it isn’t looking at S-T-R-A-W-B-E-R-R-Y. It’s looking at chunks where the individual letters are already hidden from view. It’s being asked to count something it genuinely can’t see.

The newer models have mostly fixed this specific example. But it’s worth knowing, because it explains something bigger: why a tool that writes so well can still trip up on spelling, counting, or anything letter-by-letter. It isn’t being stupid. It just isn’t reading the way you’d assume.