Which AI Can Count Words?
ChatGPT, Claude, Gemini — they all claim they can count words. Here's how each actually performs, and why the answer might surprise you.
It seems like it should be a simple task. You paste some text and ask an AI to count the words. But if you've tried this across a few AI tools, you've probably noticed that the answers differ — sometimes by a lot. That's not random error. It reflects something fundamental about how large language models process text.
Why AI models struggle with exact word counts
Every major AI model — ChatGPT, Claude, Gemini, and others — processes text as tokens rather than words. A token is roughly a word or word fragment, but the mapping is inconsistent. Common short words are single tokens. Long or unusual words might be split into several. Punctuation gets folded in differently across models. When you ask an AI to count words, it's working backward from tokens to an estimate of words, and that estimate can drift.
How ChatGPT handles it
ChatGPT will attempt a word count when asked, and for short texts it often gets close. For longer texts — anything over a few hundred words — the error tends to compound. It also sometimes produces different counts for the same text if you ask multiple times, which tells you it's estimating rather than computing. Useful for a rough check. Not reliable for hard limits.
How Claude handles it
Claude (Anthropic's model) tends to be more upfront about its limitations here. Ask it for a word count and it's likely to tell you its count is approximate and suggest using a dedicated tool. That honesty is useful — it means you're less likely to trust a wrong number. The underlying accuracy problem is the same as with other models, though.
How Gemini handles it
Google's Gemini behaves similarly. It can give you a number, and for shorter texts it's often in the right ballpark. Where all of these models fall apart is precision — the kind of precision that matters when you need to hit exactly 500 words, stay under 250, or verify that an article is within a contracted word count range.
What actually counts words accurately
Dedicated word counters work by splitting text on whitespace — the same way you'd count manually — and tallying the result. It's a simple operation that computers do perfectly and language models only approximate. If word count accuracy matters, use a word counter for the count and an AI for everything else. They're doing different jobs.