May 8, 2026

Spellcheck, but make it expensive

Human typing habits and token counts

Your typos might be quietly charging you extra — and the comments are spiraling

TLDR: The article says everyday typing habits like typos, shorthand, and extra fluff can quietly raise the cost of using AI chat tools. Commenters were split between laughing off the tiny savings, worrying about their typo habits, and accusing the post itself of sounding AI-written.

A delightfully petty internet panic is brewing: your sloppy typing may be costing money. The article’s big reveal is simple enough for anyone who’s ever fired off a rushed message — small habits like typos, extra punctuation, filler words, shorthand, random pasted links, and even weird spaces can make an AI chat count more chunks of text, which means a bigger bill. In one tiny example, a five-word prompt with two spelling mistakes counted as 13 billable chunks, while the corrected version dropped to 6. Ouch.

But the real show is in the comments, where the crowd instantly split into camps. One side basically said, absolutely not, I am not turning into a hall monitor for my own typing just to save a fraction of a cent. That was the loudest eye-roll of the thread: people are happy to trim automated systems and giant documents, but obsessing over every live chat typo? “making yourself crazy” was the vibe. The other side had a mini-existential crisis, admitting they typo constantly and had never considered their bad spelling a budget item. Still, even the penny-pinchers drew a line at manners, with one commenter declaring they’ll keep their “please” and “thanks” no matter what, because saving a few text-chunks isn’t worth becoming rude in real life.

And then came the spiciest detour: a commenter accused the post itself of sounding suspiciously AI-written, turning a nerdy pricing discussion into a classic internet side-quest about authenticity. So yes, the story started with spelling mistakes — and ended with people side-eyeing the author’s entire voice. Peak comment-section energy.

Key Points

  • The article says token billing can rise because human typing habits such as typos, shorthand, fillers, and stray whitespace change how text is split into tokens.
  • A short prompt example in the article dropped from 13 tokens to 6 after spelling was corrected, according to the authors measurements.
  • Misspellings like swapped, dropped, or doubled letters often produce more tokens than standard spellings under both OpenAI and Claude tokenizers.
  • Conversational padding and expressive punctuation or elongation, such as "...", "!!", "yesss", and "reeeally," add tokens while rarely improving the task.
  • Technical strings such as UUIDs, RFC 3339 timestamps, URLs, file paths, and boundary whitespace can quietly inflate token counts in normal work.

Hottest takes

"making yourself crazy to save a fraction of a cent" — Analemma_
"I can't write any prompt without typos, never thought it will increase my LLM cost" — hemc4
"heavily filtered through an LLM's voice, if not completely written by one" — rgovostes
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.