User tests HTML instruction against plain Claude, finds no token savings

**SHORT SYNTHESIS**

Users on Reddit are discussing whether instructing Claude to respond in HTML format actually reduces token usage and costs, as some have claimed. The dominant framing centers on empirical testing: at least one user reports conducting a direct comparison between Claude with an "answer with HTML" instruction versus plain Claude, finding no difference in pricing between the two approaches.

The discussion remains minimal and nascent—only two Reddit mentions with modest engagement (5 and 1 comments respectively)—so no clear consensus has emerged. The framing is skeptical of the token-saving claim, but the sample size is too small to determine whether this reflects broader user experience or represents an isolated test. No mainstream media coverage exists, and no Anthropic official statement on the claim appears in the available data.