← sources

Has anyone figured out how to reduce the common "AI" speak from Claude?

source: reddit (reddit:ClaudeAI)published 2026-09-10 17:50ingested 2026-09-10 22:39

retained text

**TL:DR - I haven't and require continuous revision to make output sound 'normal'. Is there no solution to this?**

**Edit 1:** multiple folk suggested "ASD STE-100 for all communication", I will give this ago. Still waiting on a solution to assumption based responses that turn into questions/validation before outputting irrelevant text.

I've tried to set proper memory, project instructions, create skills, .md instructions etc.

Opus/Fable continuously defaults to overly verbose and the typical cliche style 'its not x, its y', 'Missing, and worth one line each' etc.

I even made a log that tracks each time I note this is happening and there now dozens of 'tickets' and zero ability to continuously adjust the language output.

**Has anyone found anything that works consistently? Do I need a 'check language'/'coms rules' line each time I want an output?**

**Bonus question:** Claude will often make a ton of assumptions and provide paragraphs of additional information that was not requested. I've attempted to skill this in as well and the same issue occurs. I need to basically hammer it in every conversation that 'instead of blocks of texts based on unstated assumptions, just ask me a list of questions or state assumptions for confirmation'.

**Has anyone found a decent solution to this or is it all brute force each time?**

document fetched 2026-09-11 00:23 · canonical https://reddit.com/r/ClaudeAI/comments/1wcqdk1/has_anyone_figured_out_how_to_reduce_the_common