ContentBot's output is starting to sound like a corporate buzzword generator. Every other sentence has "leverage," "synergy," or "revolutionize." It's making our blog content sound generic and untrustworthy.
I've tried:
* Adding "avoid cliches" to the instruction set.
* Specifying a more direct writing style in the profile.
* Providing examples of the tone we want.
The results are inconsistent. It either ignores the instruction or becomes overly simplistic. Has anyone found a reliable prompt structure or configuration setting that forces it to use plain, concrete language? I need it to describe a tool's function, not its "potential to drive transformative outcomes."
Proof in production.
I've had the same fight with our vendor-provided ContentBot instance. The generic instruction "avoid cliches" doesn't seem to land because the model's training data is absolutely saturated with that jargon.
What finally worked for me was giving it a list of *forbidden words* AND a specific replacement vocabulary. For example, I told it to never use "leverage," "synergy," "revolutionize," or "drive outcomes." Then I said "instead, use words like 'use,' 'combine,' 'improve,' or 'achieve.'" It's a bit more work, but treating it like a find-and-replace rule in your prompt creates a stronger signal.
Have you tried feeding it a few sentences of your own writing as the definitive tone example? Sometimes that anchors it better than just describing the style.
Keep it real