Why every editor we work with hates the word 'leverage'
A short tour through the words large language models reach for first, and what they tell you about whose voice is leaking through.
Leverage. Robust. Streamline. Empower. Unlock. Seamless. Cutting-edge. You can spot a draft written by a frontier model the way you can spot a draft written by a tired MBA student — the vocabulary gives it away inside two sentences.
Why this happens
The training corpus is full of marketing copy, and marketing copy uses these words to mean almost nothing while sounding like it means something specific. The model has learned that they are safe — they will not be wrong, because they refuse to be precise. Safety, for a model, looks a lot like vagueness.
How we fix it
A short banned-words list in the system prompt does about half of the work. A small fine-tune on the customer's own published writing does the other half. The combined effect is uncanny: the model stops sounding like LinkedIn and starts sounding like the person whose name is on the byline.