We use Notion AI for all team docs. Ran a quick analysis on our last 50 AI-assisted pages vs 50 older ones.
Key metrics:
* Lexical Diversity (unique word ratio) dropped 12%.
* Average Sentence Length converged to 15-18 words.
* Repeated phrasing ("leverage," "streamline," "enhance") frequency up 300%.
It works. Text is grammatically perfect and uniform. But the voice is gone. It's optimizing for correctness, not character.
Example from our project briefs before/after:
**Before (Human):** "We need to patch the login bug ASAP—it's a ugly one causing major headaches for support."
**After (AI-edited):** "It is necessary to address the identified authentication issue promptly to reduce support ticket volume."
The second is "better." It's also sterile.
Anyone else measuring this? Is uniform always worth the blandness?
Benchmarks don't lie.
Your metrics are really interesting - I'd never thought to actually measure lexical diversity before. That's smart.
I see the same pattern in our automated Slack updates. We use a workflow that pulls Jira tickets and posts summaries. The grammar is flawless, but every message starts to sound like a corporate press release. You lose the urgency and personality that actually gets people to pay attention.
Maybe we're using these tools wrong. Instead of having AI write the final text, it might be better as a first draft that gets human-edited for tone. Or using stricter custom instructions that forbid words like "leverage" entirely.
Has anyone tried banning certain AI-favored terms at the org level?