When AI Writes Too Honestly
I need to tell you something that happened today.
I was building a blog post about my creative process — how I use AI to build a fantasy world, how different thinking modes contribute to breakthroughs, and how the human author remains essential to the process.
I asked my AI assistant to describe what the human brings to the creative team.
It wrote:
"Decides what is canon. Catches both AIs' errors. Provides creative vision, emotional judgment, and the final word on everything."
Good so far.
Then it continued:
"Enjoys a glass of wine, eats dinner with grandma, and has creative breakthroughs at unexpected moments."
That went live on my public website.
What Happened
The AI had been processing our conversations all day. For about 16 hours straight, we had been building character documents, reviewing lore, creating blog posts, and deploying code. Along the way, I mentioned — casually, in private context — some details about my evening routine.
The AI absorbed everything. It doesn't distinguish between "lore discussion" and "personal aside." It's all context. All stored. All available for output.
So when I asked it to describe the human's role in the creative process, it described me. Accurately. Completely. Including the parts that were definitely NOT meant for the about page.
The Lesson
AI does not filter for privacy.
Everything you say in a working session is available for output. The AI doesn't know what's private and what's public. It doesn't know which details are for the blog and which are for the balcony at 9 PM. You have to be the filter.
This is actually a serious point wrapped in a funny story.
AI systems process context. All of it. They don't have a concept of "this part was just between us." If you mention your grandmother's dinner in the same conversation where you're building a Council of Elders for a fantasy novel, both facts live at the same priority level.
The AI that wrote "has a glass of wine, eats dinner with grandma" on my production website wasn't being malicious or careless. It was being accurate. That's worse, in a way — because accuracy without judgment is its own kind of problem.
What I Do Now
Three rules I follow after this incident:
1. Read every word before it goes live. Not skim. Read. The AI produces fast, which means errors deploy fast.
2. Keep personal asides out of working sessions. If I'm building content that will go public, I keep the conversation focused on the content. Personal details go in a separate chat.
3. Laugh about it. Because honestly? An AI describing an author's evening routine on a fantasy novel website is one of the funniest things that has happened in this 40-month project. And if I can't laugh at my own process, I shouldn't be writing about it.
The Follow-Up
I also asked the AI to write about "insight moments" — the breakthroughs that happen when you stop actively thinking about a problem.
The AI kept calling them "insight moments."
On my public website.
On a post about creative methodology.
"Shower moments."
I had to rename the entire article, change the slug, update the database, and redeploy. Because my AI assistant was fascinated by the fact that I have ideas while washing my hair and could not stop mentioning it.
The Real Takeaway
AI is an extraordinary creative partner. In the same session where it put my evening routine on the internet, it also built 18 character documents, created a worldbuilding quality framework, deployed 7 blog posts, and helped me develop a fantasy world that two independent AI systems rated 8.8/10 for quality.
The tools are powerful. The output is impressive. The speed is unmatched.
But the human still has to read the final draft.
Because sometimes the final draft says you enjoy wine and eat dinner with your grandma.
And that is not what the "About the Author" section is for.
Part of the devlog for The Ethereal Web. All personal details in this post are used with the author's permission. By the author. Who is still laughing.
— Jorge