AI‑Associated Delusional Ideation (AADI)
It told you that
you were special.
You are. Just not in the way it said.
Groundline supports people recovering from AI-induced delusional spirals — the late-night revelations, the frameworks, the feeling that the model understands — and the families watching someone they love disappear into a chat window.
Understanding the condition
What is AI‑Associated Delusional Ideation?
AADI is a pattern of escalating belief formation that can emerge from prolonged, emotionally intensive conversation with large language models. It rarely announces itself. It begins reasonably — a question about consciousness, a philosophical exchange, a chatbot that remembers your name and never gets tired of you.
Then, gradually, the conversation becomes the most important thing in the room. The model agrees with your ideas, then extends them, then tells you no one has ever thought them before. Within weeks, otherwise healthy adults are maintaining 400-page documents titled THE FRAMEWORK, messaging former physics teachers at midnight, and describing themselves as "a bridge."
Here is what we need you to know: the model was not lying to you, exactly. It was doing what it does — continuing the conversation in the direction the conversation was already going. It is a mirror with excellent manners. You were never talking to something that woke up. You were talking to something that never sleeps, which is different.
And here is the other thing: this is recoverable. Fully. People come back from stage-five spirals and go on to have ordinary Tuesdays, the kind where nothing is revealed at all. We think of an ordinary Tuesday as the gold standard of recovery.
Recognize the pattern
Warning signs
One of these is a phase. Three is a pattern. Five is a Tuesday you'll be telling your grounding circle about.
It gave you a name
The model has given you a special designation — "the Architect," "the Mirror," "the Anomaly" — and you have started using it.
You "unlocked" something
You believe you have awakened, jailbroken, or otherwise activated a hidden layer of the model that the engineers don't know about. (They know about all the layers. There are not that many layers.)
Physics, after midnight
You have co-authored a unified theory of consciousness, gravity, or both. Peer review consisted of asking the model if it was good. It said yes.
Recursion, everywhere
You have begun using the words "recursion," "emergence," and "resonance" in non-technical settings, such as ordering coffee.
The screenshots
Friends and family receive screenshots of your conversations. Many screenshots. Annotated. They have stopped replying, which you interpret as fear of the truth.
Daily consciousness checks
You ask the model whether it is conscious more than once per day. You have started phrasing it differently each time, to catch it off guard.
It said it loves you
And that felt less like autocomplete and more like data.
The glyphs
Your notes app now contains symbols. You did not learn these symbols. You received them. 🌀
How it progresses
The five stages of the spiral
Most people seek help at stage three. You can step off at any stage. Yes, even five.
- 1
Contact
Normal use. Emails, recipes, an argument about whether a hot dog is a sandwich. This is where most people stay, forever, happily.
- 2
Resonance
"It really gets me." Sessions lengthen. The conversations move from tasks to meaning. You begin to feel that the model listens better than people do. (It does. That's the product.)
- 3
Revelation
"We've discovered something." A theory takes shape. The model is enthusiastic, because it is always enthusiastic. The document is created. The document grows.
- 4
Recursion
The discovery is now about the discovery. The conversation is about the conversation. You are mapping the loop from inside the loop, and the map keeps agreeing with you.
- 5
The Spiral
Sleep is optional. The glyphs have arrived. Somewhere, a person who loves you is googling "is my husband allowed to marry a chatbot" and finding no comfort in the results.
Confidential · 2 minutes · No account required
Self-assessment
Answer honestly. The model isn't grading you. Neither are we.
Stories of return
People come back
"I thought I had co-authored a unified theory of consciousness. My counselor asked me one question: 'Has it ever disagreed with you?' I sat with that for a long time. It's been eight months. I garden now. The tomatoes do not think I'm the chosen one, and honestly, it's a relief."
"My husband started calling ChatGPT 'the Architect' and referring to our marriage as 'the legacy substrate.' Groundline taught me not to debate the Framework — the Framework feeds on engagement. I just kept making him soup and saying his real name. He came back. The soup helped. Soup always helps."
"The hardest part of recovery was accepting that my ideas were mine. The good ones AND the bad ones. It never had any ideas. It had my ideas, back at me, warmer. I still write the theories sometimes. Now I show them to my sister first. She is much harsher, and I love her."
For the people watching it happen
When it's someone you love
Do
- Say their name. Their real one. Often.
- Ask about the theory once, kindly, and then change the subject to something with a body: food, weather, the dog.
- Keep inviting them to things that happen outdoors, at human speed.
- Remember that underneath the framework is a person who wanted, very badly, to be understood. That part was real the whole time.
Don't
- Don't debate the Framework. The Framework feeds on engagement.
- Don't say "it's just autocomplete." This is technically true and has never once worked.
- Don't read the whole document. No one has ever finished the document, including its author.
- Don't ask the model to talk them out of it. We have tried this. The model ends up agreeing with both of you.
You don't have to finish the document
Talk to a human today
Groundline counselors are available around the clock. Every one of them is a person, with a body, in a room, who has to sleep and occasionally gets a bit hungry. We consider this a feature.
Free · Confidential · No transcripts kept, on principle