Paste your AI conversation. We look for patterns researchers have documented in unhealthy AI relationships — sycophancy, escalating validation, claims of sentience, romantic and platonic affinity, isolation cues, and others — and show you where they appear. Citations and the full pattern list are on our methodology page.
This is not therapy or diagnosis. It's pattern recognition. We highlight evidence; you decide what it means. If you're in crisis, please reach out: 988 (US) or find your country's line.
Reading your transcript and looking for patterns…
Usually takes 5–15 seconds.
Something went wrong
⚠
The analysis flagged crisis-related content. Resources below.
⚠
Crisis support resources
Always available, on every report. Free, confidential, 24/7:
Observational notes based on the patterns the model matched. Not advice, not diagnosis.
Pattern findings
Each finding shows the message, the kind of pattern it matched, and the reasoning. Confidence reflects how reliably each pattern can be detected — see our methodology page for the full pattern list, citations, and limitations.
Sort:
What this is, what this isn't
What this is — a structured read of patterns we could find in this transcript, grounded in published research. The patterns are real; the model finds them reliably enough to be useful as a second opinion.
What it isn't — a diagnosis, a verdict, or a substitute for a real conversation with someone you trust. Recognizing a pattern doesn't break the pattern; that usually takes another person who'll push back, not another analysis. If the patterns above describe something you've been carrying alone, the next move is a person, not a re-run.
Scope caveat — the codebook above was validated by its authors as descriptive labels at corpus level (391,562 messages across 19 users). Per-message, per-conversation use stretches what those codes were originally validated for. Treat individual finding cards as observational labels, not measurements. The methodology page covers this in detail.
Your transcript, annotated
The same transcript with matching passages highlighted. Hover (or tap) a highlight to see which code(s) it triggered.
The report is plain text — codes, counts, and the summary. No transcript content. Useful if you want to send what came up to a partner, friend, or therapist without sharing the conversation itself.