AI Relationship Conversation Analysis: What to Check Before You Pick a Tool
As a practitioner trained in cognitive-behavioral therapy, I regularly get asked how trustworthy AI-based conversation analyzers actually are. It's a young, fast-moving market — new tools show up every month — which makes it genuinely hard to tell a serious analysis apart from a novelty. This guide breaks down what to check, and walks through the method used on this platform.
Why "just ask ChatGPT" isn't the same thing
The easiest option is pasting your conversation straight into a general-purpose AI assistant. You can — but three limits apply:
- No dedicated privacy guarantee: a general assistant isn't built around handling sensitive relationship data as its core use case, and its retention policy covers everything it's used for, not a policy designed for this specific situation.
- No structured clinical framework by default: without precise instructions, the analysis stays impressionistic — a general tone summary, not a dimension-by-dimension read grounded in validated models.
- No structured output: a dedicated tool produces an organized report (profiles, dynamics, recommendations); a chat conversation produces a running reply you have to piece together yourself.
Checklist before you pick a tool
| Criterion | What to check | Why it matters |
|---|---|---|
| Privacy | Is your data deleted after analysis? Is it used to train a third-party model? | A relationship conversation is sensitive data; reuse should be explicitly ruled out |
| Named method | Does the tool cite identifiable theoretical frameworks? | A named model (Gottman, Bowlby, Young, NVC) can be checked and challenged; an unexplained "score" can't |
| Pricing transparency | Is the price clear before upload, or only revealed after? | A hidden price until payment is a classic red flag |
| Stated limits | Does the tool say what it can't do? | No AI replaces clinical judgment; a tool that never says so is overpromising |
The method behind ScanMyLove
The report is built on 14 published clinical frameworks, including:
- Gottman's Four Horsemen model (criticism, contempt, defensiveness, stonewalling)
- Bowlby and Ainsworth's attachment styles
- The Walker cycle of abuse
- Principles of nonviolent communication
- Beck's cognitive distortions
The limits, stated plainly
No automated analysis — this one included — replaces support from a professional when the situation calls for it. A report offers a structured lens on communication patterns; it doesn't diagnose anything and should never be treated as definitive proof in a conflict. If there's any doubt about real danger, the priority is always a qualified human contact, not a tool.
Frequently asked questions
Is my data kept after the analysis? No: uploaded conversations aren't stored beyond what's needed to generate the report, and aren't used to train any other model. Do I need an account to upload a conversation? No, upload is anonymous and requires no sign-up. Can the report be used as evidence in a dispute? No — it's a tool for personal understanding, not a document with evidentiary value. For professional use requiring a dated record of facts, a separate product exists (ScanMyJob, for workplace exchanges).Gildas Garrec is a practitioner trained in cognitive-behavioral therapy (CBT). The analysis method rests on published, referenced theoretical frameworks; its limitations are stated plainly in every report.
In brief: Dozens of tools now claim to analyze a relationship conversation with AI — some are general-purpose chat assistants repurposed for the job, others are dedicated products built on a published clinical method. Before pasting a private conversation into any of them, three criteria help you sort the real from the gimmick: privacy (are your messages stored, resold, or used to train another model?), method (does the analysis rest on named frameworks — Gottman, attachment theory, nonviolent communication — or does it just hand back a generic summary?), and specialization (does a tool built specifically for relationship texts outperform a general chatbot asked the same question?). ScanMyLove, built by a CBT-trained practitioner, runs on 14 published clinical frameworks and doesn't retain any conversation after analysis.
Join our exchange community
Join the Community (waiting list)Would it help to talk this through?
Gildas Garrec, CBT practitioner — sessions by video call (90 € / 75 min), or in person in Nantes, France.
Book a video session (page in French) →Related articles
Zombieing: When Your Ghoster Returns & What to Do
Understand zombieing, a post-ghosting phenomenon in modern dating. Learn effective strategies to respond when your ghoster returns and protect your emotional health.
Orbiting: Staying in Your Orbit Without Ever Texting (Likes, Story Views)
An ex who no longer texts you but likes your posts and watches all your stories: orbiting. This presence without contact has a logic the timeline of interactions reveals.
How ScanMyLove Works: A Guided Tour of the Tool
Prepare a conversation, understand how the steps depend on your files, compare reports and review the findings: a practical guide to ScanMyLove before paying.
Henry Hill: Why the Gangster Life Is So Seductive — The Psychology of Goodfellas
Psychological analysis of Henry Hill (Goodfellas): seduction of the criminal world, need for belonging, adrenaline dependence, and abandonment trauma.
