How Reloveo analyses your chats

Before any AI is involved, we compute hard facts directly from your chat — the kind of numbers an AI would otherwise have to estimate. For every analysis we measure:

• How many messages each person sent (and the share of the total).
• Average message length per person, in characters.
• Question-asking rate — how often each person asks something.
• Conversation initiations — who restarts the chat after a long gap.
• Median reply time for each person.
• Emoji use and the overall span of the conversation in days.

These numbers go into the analysis as facts, and the “effort share” percentage you see on your report is calculated from them — not estimated by an AI. Same chat, same number, every time.

Where the AI does interpret, we tell it to draw on specific, well-known frameworks — by name — so the reads are research-informed rather than vibes.

The Gottman Institute

Decades of research on what predicts whether relationships thrive. We watch for:

• The Four Horsemen— criticism, contempt, defensiveness, and stonewalling — patterns Gottman's research ties to relationship erosion.
• The 5:1 ratio — stable, satisfied couples show roughly five positive interactions for every negative one. We use this when assessing tone balance and trajectory.
• Bids for connection — small attempts to connect (a question, an observation, a check-in). We note whether each person turns toward, away, or against those bids.
• Repair attempts — small efforts to de-escalate during friction (humour, apology, softening). Their presence is a strong positive signal.

Attachment theory

We use Bartholomew's four-category model — the standard adult-attachment framework in modern research:

• Secure — comfortable with both closeness and autonomy.
• Anxious-preoccupied — high need for closeness, sensitive to perceived distance.
• Dismissive-avoidant — values independence, deactivates emotion under stress.
• Fearful-avoidant — wants closeness and fears it; visibly mixed signals.

Crucially, we frame these as “patterns consistent with…” — never as a diagnosis. Texting is a thin slice of a person.

Other lenses we use

• Demand-withdraw pattern (Christensen) — one person pushes for engagement while the other withdraws. Naming this loop out loud is often the first step out of it.
• Self-disclosure reciprocity(Altman & Taylor) — healthy connection deepens through mutual, gradual openness. Imbalance here is a useful signal.

Every analysis covers these:

1. Connection strength score — a single 0–100 read of the overall dynamic.
2. Communication style — how each person communicates, with quoted evidence.
3. Emotional investment balance — driven by the measured effort share.
4. Response patterns — initiation, reply times, conversation flow.
5. Attachment indicators — likely attachment patterns for each person.
6. Tone trajectory — how the emotional tone has moved over time.
7. Three key observations — specific patterns you might not have noticed.
8. One honest recommendation — a single, specific next step.

1. You paste a chat, upload a WhatsApp export, or attach screenshots.
2. We parse the messages, strip system noise (deleted messages, media markers), and anonymize names — replacing them with “Person A” and “Person B” before anything else happens.
3. We compute the hard metrics directly from the structured messages.
4. We send the anonymized transcript and the computed metrics to our AI model (Gemini 3.5 Flash, with 2.5 Flash as automatic fallback) with framework-grounded instructions.
5. We override any AI-estimated percentages with the measured ones, so the numbers in your report are real.
6. You see a free preview, with crisis information always unlocked. The full report is paywalled at a launch price.
7. After unlock, you can download a branded PDF and we email a copy to you.

• Names are anonymized before the AI ever sees the conversation.
• Phone numbers and emails inside messages are scrubbed.
• We do not store your conversation on our servers. The analysis is held encrypted in your own browser until you unlock it.
• Reloveo is built to be aware of India's DPDP Act and the GDPR.

For the full detail, read our Privacy Policy.

We try to be honest about limits, because a tool that pretends to be more than it is, isn't worth using.

• This is informational analysis — not medical or psychological diagnosis.
• A single conversation is a slice, not the whole story. Context outside the chat matters too.
• AI analysis has real limitations and can miss tone, sarcasm, or shared context.
• If a conversation shows signs of crisis or abuse, we flag it and surface professional helplines — not a verdict.

Reloveo is India-first, and the model is explicitly told to read with South Asian context in mind:

• Indirectness can be a politeness norm, not avoidance.
• Hindi-English code-switching is normal expressive behaviour, not confusion.
• Family considerations and shared context often shape what's said and unsaid.
• Early-relationship formality doesn't always indicate emotional distance.