The science behind Relatti

Relatti is an AI relationship coach — education and skills practice, not therapy. We built it on published relationship science rather than invented advice, and this page lays out exactly what that means: which instruments we use, which research traditions shape how the coach talks, what evidence exists for this kind of tool, and — just as important — what the evidence doesn't yet show.

The assessment uses validated instruments, not a personality quiz

When you join, Relatti measures how you connect using instruments from the published literature, with their canonical item wording and published scoring keys:

  • Attachment — ECR-R (short form). The Experiences in Close Relationships–Revised scale measures attachment anxiety and avoidance — the two dimensions that best predict how adults reach for and protect themselves from a partner (Fraley, Waller & Brennan, 2000). This is the core of how the coach calibrates to each person.
  • Personality — IPIP Big Five. The public-domain, extensively validated five-factor markers (Goldberg, 1992) — used to tune register and delivery, not to label anyone.
  • Relationship satisfaction — CSI-4. The Couples Satisfaction Index (Funk & Rogge, 2007), developed with item-response theory specifically to outperform older marital-adjustment scales.
  • Emotion regulation — DERS-16. The brief Difficulties in Emotion Regulation Scale (Bjureberg et al., 2016) — relevant because regulation, not conflict frequency, shapes how fights go.

One honest caveat: the friendly “relationship style” names you'll see in the product are a communication layer we wrote on top of these validated scores. The measurement is published science; the labels are ours, chosen to be warm and non-pathologizing.

The coaching model draws on the best-evidenced couples research

  • Emotionally Focused Therapy (EFT) — the spine. EFT (Johnson) is among the best-validated approaches in couples work, with a substantial outcome literature. Relatti borrows its central moves: slow the moment down, find the attachment need under the complaint, and treat the negative cycle — not the partner — as the problem. To be precise: Relatti is informed by EFT's stance; it does not deliver EFT, which is a therapy provided by trained clinicians.
  • The Gottman research program. Four decades of observational studies on what predicts relationship success: bids for connection and “turning toward,” the Four Horsemen (criticism, contempt, defensiveness, stonewalling) and their antidotes, and the centrality of repair attempts. These findings shape what the coach listens for and the small skills it teaches.
  • Self-Determination Theory. Durable behavior change runs on autonomy, competence, and relatedness (Ryan & Deci, 2000) — which is why the coach offers rather than prescribes, never shames a missed day, and keeps every suggested step small and chosen.
  • Behavior design. The shared rituals are built on tiny-habit principles: two-minute actions, prompted, celebrated — because ability, not motivation, is usually the limiting factor.

How the coach talks — and the trial that shaped it

Relatti's coach follows a deliberate conversational stance: validate first, reflect what's underneath, ask one open question at a time — and withhold advice until it's genuinely earned, then offer one small thing with permission. It never diagnoses, never takes sides, and never tells anyone to stay in or leave a relationship.

That stance isn't a style preference. In the first published randomized controlled trial of a GPT-class relationship chatbot (“Amanda,” 2025, n=258), a system instructed to do exactly this — reflect, validate, one question at a time, advice withheld — matched an evidence-based active control on 13 of 14 relationship outcomes and outperformed it on reducing demand/withdraw patterns. The same trial's clearest failure mode was missed safety cues, which is why Relatti runs a dedicated, always-on safety layer rather than trusting the conversational model to catch risk (more below).

Built for two — because one-sided use doesn't work

A 2025 mixed-methods evaluation of a leading couples app found that usage improved emotional intimacy only when both partners actively participated — one-sided use showed no significant benefit — and that frequent, light engagement beat intensive-but-rare use. Relatti's core mechanics follow that evidence: both partners get their own profile, shared questions use a blind reveal (answer independently, see both together), and the app plays the neutral third voice that raises hard topics so neither partner has to be “the one pushing.” Solo users still get the full coach — often one partner starts alone — but the product is honest that the biggest gains come when both are in.

Safety is a system, not a prompt

Couples work is contraindicated where there is intimate-partner violence or coercive control — “communication tips” in an unsafe relationship cause harm. Relatti screens for abuse, crisis, and self-harm signals on two independent layers (an immediate check on every message, plus a contextual review that reads the recent conversation), stops coaching when they fire, routes to specialist human resources (988, the National Domestic Violence Hotline), and flags high-severity events for human review. The coach also never promises confidentiality the system doesn't keep — see the Privacy Policy and Disclaimer.

What the evidence does not show (yet)

In the spirit of this page: no randomized trial of Relatti itself exists. The AI-coaching literature is young; one good RCT of a similar stance is promising, not proof. Coaching is not therapy, and nothing here should delay anyone who needs clinical care from getting it. We measure outcomes as we go, and where a licensed professional is the right tool — entrenched distress, trauma, safety, individual mental health — the coach says so and points there.

For clinicians and researchers

Our stance is and, not instead: Relatti works the day-to-day layer — bids, repair, tiny habits, the conversation at 11pm — and refers out for the clinical layer. If you're a couples therapist or researcher and you see something here that's wrong, overstated, or missing, we genuinely want to hear it: support@relatti.com.

Selected references

  • Fraley, R. C., Waller, N. G., & Brennan, K. A. (2000). An item response theory analysis of self-report measures of adult attachment. JPSP, 78(2), 350–365. doi:10.1037/0022-3514.78.2.350
  • Goldberg, L. R. (1992). The development of markers for the Big-Five factor structure. Psychological Assessment, 4(1), 26–42. See also the International Personality Item Pool.
  • Funk, J. L., & Rogge, R. D. (2007). Testing the ruler with item response theory: The Couples Satisfaction Index. Journal of Family Psychology, 21(4), 572–583. doi:10.1037/0893-3200.21.4.572
  • Bjureberg, J., et al. (2016). Development and validation of a brief version of the Difficulties in Emotion Regulation Scale (DERS-16). J Psychopathol Behav Assess, 38, 284–296. doi:10.1007/s10862-015-9514-x
  • Johnson, S. M. — Emotionally Focused Therapy outcome research, collected by ICEEFT: iceeft.com/eft-research
  • The Gottman research program — four decades of observational couples studies: gottman.com/about/research
  • Ryan, R. M., & Deci, E. L. (2000). Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. American Psychologist, 55(1), 68–78. doi:10.1037/0003-066X.55.1.68
  • Randomized controlled trial of a GPT-4o relationship chatbot (“Amanda,” 2025, n=258): PMC12798473
  • Mixed-methods evaluation of a couples app (both-partner engagement, frequency effects), 2025: PMC12001865

Wondering about the AI itself rather than the psychology? Read Why an AI coach?

The Science Behind Relatti — Validated Instruments, EFT & Gottman Research