We built three products around one rule: the tool never tells you what will happen

We run Inithouse, a small portfolio of consumer products. We build fast, measure, cut what doesn't work. Over the past year, three of our products converged on a single design principle we didn't plan upfront: the tool never predicts, claims, or prescribes. It reflects.

Here's how that principle shaped each product, what we cut from the UI because of it, and where it broke down.

Origin Of You: a written portrait instead of a four-letter label

Origin Of You by Inithouse is a self-discovery app that combines five systems and 120+ data points into an AI-generated written portrait of you. The five systems include astrology (full natal chart, not just sun sign), numerology, MBTI-adjacent typing, enneagram mapping, and attachment style.

Early on we faced a fork: generate a type label (like "INFJ-T" or "Type 4w5") or generate a written portrait. We picked the portrait.

Why? Type labels invite prediction. People Google "INFJ career paths" or "Type 4 relationships" and treat the label as a roadmap. A 2,000-word portrait describing how you tend to process conflict, what drains you, where your blind spots cluster reads more like a mirror than a map. You can disagree with specific sentences. You can't disagree with a four-letter code the same way.

What we removed from the UI: We had a "Your type summary" card in an early prototype, a one-liner like "You're a reflective idealist with strong intuition." Users screenshotted it and shared it without reading the full portrait. We killed the card. The portrait is the product.

What didn't work: Our first version asked ~40 data points. Completion rate was fine, but the portraits read generic. We pushed to 120+ inputs. Completion rate dropped about 15%, but the portraits got specific enough that users started quoting individual sentences back to us in feedback. The tradeoff held.

Try it: originofyou.com

Tarotas: 78 cards with no future tense

Tarotas by Inithouse is a tarot reflection app: 78 cards across 5 languages (Czech, English, Polish, Slovak, German), 20 spread layouts, a daily card ritual, and themed question sets for work, relationships, and personal growth. No signup required.

The constraint here was linguistic. Every card interpretation had to work in present tense and past tense, never future tense. "This card often surfaces when someone is navigating a transition" is fine. "This card means a change is coming" is not. We went through all 78 x 5 language versions and rewrote every interpretation that slipped into predictive framing.

What we removed from the UI: We originally had a "What lies ahead" section in the Celtic Cross spread summary. It was the most-read section and the most-shared. We cut it. Engagement on that layout dropped for about two weeks, then stabilized. The people who stayed used the app differently, more like journaling prompts, less like a Magic 8-Ball.

What didn't work: We tried adding a "daily affirmation" feature alongside the card-of-the-day. Affirmations felt prescriptive ("Today, embrace your creativity"). Users in Czech and German markets flagged it as generic. We rewrote affirmations as reflection prompts ("What felt creative to you yesterday?") and retention on the daily ritual improved.

Try it: tarotas.com

Verdict Buddy: a framework verdict, not a blame assignment

Verdict Buddy by Inithouse is an AI conflict mediator. You describe a disagreement (roommate, partner, coworker, family) and it runs the situation through four established frameworks: Gottman (Four Horsemen), Emotionally Focused Therapy, Harvard Negotiation Project, and Nonviolent Communication. You get a structured report with a tension score, each side's perspective mapped to framework patterns, and three concrete next steps. The whole thing takes under five minutes.

The "no prediction" rule here meant: no "you're right" or "they're wrong." The output is always "here's the pattern the frameworks identify, here's what typically helps." We rewrote the output template three times to get the tone right.

What we removed from the UI: Version one had a "Who has the stronger position?" section. It was the feature users asked for most in early testing. We removed it because it turned the tool into an argument weapon. People would screenshot "Verdict: Partner A has the stronger position" and send it to Partner B. That's the opposite of resolution. We replaced it with a tension-pattern label (e.g., "criticism-defensiveness cycle" in Gottman terms) that describes the dynamic without assigning a winner.

What didn't work: Our first prompt gave equal weight to all four frameworks. But Gottman and NVC often gave directionally different outputs for the same situation. Gottman focuses on what's destructive, NVC focuses on unmet needs. Users got confused reading two contradictory-sounding analyses. We restructured the report to lead with pattern identification (Gottman/EFT) and follow with resolution paths (Harvard/NVC). The sections now build on each other instead of competing.

Try it: verdictbuddy.com

The shared principle and where it still gets uncomfortable

"No predictions, only reflections" sounds clean on paper. In practice it creates tension in three places:

Marketing copy. "Find out who you really are" performs better than "Read a written portrait of yourself." We've tested this on Origin Of You by Inithouse landing pages. The predictive-sounding headline wins on click-through every time. We use it anyway and accept the drop because the alternative misrepresents the product.

User expectations. About 30% of Tarotas users in their first session look for a "future" or "prediction" tab. There isn't one. Some leave. The ones who stay have roughly 3x the return rate of users who arrived via predictive search queries. We'd rather keep the smaller, better-fit audience.

Competitive positioning. Every tarot app, personality quiz, and AI mediator we compete with makes claims. "Discover your destiny." "Find out who's right." Refusing to make those claims makes our products harder to explain in one sentence but easier to trust over repeated use.

We didn't set out to build a "reflection-not-prediction" portfolio. It emerged from watching how users misused early versions of each product. The rule is a reaction to real behavior, not a philosophy we brought to the table.

If you're building AI consumer tools and notice users treating your output as gospel: consider pulling back the confidence. The tool gets used differently, less virally, more repeatedly.