Why we are publishing our prompt approach

If you have ever pasted a career question into ChatGPT and asked it to "read tarot," you already know the gap between a one-line prompt and a product that feels coherent session after session. TarotCareer is not magic software — it is prompt engineering, card data, and UX wrapped around the same large language models many people already use. We think that deserves daylight, not a mystique layer.

This post explains how our built-in AI prompts work at a high level: what the system rules enforce, how each of the 13 career scenarios gets its own brief, and why topic tags, optional zodiac flavor, and reshuffle variation prompts exist. We are not dumping proprietary source code; we are describing behavior so you can judge whether the labor we front-load is worth your time compared to DIY.

Tarot is entertainment and reflective practice. Nothing here predicts hiring outcomes, replaces therapy, or fixes your fortune. If you are in crisis, reach qualified professionals first. If you are curious about how a career-native tarot product keeps reports structured without sounding like a horoscope spam bot, read on — and tell us where we are wrong at famanoder.eryue@gmail.com.

We linked this explainer to the career direction scenario because that is where many first-time users land. The same prompt stack powers every scenario; only the scenario brief and spread framing change.

Layer 1: EN_BASE_RULES — the system contract

Every TarotCareer reading starts from a shared system ruleset we call EN_BASE_RULES in our codebase (with a Chinese mirror for zh-CN locale users). The rules tell the model who it is: a tarot assistant for work and career mindset — job search, burnout, pivots — not a human psychic and not an astrologer making birth-chart claims.

Two values repeat in every scenario. **Clarity** means helping you build a coherent frame: patterns and inner tendencies, not prophecy. **Emotional support** means acknowledging anxiety, excitement, or doubt before advice lands — warm tone, no cold lecturing, no doom spirals. The model must never promise guaranteed success or failure, and it must avoid fortune-fixing language like rituals that "change fate."

Output shape is non-negotiable. Reports include **Overview**, **Spread reading** (one bullet per card with position and orientation), **Deep dive**, **Action steps** (three numbered items with time scale), **Four-week rhythm**, and **Mindset note** reminding you that you stay the decision-maker. That structure is why a TarotCareer report feels like a document rather than a chatty paragraph — and why critics sometimes say it still hides the full prompt chain behind the UI.

Supporters argue this system layer saves hours of re-prompting in ChatGPT. Skeptics counter that without seeing every token, you are trusting a black box with the same opacity as any other AI product. Both can be true: structure reduces variance; opacity remains unless we keep publishing like this.

Layer 2: thirteen per-scenario prompts

Generic tarot prompts fail career users because "tell me about my future" collapses into vague encouragement. TarotCareer ships a dedicated brief for each of its 13 career scenarios — job-change, interview, career-direction, burnout-quit, multi-offer, layoff-warning, mid-career-transition, and others — each with its own spread logic and tone guardrails.

Examples: the **job-change** prompt stresses stay-vs-leave trade-offs without declaring a winner. **Interview** emphasizes preparation and performance over luck. **Burnout-quit** warns against pushing quit without runway. **Multi-offer** compares options structurally without "you must pick X." **Layoff-warning** focuses on practical prep, not severance prophecy. These briefs sit on top of EN_BASE_RULES so the model cannot drift into unrelated life coaching.

When you open job change versus burnout and quit, you are not just changing labels — you are changing the scenario question, spread positions, and the instructions the model receives before your cards are interpreted. That is the main product moat versus a single ChatGPT thread where you must re-explain context every time.

Fans say scenario prompts feel like talking to someone who already knows HR seasons and offer timelines. Critics say thirteen strings could still be edited in an afternoon — fair challenge. Our bet is that maintaining them alongside card copy, topic tags, locale mirrors, and reshuffle logic is ongoing labor, not a one-time paste.

Layer 3: topic tags, zodiac flavor, and reshuffle variation

After you pick a scenario, you choose a **topic tag** — a short phrase like "offer comparison" or "imposter syndrome in interviews." That tag flows into the user prompt so the report anchors on your noun, not a generic spread essay. Topic tags are one reason two people in the same scenario can receive meaningfully different reports without changing the underlying cards.

Optional **zodiac** input is explicitly style-only: it may tint metaphors or tone, but prompts forbid treating it as astrological proof or hire-date prediction. We label it as flavor because many users enjoy symbolic language; skeptics rightly ask why it exists at all. Answer: it is optional, skippable, and constrained so it cannot override career facts.

**Reshuffle variation** prompts kick in when you draw again within the same scenario. Core advice may align — if three cards keep pointing toward rest, rest stays thematic — but wording, examples, and card ties must change so you are not reading copy-paste déjà vu. That balances consistency with freshness and reduces the "AI said the same thing twice so it must be fate" trap.

Chinese users on zh-CN locale get **mirror prompts** — not machine translation of English strings, but parallel rules and scenario briefs tuned for Mandarin career context. We treat locale as a first-class prompt fork because job-market vocabulary and emotional register differ; a literal translate would sound stilted and miss nuance.

What supporters say

Supporters — including career coaches who recommend TarotCareer as a journaling warm-up — say the built-in prompt stack removes friction. You do not spend twenty minutes engineering a ChatGPT system message, fighting format drift, or re-uploading spread photos. You pick a scenario, tag a topic, shuffle, and receive the six report sections every time.

They also like that guardrails are opinionated in the right direction: no guaranteed outcomes, no ritual superstition, explicit mindset note that you decide. For people who want symbolic language without woo escalation, that boundary matters. Several compare it to a structured reflection worksheet that happens to speak in tarot vocabulary.

Power users note that browsing examples before a draw sets expectations — you see how Deep dive links cards to workplace patterns, not lottery numbers. Combined with open tarot meanings for card study, the product spans instant AI interpretation and self-education.

Supporters acknowledge limits: it is still a language model, still entertainment, still not your union rep or therapist. They just prefer paying for maintained prompts over maintaining their own.

What critics say

Critics argue that publishing a blog post about prompts is not the same as open-sourcing them. Without the exact strings, token order, and model version, users cannot audit bias, test reproducibility, or verify that zodiac flavor stays truly inert. "Trust us, we have rules" sounds like every other AI app marketing deck.

Some skeptics say DIY ChatGPT with a good personal prompt beats a walled garden — especially if you already know tarot spreads and want full control over temperature, model choice, and data retention. They worry TarotCareer adds mystique packaging to what is fundamentally autocomplete with tarot nouns.

Others question whether scenario prompts can handle edge cases — immigration status, disability accommodation, toxic leadership with legal risk — without overstepping into HR or legal advice. EN_BASE_RULES forbid that, but models drift; critics want clearer failure modes when users bring high-stakes questions.

We take that seriously. Email famanoder.eryue@gmail.com with prompt gaps you have seen — hallucinated policy advice, tone-deaf cheerleading, repetitive reshuffles. Critical feedback improves the stack faster than fan mail.

TarotCareer vs DIY ChatGPT: honest comparison

Neither option is "more true." Tarot does not become scientific because it runs inside a branded product, and DIY is not automatically safer because you typed the prompt yourself. Compare on workflow, structure, and how much maintenance you want to own.

If you enjoy prompt tinkering, DIY may win. If you want career-native spreads, topic tags, locale mirrors, and report sections without weekly prompt edits, TarotCareer wins. Many users do both: study cards on our meaning pages, run a structured session here, export a share image, then journal offline.

Try one career direction reading as a guest and compare the output to your own ChatGPT thread on the same question. Judge structure, tone, and whether the scenario brief saved you time — not whether either tool predicted the future.

What we will and will not claim

We will claim that TarotCareer invests in layered prompts so career readings stay structured, emotionally grounded, and scenario-aware. We will claim that publishing this overview is an invitation to scrutiny, not a flex.

We will not claim our prompts are unique magic, legally audited for every jurisdiction, or immune to model updates. When providers change base models, behavior shifts — we regression-test, but perfection is impossible.

We will not claim tarot replaces professional advice. Entertainment and reflection only. For compensation negotiation, discrimination, safety, or mental health crisis, use qualified humans.

Tell us what to improve: missing scenario nuance, weak reshuffle variation, zh-CN mirror gaps, or places where the black box still feels too black. Reach us at famanoder.eryue@gmail.com — supporters and skeptics equally welcome.