AI Disclosure
Effective Date: October 7, 2026
1. Readings Are Written by AI Models
CelestKin uses artificial intelligence to write personalized astrological readings. The readings are written by large language models that we call through Amazon Bedrock, the AI model service of Amazon Web Services (AWS). No human astrologer reviews or edits individual readings.
1a. Which Models Receive Your Data
As of the effective date above, our code can send your data to the following models. All of them run inside Amazon Bedrock.
• Anthropic Claude (Claude Sonnet 4.6, Claude Opus 4.6 and Claude Haiku 4.5). Claude writes most kinds of reading, the replies in chat and dream readings, and it reads the photos you submit for palm and body readings, the aura scan and the Feng Shui room scan. In chat, Claude also keeps short notes on what you mention (such as topics, people and life events) and summarises the earlier part of a long conversation, so that later replies can draw on them. Claude Haiku 4.5 also writes the chat suggestions described below in Japanese and Korean, and steps in when a reading, digest, briefing or preview call to Nova fails. The name-suggestion tool falls back to the earlier Claude Sonnet 4.5 or Claude Sonnet 4 if Sonnet 4.6 is unavailable.
• Meta Llama 4 Maverick. Llama writes the follow-up questions suggested in chat, working from your latest question and the reply to it, in every app language except Japanese and Korean. Our backend also has settings that, when switched on, have Llama write readings and chat replies in those same languages.
• Amazon Nova Micro and Amazon Nova Lite. Before a chart or compatibility reading is written, Nova reads the question you typed for it. It can pass the question on, turn it away, or reword a question it judges off-topic or inappropriate into a question about your chart; the reading then answers the reworded question, not the one you typed. Not every question you type passes through Nova: follow-up questions about a palm or body reading, and chat messages, are not sent to it. Nova also writes biorhythm readings; short answers about today, this week or this year; the daily and weekly digests; the translation of the daily briefing into your language; and short previews drawn from your chart.
Which model writes a given reading depends on the kind of reading, its language, your plan and our current settings. We change these settings from time to time without notice, so we do not promise that any particular reading was written by a particular model.
1b. Where AI Is — and Is Not — Used
Where AI is used: writing readings (chart, compatibility, dream, palm and body), chat replies, the daily and weekly digests and suggested questions; suggesting names; reading the photos you submit; checking, and sometimes rewording, the questions you type for a chart or compatibility reading; and drawing some of the Universe Feed pictures (see section 7).
Where AI is NOT used: chart computation (deterministic math against the Swiss Ephemeris and the Lahiri, KP and Four Pillars rule sets); the reply shown when a chat message matches our crisis keyword list, which is a fixed message with helpline numbers, chosen without calling any model; the filter that strikes medical terms from readings, which is a fixed word list; and our legal pages. These never call a model and cannot be changed by what a prompt says.
2. How It Works
When you request a reading:
1. Your birth data (name, date, time, location) is used to compute chart positions mathematically on our servers.
2. The computed chart data (planetary positions, house placements, aspects) is assembled into a structured prompt along with tradition-specific reference guides.
3. This prompt is sent through Amazon Bedrock to one of the models listed in section 1a, which writes a personalized interpretation.
4. The model's response is returned to you as your reading.
The model sees your data only for the request it is answering. It does not retain or learn from your individual data between sessions.
3. Data Sent to AI
The following data is included in AI prompts:
• Name (for personalization)
• Birth date, time, and location
• Age, worked out from the birth date
• For palm, body and dream readings: the gender on the profile, if one is set
• Computed planetary positions and chart data
• Selected question or reading type
• With many readings, what you told us during onboarding, if you answered: why you came, how you were feeling, and a preferred tradition
• In chat: your messages, the conversation so far, and any reading you carry into the chat
• For a dream reading, the dream you describe; for a photo reading, the photos you choose to submit
• For name suggestions: results worked out from the birth chart the names are for (such as its Moon sign), and the choices you make in the tool, such as boy, girl or neutral names, the traditions and cultures, and names to leave out
• Tradition-specific reference materials (these are static guides, not your data)
We do NOT send: your email address, account ID, payment information, or data from other users (except in compatibility readings, where both profiles' chart data is included). Earlier readings are sent in three cases: when you carry one into a chat; when you ask a follow-up question about a palm or body reading, which sends that first reading with your question; and when you ask for your Narrative, which is written from your five most recent saved readings.
4. AI Limitations
• AI-generated readings are non-deterministic — requesting the same reading twice may produce different text
• Different models can read the same chart differently, and the model that answers can change (see section 1a)
• The AI may produce inaccuracies, hallucinations, or culturally insensitive content
• Readings reflect pattern matching on training data, not genuine astrological expertise or supernatural insight
• The AI cannot verify or validate astrological claims — it presents traditional interpretive frameworks
• Some Claude readings use extended thinking (reasoning before answering), which does not guarantee correctness
5. No Training on Your Data, and What AWS Keeps
Amazon states that neither AWS nor the third-party model providers use inputs to or outputs from Amazon Bedrock to train Amazon Nova, Amazon Titan or any third-party model, and that your inputs and the models' outputs are not shared with the model providers: Anthropic and Meta do not receive your data. Amazon's documentation also says Bedrock does not store model inputs or outputs by default, and lists the exceptions it makes for abuse detection. As of the effective date above, those exceptions cover models we do not use, and images that its automated child-safety check flags. See Amazon Bedrock's FAQ, data protection and abuse detection pages. What CelestKin itself stores, such as your saved readings, is described in our Privacy Policy.
6. Human Oversight
Each day our team reads a small sample of questions and answers to check their quality (questions flagged as a crisis are never included). CelestKin's development team also periodically audits reading quality, reviews prompt engineering, and monitors for harmful or inappropriate outputs. If you receive a reading that contains harmful, offensive, or clearly erroneous content, please report it to us.
7. How the Blog Guides and the Universe Feed Are Made
The guides on the CelestKin blog are drafted with the help of an AI model (Claude, by Anthropic) and published by the CelestKin team. Every guide is published under the CelestKin name, because the same small team publishes all of them.
The Universe Feed is not written fresh each morning. It is assembled from a fixed library of short passages: one set for each day sign and number of the Maya Tzolk'in, each lunar mansion and phase group of the Vedic month, each Chinese day stem, each Moon sign and phase, and each day number. Those passages were also drafted with AI assistance. Which passages appear on a given date is decided by calculation, not by a model: the Moon's position, the Tzolk'in count and the day pillar are computed for that date by the same code the CelestKin app uses.
Some Universe Feed pictures are drawn by an AI image model, Stability AI's Stable Image Style Guide, which also runs inside Amazon Bedrock. When a post is published, the model is given a short scene description chosen from the post's topics and one of our own pictures to follow for style, and it draws that post's picture once. No reader data is sent to it. See Amazon's page on this model.
8. Contact
AI-related questions: [email protected]