Editorial policy

Who is responsible

Love Oracle is a one-person independent project. It is written, built and published by askloveoracle, who runs the same handle on Instagram and YouTube, and publishes this project’s open dataset and MCP server on GitHub. There is no editorial team, no contributor network and no sponsored placements. One person decides what ships, and that person is answerable for everything on the site.

No professional credentials are claimed and none are implied. Nobody here is a licensed therapist, counsellor, doctor, lawyer or financial adviser, and no reader on this site is a human psychic. Tarot here is entertainment and a vocabulary for thinking, not divination and not advice.

What wrote the English

The English source text — the 78 card pages, the question pages, the zodiac and birth-chart pages, the articles and these policy pages — was drafted with a large language model, directed section by section and edited by a human before it went live. It is not scraped, spun, or compiled from other tarot sites, and the love meanings were written for this product rather than borrowed from an existing reference.

Two parts are deliberately mechanical rather than written. The 22 Major Arcana names follow the 1909 deck’s own tradition; the 56 minor cards have their names generated from rank × suit tables, so the Two of Cups and the Two of Wands cannot drift apart inside a language. Where a correspondence is disputed between tarot schools — the elemental attribution of the Major Arcana, for instance — the site says nothing rather than picking a side, and where it does follow a school it names it.

The other 21 languages are machine-translated

English is the source. Every other language on this site was produced by machine translation and has not been read by a native speaker. That is a real limitation and it is stated here, on the about page, and in the site’s machine-readable documentation, rather than left for a reader to discover.

Because of that, 7 languages are currently offered to search engines — English plus the 6 with an established readership. The rest remain fully readable and fully usable on the site, and are marked so that search engines do not index them. A language is put back into search when it has been proof-read, not before.

What is checked before publishing, and what is not

Machine translation fails in ways that reading the file will not reveal, so the checks are machine checks, and they run over all 22 languages on every change. They reject a document that:

• names the wrong card or the wrong zodiac sign — the single most common and most invisible failure;
• leaves English, or another language’s script, sitting inside the translated prose;
• collapses the paragraph structure of the source, or silently drops a section;
• loses a key, an interface placeholder or a number that the source stated;
• invents fields that the source never had, or overruns the length a search result can show.

What those checks cannot see is idiom, register and tone. A sentence can pass every one of them and still read like a translation. That is the gap human proof-reading closes, and until it is closed for a language, that language stays out of search.

Readings are generated, not written

A reading is produced live, for you, by a language model, from the cards your shuffle actually produced. There is no answer bank and no human in the loop: nobody writes, reviews or approves a reading before you see it. The model is told which cards came up, which way round they landed, in which positions, and what is physically drawn on those specific cards — that last part exists because without it a model will confidently describe a figure that is not on the card.

The five readers are writing styles, not people. They are prompt personas with different registers; the same spread reaches the same verdict through any of them. None of them is presented as a real psychic, and none of them has a biography, because inventing one would be a lie.

Readings are stored so you can return to them, and the operator can access stored readings for support and for investigating abuse — that is how the prompt problems listed above were found. Readings are never published, never indexed and never served to crawlers unless you deliberately share one with the share link. What happens to that data, and how to delete it, is in the privacy policy.

What the readers refuse

Some questions should not be answered by a card generator, and the refusal is built into the prompt rather than promised here. The readers will not give medical, pregnancy, financial or legal readings, will not predict death or a diagnosis, and will not state a fact about a third person — whether someone cheated, lied or is hiding something. They can read a pattern in your situation; they cannot deliver a verdict about somebody who is not in the room and cannot answer back.

A question that suggests someone is in danger gets a support response, not a spread.

Sources, and what we give back

The card artwork is the 1909 deck, which is in the public domain; the scans come from Wikimedia Commons. Birth-chart positions are computed from published ephemerides rather than looked up from a service, and place and time-zone data comes from GeoNames under CC BY 4.0. Where an article cites research it names the study and links it; where it compares this product with a competitor it names the competitor, links to it, and discloses the conflict of interest in the first line.

The keyword layer of the card data — all 78 cards, upright and reversed, in every language — is published as one open JSON file under CC BY 4.0, so that anyone can check what the site actually claims a card means.

Mistakes, and how they get fixed

Report anything wrong — a mistranslation, a card described incorrectly, a reader saying something it should have refused, a factual error in an article — to hello@loveoracle.app. A person reads it.

Corrections are made by fixing the source text and redeploying, which usually means the page is right within a day. A translated page that is wrong is deleted in that language and regenerated rather than patched by hand, because editing a machine-translated paragraph without reading the whole document tends to produce a second, quieter error. A correction that changes the meaning of an article is noted in the article itself.

What this site will not do

No countdown timers, no invented scarcity, no “three people are reading this now”. No fabricated reviews or testimonials. No reading that stops halfway to sell you the ending. No claim that a human psychic is on the other side. No affiliate links, and no page written to rank that a reader would not want to read. How the product actually makes money is on the pricing page, in numbers.