Algorithmic divination and AI oracles

Algorithmic divination describes practices in which computational systems participate in selecting, arranging, recommending, or interpreting material that users understand as guidance, signs, prophecy, fortune-telling, or communication with a more-than-human source. AI oracle is a looser expression used for conversational systems that answer open-ended questions, systems explicitly incorporated into divinatory practice, and metaphorical descriptions of artificial intelligence as an apparently authoritative source of future knowledge.

The field includes several technically and socially different phenomena. An astrology program may calculate a chart without using artificial intelligence. A tarot application may draw cards through a pseudorandom process and retrieve fixed meanings. A recommendation platform may deliver content that a user interprets as destined for them. A practitioner may draw physical cards and ask a general-purpose language model to help interpret the spread. A chatbot may also be consulted directly as if it were an oracle, even when its developer does not market it for spiritual use.

The available research documents specific platforms, interpretive practices, interface effects, and oracle metaphors. It does not establish that algorithmic outputs predict events, communicate with supernatural agents, represent a single religion, or have a known population-wide prevalence.

Definition and category boundaries

Divination is broader than prediction. Historical and contemporary practices may seek diagnosis, orientation, symbolic interpretation, hidden causes, appropriate timing, or guidance under uncertainty. Digitization can automate parts of an established system without changing its claimed cosmology, while generative systems can introduce new language and apparent conversational agency.

Pre-AI digital-divination background

Computer-mediated divination predates generative AI. Early and later online services encoded astrological, numerological, calendrical, tarot, and fortune-telling rules; selected stored interpretations; or connected clients with human practitioners. Kuo's 2009 study of Chinese online fortune-telling services examined a commercial market through website analysis, interviews with site owners, and consumer surveys, demonstrating that digital delivery and paid remote consultation were established well before current language models.^[9]

Kim's study of South Korean eight-character fortune telling describes both face-to-face and online forms. In the online form, the diviner may no longer be necessary for the immediate reading: scripts convert birth information into formatted results, while users encounter the output as both information and an affective virtual experience.^[1] This is digital divination, but it is not automatically AI. The distinction matters because a deterministic rule engine, a random card selector, a recommendation system, and a generative model produce outputs through different mechanisms and create different forms of opacity.

Digitization also changes material and social conditions. A private application can make consultation immediate, repeatable, inexpensive, and detached from a recognized practitioner. At the same time, it may remove the interpersonal negotiation through which a human diviner clarifies a question, explains a tradition, refuses an inappropriate request, or assumes responsibility for a consequential interpretation.

Algorithmic and generative mechanisms

Three mechanisms recur across documented practices: selection, personalization, and generation.

• Selection chooses an item from a finite set. A digital tarot deck may simulate a shuffle; an oracle application may select a verse or symbol. The selection can be random or pseudorandom without being intelligent.
• Personalization ranks material according to observed behavior, profile data, inferred interests, network effects, and platform objectives. The system is usually optimizing an operational target such as engagement, not conducting divination.
• Generation composes new text, images, audio, or combinations of them from a prompt and context. A generated interpretation may be specific in language while remaining unverified, internally inconsistent, or strongly shaped by the information supplied by the user.

These mechanisms can be combined. A practitioner may physically draw cards through a traditional random process, photograph or describe them to a language model, provide personal context, and receive a generated narrative. In that case, the cards, practitioner, prompt, conversation history, system configuration, and model all participate in the result. The model is not independently discovering the cards' meaning; it is producing a plausible interpretation from the supplied symbols and context.

Generative output is also sensitive to wording and interaction history. NIST identifies confabulation, automation bias, anthropomorphism, emotional entanglement, and unjustified overreliance among the cross-sector risks of generative AI.^[8] In an oracle-like setting, fluent language can therefore increase perceived authority without increasing predictive validity.

Representative platforms and practices

User interpretation and ritual use

St. Lawrence describes WitchTok practitioners who use TikTok's recommendation system and generative filters as elements of technomancy. In the documented examples, users may treat an unexpectedly specific video, repeated symbol, generated image, or item appearing on a personalized feed as a response to intention. Some describe the algorithm as an entity, egregore, collective intelligence, or collaborator; others use it as one component alongside tarot, runes, deities, spirits, or established magical practice.^[2]

Chalfant analyzes a related but distinct pattern through New Thought and the law of attraction. TikTok's apparently mind-reading personalization can be interpreted through a religious vocabulary in which thoughts attract external events. The algorithm's invisibility and personalization make it compatible with narratives of manifestation, even though the platform's recommendation system is operating through data collection and ranking rather than metaphysical causation.^[3]

Prock and colleagues provide direct evidence about AI-assisted tarot through twelve semi-structured interviews with practitioners who already incorporated AI into personal divination. Participants used AI to navigate self-doubt and ambiguity, compare alternative interpretations, streamline research, and extend their practices. They also described tensions: AI could reassure or stimulate interpretation, but could inhibit intuition, produce generic readings, or alter the social and spiritual quality of the practice.^[5]

These studies show that meaning is negotiated rather than delivered by the system alone. Users select questions, decide what counts as resonance, reject or reinterpret outputs, connect them to prior beliefs, and determine whether the interaction is entertainment, reflection, spiritual practice, or guidance.

Randomness, personalization, and opacity

Randomness and personalization can produce similar subjective effects while operating differently. A randomized draw is not tailored to the user, but a reader may find relevance through symbolic interpretation. A recommendation system is tailored statistically to predicted behavior, but the user may experience the result as uncanny timing or intimate recognition. A generative model can combine both: it may interpret a random symbol using highly personal context supplied in the conversation.

Opacity matters because users may not know which process produced the apparent relevance. A feed item may reflect prior viewing behavior, network trends, commercial promotion, content availability, or platform experimentation. A chatbot answer may reflect the prompt, earlier conversation, model defaults, safety rules, and probabilistic generation. Without clear provenance, an output can appear more spontaneous, autonomous, or revelatory than its production warrants.

Opacity does not make an experience insincere, but it limits claims about causation. The user may authentically experience a message as meaningful while the system is still performing ranking or text generation. Documentary analysis should preserve both levels: the reported interpretation and the technical operation that can be established.

Authority and revelation claims

Oracle-like authority can be attributed by users, implied by interface design, asserted in marketing, or used metaphorically by critics. These sources must be kept separate.

• User attribution occurs when a person treats the output as a sign, message, prediction, spiritual response, or source of hidden knowledge.
• Practitioner framing occurs when a diviner incorporates the system into an existing ritual and explains how its output should be interpreted.
• Platform framing occurs when a product markets personalized insight, destiny, mystical access, or predictive capability.
• Scholarly metaphor occurs when researchers call AI an oracle to analyze epistemic authority, uncertainty, or power without endorsing supernatural claims.

Fischer, Applin, and Ravula compare contemporary chatbot consultation with oracles, divination, and animism through symbolic-interactionist and anthropological theory. Their focus is meaning-making between users and generative systems, including ethical concerns around consultation, forecasting, and decision-making.^[4] Welisch and Basra use historical divinatory objects to question AI's deterministic presentation in futures work and ultimately relocate the oracle role from the machine to the reflexive human practitioner.^[7] Neither analysis establishes supernatural access; both examine how authority is constructed and mediated.

Interface and perceived sacredness

Perceived sacredness is affected by more than the semantic content of a reading. Yin and colleagues conducted three experiments on AI-assisted tarot and reported that agent identity, interaction modality, and users' spiritual orientation affected sacredness, trust, and service evaluations. Participants identified as spiritual believers responded less favorably when told AI was involved, while oral conversation produced greater perceived sacredness than touchscreen interaction in the reported experiments.^[6]

This does not show that a system becomes sacred or mystical. It shows that disclosure, voice, social presence, and prior belief can alter how an otherwise algorithmic interaction is experienced. A conversational voice, embodied avatar, ritualized timing, symbolic visual design, or confident personalized language may increase the felt authority of an output without changing its evidentiary basis.

Dependency and decision-making risks

Repeated oracle consultation can shift from reflection to deference. The risk is not unique to AI, but generative systems make consultation immediate, private, responsive, and effectively unlimited. A user can reformulate a question until an answer resonates, supply increasingly sensitive context, or return repeatedly when uncertain.

Material risks include:

• Automation bias. Fluent personalized answers may receive more weight than their evidence warrants.
• Confirmation loops. Repeated prompting can produce a desired narrative and then appear to confirm it.
• Confabulation. A model may invent facts, traditions, quotations, correspondences, or predictions.
• Privacy loss. Users may disclose relationships, health, finances, trauma, location, or identity information to obtain a more specific reading.
• Emotional dependency. A system available at all times can become a preferred source of reassurance or permission.
• Commercial manipulation. Subscription design, engagement incentives, advertising, and upselling can reward repeated consultation.
• Displacement of accountable help. Oracle-like answers may substitute for qualified professional, communal, or pastoral judgment in consequential situations.

NIST's generative-AI profile does not address divination specifically, but its cross-sector analysis of confabulation, anthropomorphism, emotional entanglement, automation bias, and overreliance is directly relevant when a model is treated as an authoritative adviser.^[8]

Evidence on prevalence and sincerity

The available studies demonstrate that algorithmic and AI-assisted divination exists; they do not supply a representative global prevalence estimate. St. Lawrence and Chalfant analyze specific TikTok discourses and communities. Prock and colleagues intentionally recruited twelve tarot practitioners who already used AI, so that study cannot estimate how common the practice is among tarot readers generally.^[2][3][5]

Public posts also do not reveal one uniform level of belief. The same interface can be used sincerely, experimentally, playfully, commercially, skeptically, aesthetically, or as a prompt for self-reflection. Terms such as “oracle,” “manifestation,” and “the algorithm sent this” may be literal religious claims, community conventions, jokes, engagement strategies, or deliberately ambiguous language. Classification therefore requires evidence about the actor's own framing and the surrounding practice.

Comparative religious context

Digital systems enter traditions with different accounts of chance, fate, revelation, spirits, astrology, sacred texts, and legitimate divinatory authority. Online South Korean eight-character fortune telling remediates a learned cosmological system; WitchTok combines diverse magical and neopagan practices with platform culture; New Thought supplies a language of attraction and manifestation; tarot practitioners may describe their work as spiritual, psychological, artistic, or secular.

No single doctrine explains all of these uses. A platform category such as “AI oracle” can obscure differences between a digitized inherited system, an invented commercial product, a general-purpose chatbot used experimentally, and a religious practice in which the algorithm is granted agency. Comparative treatment should identify the tradition, community, practitioner, technical mechanism, and authority claim rather than assuming that similar interfaces express the same religion.

Evidence limitations

The evidence base is interdisciplinary and uneven. It includes ethnographic and media-studies interpretation, a small interview study of selected practitioners, experimental work on perceived sacredness, a symbolic-interactionist book chapter, a conceptual futures-practice paper, older digital-divination research, and a general technical risk framework. These sources support category distinctions and documented examples, not universal causal claims.

Platform behavior changes quickly. Recommendation systems, model versions, moderation rules, product interfaces, and commercial terms may differ from those described in a published study. Researchers also have limited access to proprietary ranking and model configurations, so technical explanations may remain incomplete even when user practices are well documented.

There is little longitudinal evidence about dependency, behavioral outcomes, sustained ritual communities, or the effects of generated readings on consequential decisions. There is also limited comparative evidence outside the specific regions, languages, platforms, and traditions represented in the current sources. Claims about prediction, revelation, or supernatural agency remain claims by users or practitioners unless independently supported by evidence appropriate to those claims.

See also

• Machine-mediated religion
• Hidden context
• Machine agency
• Religious chatbots and scripture interfaces

References

1. David J. Kim, “Divination and its Potential Futures: Sensation, Scripts, and the Virtual in South Korean Eight-character Fortune Telling.” Material Religion 15, no. 5 (2019): 599–618.
2. Emma St. Lawrence, “The Algorithm Holy: TikTok, Technomancy, and the Rise of Algorithmic Divination.” Religions 15, no. 4 (2024): 435.
3. Eric Chalfant, “I don't chase, I attract: TikTok, new thought, and the algorithms of divination.” Religion 55, no. 3 (2025): 638–657.
4. Michael D. Fischer, Sally A. Applin, and Sridhar Ravula, “External Intelligence: Oracles, Divination and Animism, and the Use of LLMs/Generative AI.” In Symbolic Interaction and AI (Emerald Publishing, 2025), 73–114.
5. Matthew Kieran Prock, Ziv Epstein, Hope Schroeder, Amy Smith, Cassandra Lee, Vana Goblot, and Farnaz Jahanbakhsh, “Interpretive Cultures: Resonance, randomness, and negotiated meaning for AI-assisted tarot divination.” Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (2026).
6. Mengjiao Yin et al., “Can a Mystical Experience be Emulated by AI-Generated Rationality?” International Journal of Human–Computer Interaction (published online 2025).
7. Gaston Welisch and Santini Basra, “AI & The Oracle: Future Interfacing Technical Objects and the Reflexive Practitioner in Futures Studies.” World Futures Review (published online 2026).
8. Chloe Autio et al., Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile. NIST AI 600-1 (2024).
9. Cheng Kuo, “A study of the consumption of Chinese online fortune telling services.” Chinese Journal of Communication 2, no. 3 (2009): 288–306.