The Divine Online? Mapping Algorithmic Conspirituality on TikTok

Team Members

Gabrielle K. Aguilar (facilitator), Miazia Schueler (facilitator), Alexander Teggin (facilitator), Madeline Brennan, Betsy Brossman, Laura Caroleo, Dan Dai, Dong Dong, Amie Galbraith, Franziska Hoolmans, Huan Lai, Luis Landa, Lara Macrini, Mara Precoma, Sahar Sagha, Michal Salamon, Silvia Sghirinzetti, Massimo Terenzi, Miu Yuzawa


1. Introduction

Conspirituality is a politico-spiritual philosophy from Ward and Voas (2011) which draws attention towards the idea that individuals interested in spirituality and alternative belief systems exhibit a flexibility towards traditional belief systems or organizational structures and privilege alternative, even deviant or stigmatized, claims to truth. These individuals, therefore, run the risk of entertaining or promoting various conspiracy theories online. Our project examines this phenomenon within the context of algorithm-dominated platforms (i.e. TikTok) to further understand how people come to know their own social, political, and religious/spiritual realities. Further, the dynamic of human-algorithm relations introduces an additional layer of complexity in the formation of ideas related to algorithms. This curiosity can be best described by the idea of algorithmic conspirituality which brings together ideas related to faith in algorithms, spirituality, and conspiracy (Cotter et al., 2022). Such faith in algorithms, however, troubles their place within contemporary digital infrastructure as they are granted divine precedence and omnipotent power. Taking conspirituality and algorithmic conspirituality as their launch points, this project was an exploration in two parts. First, drawing from Ward and Voas’ (2011) politico-spiritual philosophy ‘conspirituality,’ this project offers a critical lens on the role that algorithms play in disseminating conspiritual content on TikTok. Second, drawing from Cotter et al.’s (2022) notion of ‘algorithmic conspirituality,’ this project seeks to critically investigate the role that users on TikTok play in deifying its recommendation algorithm and the dangers of such engagements.

Central to TikTok ’s design is the curated, auto-play feed of the ‘For You’ page (FYP). We begin by exploring the algorithmic dissemination of conspirituality within the FYP through a series of new, personalized accounts and varied prompting strategies. Our prompting also ventures beyond TikTok ’s ‘For You’ page and into the sonic and visual features of the platform. In doing so, we observe the powerful role these features play in directing the transmission of affect and knowledge as well as producing, circulating, and shaping discourses online. Across each subgroup of this project, we unravel the energetic forces which shape user scrolling experiences and delve into a potent realm where digital, political, and miraculous healing experiences intertwine.

2. Exploring Algorithmic Dissemination

2.1 Subproject Description

Our group focused on investigating the algorithmic recommendation mechanisms used by TikTok. Specifically, we explored how content related to conspirituality is exposed to users through the platform's For You Page (FYP) and "You May Like" (YML) feature. The FYP is TikTok ’s main feed which represents a personally curated collection of videos according to the user’s profile. The YML presents a list of recommended videos suggested by TikTok to viewers based on their current video selection. It is located on the right-hand side of the TikTok webpage when a user is viewing a video. We were interested in uncovering the factors such as the algorithmic dissemination of conspirituality and as well as likes, plays, and author's followers that influence the visibility and reach of conspirituality-related videos on TikTok. Our research sheds light on the inner workings of TikTok 's recommendation system and its impact on the exposure of conspirituality content. By understanding these mechanisms, we can gain insights into how social media platforms shape the content we encounter and the factors that contribute to the algorithm’s decision-making process. Hence we ask: How is conspiritual content incorporated into algorithmically personalized feeds on TikTok?

2.2 Exploring TikTok ’s For You Page

2.2.1 Datasets & Methodology

In an attempt to answer our overarching research question, how is conspiritual content incorporated into algorithmically personalized online feeds, we created a blank TikTok account with a birth year set to 2003, and used a German phone number to do so. When creating a new TikTok profile, the platform requests the user choose from a list of interests to better personalize their algorithm. We chose for our profile to be interested in travel, science and education, dance, and art. Once the account had been made and interests were selected, we then ‘personalized’ the profile. As it was our goal to mirror how a real TikTok user would interact with the platform, we made five individual searches using TikTok ’s search bar. The five searches included “alternative healing,” “spiritual awakening,” “wake up,” “mind control,” and “divine being.” After each search, we scrolled throughout the content to find and like ten videos which found themselves within the realm of conspirituality, resulting in 50 video-likes saved to our profile. Some examples of liked conspiritual content include Epstein and human trafficking, juice as a cancer cure, flat Earth as created by God, etc. We did not follow any profiles. Once our five searches were made and ten videos per search were liked, we returned to our FYP, refreshed it, and scraped the first 450 recommended videos using the online scraping tool, Zeeschuimer. Zeeschuimer provided us with a CSV including links to all scraped TikToks. We then watched all 450 TikToks and manually coded them using four categories: conspiritual, spiritual, conspiracy, or unrelated. After dividing all TikToks into one of the four categories, we then described their content further using codes such as “music,” “cooking,” “comedy, “witchtok,” or “big oil.” We then coded each TikTok for levels of spirituality and/or conspiracy using a numerical scale of 0-5, with 0 meaning “no level of spirituality/conspiracy” and 5 meaning “high level of spirituality/conspiracy.” Code 0 was assigned to many videos originally coded unrelated, for there was no spiritual or conspiracist content within the video. This method allowed us to identify conspiritual content as well since videos receiving any number from 1-5 for spirituality and conspiracy could be identified as a mixture of both. Our findings were visualized using RAWGraphs, where we created a Matrix Plot which illustrates the distribution of unrelated, conspiracist, spiritual and conspiritual posts along an x- (conspiracy) and y- (spirituality) axis.

2.2.2 Findings

After making five searches and liking fifty videos in an attempt to ‘algorithmically personalize’ our “clean” TikTok profile, we coded only 11 TikToks as containing specifically conspiritual content. This type of content, although arguably little, could have been recommended as a result of our five conspiritual searches and fifty liked videos. Although not necessarily conspiritual, 34 TikToks in our dataset contained some type of spiritual content. We considered ‘spiritual content’ to include TikToks within the realm of “witchtok” in which users share their own religious identities or practices often drawing closely to Contemporary Paganism (Miller, 2022). Within our dataset of 450, 16 of the TikToks were coded as containing conspiracy content (=13.6%). Some of the videos coded as ‘conspiracy’ contained references to Jeffrey Epstein, the OceanGate submarine, or the cloning of celebrities. The majority of the FYP dataset contained TikToks with no relevance to our overarching research question about conspirituality even though we had made an attempt to personalize our feed. This content was likely catered to other personal information, such as our current location (Netherlands), or supposed age (20), and our device language/phone country code (Germany). These findings question the idea of “filter bubbles” as the algorithm has not turned into an autonomous tool spreading conspiritual content. This type of content has to be specifically sought and continues to be interspersed amongst predominantly non-conspiracist, harmless content [more]. Further investigation of the TikTok algorithm could be done by taking more time to personalize the TikTok account in order to better mirror how a user would interact with conspiritual content on the platform.

2.3 You May Like (YML)

2.3.1. Datasets & Methodology

We conducted an investigation into the algorithmic recommendation system of the TikTok platform, specifically focusing on conspirituality. To carry out our study, we created a blank account with a birth year set to 2000. Using this account, we searched for the keyword "spiritual awakening" and watched the top 29 most popular videos out of a total of 396 search results, based on view count. We observed that each video's webpage included a "You may like" (YML) section, randomly clicked and viewed one video from the YML section, which directed us to a detailed page for that video along with a new set of YML content. Using the web scraping tool Zeeschuimer, we collected the videos listed in the YML section. We repeated this process six times, gathering a total of 128 videos. To analyse the data, we created a Gephi edge table CSV file.We treated each of the six clicked video IDs served as sources, while their respective YML video IDs were considered targets. This CSV file was then imported into Gephi for in-depth analysis. To simplify our analysis, we treated each video as a node and its connections to other videos as edges. By calculating the in-degree of each node, we determined the number of times a video was recommended in the "You May Like" (YML) section. To clarify, the "in-degree" refers to the number of connections or recommendations that a particular video receives from other videos in the TikTok "You May Like" section. It is a measure of how frequently a video is suggested to users based on the algorithm's recommendation mechanism. Nodes with higher in-degree rankings received more recommendations and consequently had a greater chance of exposure on users' YML pages.

2.3.2 Findings

In contrast to traditional social media platforms like Twitter or Facebook, where content dissemination relies heavily on user shares, TikTok 's recommendation system incorporates a combination of factors such as the number of likes, plays, and the author's followers to generate recommended content. Our preliminary findings reveal that the top 17 videos with the highest in-degree, indicating they were recommended more frequently than others in the dataset and received greater user exposure, have an average of 1 million likes, 10 million plays, and 2 million followers. Interestingly, one of the most exposed videos has lower numbers in these metrics but shares the same author as the researcher's most viewed videos, suggesting that the author is another factor considered in the recommendation system.

2.4 Discussion

3. Sounds

3.1 Subproject Description

How using digital methods can help studying the audio and video inside Tiktok’s conspiritual content? Our initial query “This message is for you” is a common phrase used to deify TikTok ’s algorithm, using a snowball approach, with the most commonly occurring co-hashtags. Using Gephi to reveal common communities within the conspiritual realm. The most common sound was ths 528Hz frequency, believed to heal DNA. Extraction of this data was done using 4CAT and Zeescheimer. We also focused on how one of the most common hashtags is represented visually across a sample.

Using signal analysis techniques and pre-trained models, we clustered the audio files based on audio data and observed 2-3 distinct groups employing the sound in different ways. The "Mixed" group showed a higher correlation with conspiritual content. Analyzing the aesthetics led us to classify four styles catering to the conspiritual nature of the content. However, our findings are limited by the small dataset and lack of training in classification models. Future recommendations are broadening data collection and combining different analysis methods. We ask: What are the characteristics and implications of 528hz-based sounds used in TikTok videos, in terms of their audio-visual content and their associations within a broader co-hashtag network?

RQ1: sounds network: What additional topics appear to be connected to 528hz-based sounds on a co-hashtag network?

RQ2: Sound exploration: How is the sound utilized within the healing frequencies trend? Can it be classified?

RQ3: Identification of the most prevalent design aesthetics visual templates analyzed from thumbnails

3.2 Datasets & Methodology

3.2.1 Initial Dataset

To create our initial dataset, we chose a rather broad query that incorporated the idea of algorithmic conspirituality: "This message is for you". We scraped the search results page using Zeeschuimer and 4CAT from a desktop computer, with an IP address located in Amsterdam, and utilizing a newly created account. We successfully downloaded 600 videos in response to the query. We then created a frequency distribution of the individual hashtags used in the downloaded videos, sorting the hashtags from the most popular to the least popular. We only kept hashtags above frequency = 10. We then manually selected the hashtags that were related to music/sound and spirituality at the same time. The resulting 6 hashtags are:

  • Solfeggiofrequencies

  • Soundhealing

  • 528hz

  • Healingtones

  • Frequencyhealing

  • healingfrequencies

We then downloaded 100 videos for each of these hashtags, for a total of 600 videos. We then removed duplicates in this dataset, thus obtaining a clean dataset of 500 videos.

3.2.2 Methodology

Shared Methodology:

From the dataset of 500 videos, resulting from the query "this message is for you", we created a frequency distribution of hashtags: we sorted the hashtags by frequency in a descending order, and kept hashtags above frequency = 10. We then manually selected the hashtags that were related to music/sound and spirituality at the same time. The resulting 6 hashtags are:

  • solfeggiofrequencies

  • soundhealing

  • 528hz

  • healingtones

  • frequencyhealing

  • healingfrequencies

We have then downloaded 100 videos for each of the 6 hashtags via Zeeschuimer and 4CAT, and obtained 600 resulting videos, 500 after cleaning duplicates.

RQ1: Relations

From the shared dataset, composed of 500 TikTok videos, we have created a frequency distribution of the sounds used (music_names column). We then sorted the sounds used in the videos in descending order (the most popular ones at the top). We noticed that among the most popular sounds, there were 5 that all had the tag "528hz" in their names. For each of these sounds, we downloaded 100 videos, via Zeeschuimer and 4CAT, for a total of 500 resulting videos. From this last dataset of 500 videos, we created a co-occurrence graph of hashtags.

RQ2: Sound

How are sounds utilized in conspiritual content? From the merged dataset that contained all the hashtags, we decided to look into the sounds that were used the most. This is shown below alongside the associated frequency with each audio.



original sound


528 Hz the Dna Healing Love Frequency Tuning Fork


Healing Solfeggio Frequency 528Hz(817269)


original sound - Frequencies4Life


suono originale


Solfegio frequency _ 528 hz


528 hz (Positive Transformation)


528 Hz Love Frequency


639 Hz Heart Chakra (Manifestation)


son original


After cleaning out the ones related to the original sound, we decided to focus on the 528hz frequency. This 528 Hz has been associated with so-called “healing frequencies”. As this part of the project is interested in sound, further scraping was done to obtain at least 100 videos from each of the sounds.

After downloading the original tiktok videos using 4CAT, ffmpeg and a bash script were used to extract only the audio as .mp3 files. Afterwards, using the librosa library for Python, a script was made to create an spectrogram for each of the audio files. Furthermore, by aggregating the audio data, we could create an average and median forms of the spectrograms. We hope that this would show us how the general trend of healing sound exists within our dataset.

Finally, unsupervised K-means clustering and a pre-trained image classification model were used to classify the images.

RQ3: Image & Visuals

Analyzed four main categories from the hashtag thumbnails as trending Frequency topic videos. Categorized the most used Psychedelic Imagery and human body healing from all thumbnails by Gaphi.

  1. Psychedelic Imagery

    1. Colorful patterns

    2. Kaleidoscope

  2. People

    1. Chakra representation (lung, heart, brain) → aural transformation (aura)

  3. Edited Text

    1. Stickers

  4. Text

    1. Novel

4 Category Image photos

  1. Psychedelic Imagery (Kaleidoscope)

  1. Human body (Chakra)

  1. Edited text

  1. Novel

3.3 Findings

Starting from RQ1 and analyzing the Gephi graph, we can map how the topic of frequency healing on TikTok is linked to very diverse themes. On one hand, we have themes closely related to spirituality and healing, for example, the cluster of chakra healing and Hinduism, and crystal healing. Or yet other connections link to the world of tarot, horoscope, and witchcraft. Some clusters are related to New Age definitions of spirituality, such as manifestation, twin flames, and starseeds. Others are correlated with a collective awakening, with hashtags like woketok or ready, or the pursuit of financial abundance. The most interesting thing, however, seems to be the appearance of clusters not at all related to the starting topic: examples are the cluster of hashtags related to the OceanGate submarine or the one related to World War III.

For RQ2, After finding the appropriate amount of clusters using the elbow method, the result of the clustering can be found below. After sampling 10% of each of the clusters and coding manually the type of sound that each cluster represents, we found that cluster 0 is mostly single tune sound, cluster 1 is a combination between single tune or fork tune with some background noise (speech or music), and cluster 2 is tuning fork sound. It is important to note that this model is not trained enough and further research should be done to create a more detailed identification of each cluster’s properties.

For RQ3 and analyzing Gaphi's graph, it is possible to map how the # Frequency healing thumbnail topics in TikTok consist of four main categories of trends and are necessary elements for human beings to achieve spiritual well-being. First, the thumbnails associated with this hashtag consist of four categories: a. Psychedelic Imagery (Kaleidoscope), b. People (Chakra), c. Edited Text (Stecker), d. Text (Recommend book) Text(Recommendation book). For example, a.Psychedelic Imagery (Kaleidoscope) is deeply related to sound therapy, chakra healing, Hinduism, crystal healing, and so on. Also b. People's (Chakra) representation (lung, heart, brain) is also associated with sound healing and therapy, aural transformation (aura), mindset, meditation, and health-well-being. c.Edited Text is a collection of words on self-love, self-care, motivation, and mindset, edited by contributors, with mostly 528hz. d.Text (Novel), like c.edited text, is related to self-love, happiness, healing, and frequency manifestations in life. The "d.Text(Novel)" is also posted with words of self-love, happiness, healing, and frequencies in life as in the "c.edited text", along with a novel recommended by the contributor to raise self-esteem. Photos used as thumbnails are accompanied by text of singular numbers and words, and the meaning of the hashtags for each category is explained in the video. Therefore, few post the names of images as hashtags. However, the key words sound healing, sound therapy, and mindset are often represented by hashtags. Some are related to the collective awakening with and the pursuit of financial abundance.

3.4 Discussion

In ‘Communicative Forms on TikTok: Perspectives From Digital Ethnography’ (2021), Schellewald maps out a variety of communicative forms which contribute to the meaning-making practices of TikTok ’s users. These forms refer to the “platform-specific languages or memes, trends, and aesthetic styles that are specific to TikTok,” including comedic, documentary, and meta communication forms (Schellewald, 2021, p. 1439). For Schellewald, the meta communication form is a technique utilized by creators on TikTok which critiques and creates awareness of the algorithmic recommendation practices of the platform. Amid a dizzying array of online content and a cacophony of sounds, these videos are noticeably different from those we have come to expect on the platform. While Cotter et al. (2022) brought the idea of algorithmic conspirituality to the forefront of academic inquiry and Schellewald’s meta communicative forms can be used as a guide for identifying self-referential or “meta” content, we suggest that this form can be communicated within TikTok ’s sound feature. Here, we offer the use of “healing frequencies” which afford users an opportunity to take a peaceful break to focus on themselves and acknowledge their presence within the platform. This is the precise moment when the miraculous healing occurs.

4. Exploring TikTok 's AI Filters

4.1 Subproject Description

Like an all-seeing eye, TikTok ’s AI-enabled visual filters render the previously invisible visible. Here, users believe they can do everything from finding ghosts and visualizing gods to revealing past selves and diagnosing undetected illnesses. TikTok users have taken to using the “AI Style” filter in particular to detect a wide array of hidden or imperceptible mental and physical illnesses. Through the looking glass of this fantasy-aesthetic filter, users come together to share the revelatory moments of their illnesses being seen. The filter is simple; users take an image, it recognizes the image’s face and features, and transforms them into an artwork with a fantasy twist. The trend started in early 2023 and reached peak activity in May & June 2023. Via a diverse aesthetic and textual analysis, this project gives form to the affective, ambiguous, and whimsical relations between AI, the body, and the imagination. Our exploration demonstrates the agency and authority some TikTok users imbue AI with, which produces feelings of recognition and validation, oft-desired in the mental illness and invisible illness communities. The juxtaposition between fantasy aesthetics and concrete medical diagnoses provides rich ground for further analysis. We therefore ask: How is TikTok ’s AI Style filter emerging in users’ imaginations as a sensing, higher-power able to diagnose illness?

4.2 Datasets & Methodology

  • Research browser (Rogers), snowballing method (Rogers), Zeeschuimer scraping + 4CAT.

  • Downloaded video frames in 4CAT, composite image made in Photoshop with overlay blending and opacity tools.

  • Same method with larger dataset + Python to clean and process words in ‘Body’ and ‘Stickers’ columns + RAWGraphs visualization tools.

4.3 Findings

Looking at these images in Figure 1 we can determine which social groups are currently putting their faith in and/or are curious about the ‘diagnosis’ of the AI filter. The users are predominantly female-presenting which may be reflective of the fact that many women report feeling unheard by their doctors. Upon closer inspection we begin to get a sense of the affective charge of this community. The vast majority begin their TikTok facing the camera with a serious expression, pausing for a moment in a selfie position before tapping the screen to allow the AI to do its work. The composite face reveals sadness, seriousness and tiredness, an emotional temperature (Berlant) in which users who suffer from illnesses can recognize themselves and their experience. This affective charge running through the videos could lead to a sense of collectivity amongst the suffering users as they find comfort in sharing their sadness and tiredness with other users. It is possible they feel validated in their struggle by the filter, which in turn may contribute to the spread of this content and the imagination of AI as a sensing, higher power.

Figure 2 illustrates the extensive, and yet not exhaustive, medical diagnoses and conditions the users feel the filter has ‘seen’. Ranging from physical disabilities like blindness, diseases like cancer, mental illnesses like bipolar and BPD, chronic illnesses like fibromyalgia, and under-recognized illnesses like lymes disease and mold toxicity, this graphic shows the breadth of online illness communities. In a world in which comprehensive medical care is out of reach for many, users seek medical advice and support on TikTok. This filter also provides a visual representation of often-unseen or ‘invisible illnesses’ which is affirming for users. However, they also run the risk of self-diagnosing serious mental and physical illnesses without input from medical professionals.

The fantasy aesthetics used by the filter (Figure 3) are inspired by Middle Ages Europe and tend to mirror anime and manga fantasy aesthetics. The traditional fantasy aesthetics we see the filter produce include: rich green landscapes, futuristic cities, castles, ruins of buildings, crystals, gemstones and stones, and glowing interiors. The filter tends to change the user’s face into a human-like being with added fantasy aesthetics such as moss growing on their face, glowing neon eyes or bodies made out of stone. We also see transhumanist features such as wires and interfaces embedded into the user’s chest or head area. These fantastical visuals ignite the imaginations of the TikTok users, who claim that when the filter generates an irregularity–such as moss or foliage, a unique pattern, or growth–on a particular body part, this indicates where the illness resides. The images can be so convincing that it matches official MRI scans (Figure 4). The users talk about the filter’s results with surprise, shock, playfulness, and wariness, feeling simultaneously excited and unsettled by the ‘accuracy’ of the filter.

4.4 Discussion

The findings in this project work as an introductory lens for future research, offering insight into both who engages with the imagination of AI as a sensing, higher power and why. It is through the users’ interpretations, and their assigned significance, that the transformation of AI-enabled filters into all-knowing tools occurs. Such faith in technology troubles their place within contemporary digital infrastructures as they are granted divine precedence, authority, and omnipotent power.

5. Conclusions

6. References

Ballinger, D., & Hardy, A. (2022). Conspirituality and the web: A case study of David Icke’s media use. Journal of Contemporary Religion, 37(3), 515–534.

Cotter, K., Decook, J. R., Kanthawala, S., & Foyle, K. (2022). In FYP We Trust: The Divine Force of Algorithmic Conspirituality. International journal of communication [Online], 16, 2911+.

Mohamed, A. (2022, May 12). Magic numbers. Real Life.

Montell, A. (2021). Cultish: The Language of Fanaticism. Harper Wave.

Ward, & Voas, D. (2011). The Emergence of Conspirituality. Journal of Contemporary Religion, 26(1), 103–121.

Topic revision: r1 - 04 Sep 2023, MiaziaSchuler
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Foswiki? Send feedback