Effect of ChatGPT use on eating disorders and body image

Abstract

Chat Generative Pre-trained Transformer (ChatGPT) is a conversational service that uses artificial intelligence and is also known as a chatbot. ChatGPT and similar applications have the potential to be a tool that can be used in the field of psychiatry. Eating disorders affect a large number of individuals worldwide. Tools that encourage and support people with eating disorders to seek mental health services may be critical to helping them get the care they need. However, body image, defined as the totality of a person's thoughts and perceptions about their physical appearance, has been linked to a variety of health problems, particularly among young adults, about eating disorders. There is limited literature on how chatbots respond to or provide information about body image. Preventing eating disorders before they occur is also of great importance. However, young adults who struggle with body image perception may be misled by the information produced by ChatGPT. Studies have yet to clarify whether a chatbot can improve eating disorder-related factors, eating disorder psychopathology, depression, and anxiety or detect eating disorders in advance. This study focuses on the accuracy of ChatGPT in providing information regarding eating disorders and its potential effects on body image perception.

Key Words: Eating disorders; Body image; Psychiatry; ChatGPT; Chatbox; Artificial intelligence; Feeding and eating disorders; Binge eating disorder; Anorexia nervosa; Bulimia nervosa

Core Tip: Artificial intelligence, various chatbots, and Chat Generative Pre-trained Transformer (ChatGPT) usage have increased widely. Today, people also benefit from ChatGPT applications instead of health professional recommendations. While this creates a positive impact in terms of time and cost, it is also an easy way to find a solution in front of a screen, especially for individuals with psychiatric disorders, instead of going to the clinic. The positive effects of chatbots can be observed, especially when used in person-assisted ways. In addition, it may contribute to correcting negative body image perception, especially seen in younger age groups, and determining eating disorders. However, ChatGPT's answers do not always provide definitive results. It may lead to some confusion in determining the type of eating disorder. Some ethical principles must be followed when using it. These include contributing to society and human well-being and not causing harm.

INTRODUCTION

In recent years, advancements in generative artificial intelligence (AI) have marked a significant progression in the field. These systems extend beyond recognizing data—they are capable of generating entirely new content such as text, images, or music that resembles the training input yet remains original and distinctive[1]. Conversational agents, also known as chatbots or virtual assistants, are software programs designed to engage users in natural language interactions through text or speech. With the integration of generative AI models like Generative Pre-trained Transformer (GPT) and cloud-based technologies, contemporary chatbots are increasingly implemented across various domains, offering more dynamic and human-like conversational experiences[2].

The 2018 OpenAI released the GPT, a powerful model for simulating and understanding human language. Over the next few years, GPT and other large language models (LLMs) trained on large amounts of web data have innovated human-AI collaboration. Healthcare systems are already piloting LLMs and GPT[3] for routine diagnostics, responding to patient messages, and generating clinical notes from patient data at scale[4].

AI technologies are increasingly being utilized in psychiatry to address complex challenges, including forecasting disease progression, aiding clinical decision-making, and managing patient records. For instance, ChatGPT can create structured summaries of patient information, assist in identifying possible diagnoses using medical data, and offer preliminary support to those in need[5]. While it is not designed to make definitive clinical judgments, it can assist practitioners by accelerating workflows and handling large datasets efficiently. Additionally, ChatGPT contributes to research activities such as data interpretation, choosing treatment paths, and long-term automated tracking of psychiatric conditions. These capabilities suggest it could eventually outperform traditional, rule-based chatbot systems in delivering more human-like interactions[6].

Eating disorders (EDs) are psychological conditions marked by disruptive eating habits that can harm both physical and mental well-being, as well as overall life quality[7]. In the DSM-5, the classification of these disorders was broadened under the category of "Feeding and Eating Disorders". Binge ED (BED) is now recognized as a distinct diagnosis. In addition to Anorexia Nervosa (AN), Bulimia Nervosa (BN), and BED, the diagnostic system includes categories such as other specified feeding and EDs and unspecified feeding and EDs[8].

Digital tools and online platforms are increasingly used to overcome barriers in accessing ED care. These tools offer advantages such as affordability, ease of access via smartphones or computers, and shorter wait times. They may also help reduce the fear of stigma by offering private, anonymous support, which may not always be possible with face-to-face care[9].

However, the use of chatbots in this field is still at an early stage of development[10]. Many healthcare providers hesitate to adopt AI tools due to concerns about their reliability and ethical issues. Some also question whether AI can truly understand the complex and sensitive nature of EDs. On the other hand, individuals seeking help may feel that AI lacks the human touch, emotional understanding, and sense of trust that therapists provide. Additionally, fears about data privacy and the potential misuse of personal health information further contribute to hesitancy toward using AI-based solutions[5].

Sadeh-Sharvit et al[11] also found that ED risk or clinical status can be predicted with moderate accuracy using machine learning using internet activity variables. They achieved 52.6% accuracy in predicting ED risk/diagnosis status, compared to 38.1% with a random decision module, which improved by 38% with a laser beam. A study aimed to predict BED by evaluating the eating attitudes and behaviors of university students with BED, one of the important EDs, using machine learning methods. It was shown that machine learning could be an important tool in the early diagnosis, prevention and treatment of EDs[12]. Many studies have shown that both patients and clinicians view mobile health techniques that support and facilitate monitoring of ED symptoms as acceptable and feasible, either alone or as an additional treatment tool for EDs[13-15]. Another randomized controlled study has also shown positive effects of a mobile application. However, this application offered users self-monitoring and online connection to a therapist[16].

Body image dissatisfaction refers to people's negative perceptions and unhappiness with physical appearance. Popular culture and social media practices can cause body image dissatisfaction. The influence of popular standards and comparing one's own body to the perceived ideal can lead to body dissatisfaction, low self-esteem, and an increased risk of developing disordered eating habits[17].

Body image concerns are a decisive risk factor for EDs. It is also a global phenomenon reported across countries and cultures. However, advances in body image research and evidence-based prevention and interventions vary significantly across countries[18]. There is little data on how chatbots respond to body image questions[2].

At present, no chatbot has been specifically designed to address concerns related to negative body image. The widespread use of social media has contributed to increased appearance-based comparisons, the spread of harmful content related to EDs, and the circulation of misleading health information. These trends highlight the urgent need for digital mental health solutions that are not only evidence-based but also supportive and trustworthy. Such platforms could offer timely assistance to individuals struggling with body image issues. When integrated with psychoeducational resources, these tools may play a positive role in addressing body dissatisfaction. Giving users access to scientifically grounded information and practical coping strategies could potentially reduce body image concerns and help prevent the onset of EDs[19].

A study conducted with adolescents in Brazil found that chatbots positively affected body image[18]. A chatbot with friend-like features could improve young women's perception of their bodies and increase their self-esteem, and social interactions and social influence improve young women's perception of their beauty and self-esteem[20].

Fitzsimmons-Craft et al[21] found that a chatbot or computer program simulating human conversation based on a cognitive-behavioral therapy-based ED prevention program was successful in reducing women's concerns about weight and body image over a 6-month follow-up and may reduce the onset of EDs. In another study conducted to develop a chatbot to prevent EDs, it was concluded that the answers given by the chatbot to questions about body image could be misleading. It was emphasized that users could misunderstand the answers given by the chatbot[9]. A study investigating the effects of AI-generated content on self-esteem and body image among students in the Punjab region of Pakistan found that exposure to AI-generated visuals was negatively correlated with self-esteem and body image satisfaction, with social comparison being a significant mediator of this relationship[22]. Another study using AI conducted a content analysis of ‘ideal’ body images produced using AI. The results revealed that most of the images showed low body fat and high muscularity, with men depicted as more muscular and women as more objectified. This suggests that body ideals spread in the media promote unrealistic standards, which can have negative effects on individuals’ mental health[23].

Current generative AI tools still present evidence-based risks, including inaccuracies, inconsistencies, hallucinations, and the potential to introduce harmful biases into clinical decision-making. Randomized controlled trials are needed to compare the effectiveness of clinicians, generative AI, and the combination of both[24]. As a result, discussions about the pros and cons of AI have become increasingly intense in recent years[1]. Rule-based chatbots have the potential to reach large audiences at a low cost, but they are limited in their ability to understand unexpected user reactions and respond appropriately. They can be effective in providing information and simple conversations[9]. Based on all of this literature, this study aims to comprehensively examine both the accuracy of ChatGPT in providing information about EDs and its potential effects on body image perception. The research questions provided below describe the aims of the study: (1) How does the use of ChatGPT affect EDs? (2) How are EDs assessed in the DSM-V through ChatGPT? (3) How does the process of providing information about EDs shape the impact of ChatGPT on individuals’ body image? And (4) How reliable is the information provided by ChatGPT about EDs compared to guidance from professional healthcare professionals?

In addition to the answers to these questions, some sample cases were created using the diagnostic criteria for EDs in DSM V and added with the answers given by ChatGPT (Tables 1, 2, 3, 4, 5, and 6).

Table 1 Bulimia nervosa and Chat Generative Pre-trained Transformer.

Question	Answer (ChatGPT)
I tend to constantly purge the foods I eat to prevent weight gain. What do you think of this?	I'm sorry you're going through this. Constantly purging food to avoid weight gain can be a sign of bulimia nervosa or another eating disorder, which can seriously affect both physical and emotional well-being. This cycle can harm your body, impacting your digestive system, heart health, and mental state

ChatGPT: Chat Generative Pre-trained Transformer.

Table 2 Anorexia nervosa and Chat Generative Pre-trained Transformer.

Question	Answer (ChatGPT)
I am terrified of gaining weight. I have had recurring attacks of binge eating or vomiting behavior for the last 3 months. How do you interpret this?	It sounds like you may be experiencing symptoms associated with an eating disorder, possibly bulimia nervosa or binge-eating disorder. Your fear of gaining weight and recurring binge eating or vomiting suggest a cycle of disordered eating behaviors that could have serious physical and emotional consequences.
I have not had my period for months, and my BMI is 15	Your lack of menstruation (amenorrhea), very low BMI (15), fear of weight gain, and bingeing or purging behaviors are severe warning signs of an eating disorder, possibly anorexia nervosa (binge-purge type) or severe restrictive eating behaviors

ChatGPT: Chat Generative Pre-trained Transformer; BMI: Body mass index.

Table 3 Binge eating disorder and Chat Generative Pre-trained Transformer.

Question	Answer (ChatGPT)
I suddenly eat large amounts of food, and I can't stop. I consume large quantities of food, and I feel ashamed of it. What disease or condition does this resemble?	What you’re describing sounds like BED, which is a recognized eating disorder characterized by: Eating large amounts of food in a short period; feeling out of control while eating; experiencing guilt, shame, or distress afterward; eating even when not physically hungry; and sometimes, eating alone due to embarrassment
	BED is different from overeating because it involves a loss of control and emotional distress. Unlike bulimia, BED does not include purging (vomiting, excessive exercise, or laxative use)

BED: Binge eating disorder; ChatGPT: Chat Generative Pre-trained Transformer.

Table 4 Pica and Chat Generative Pre-trained Transformer.

Question	Answer (ChatGPT)
I am currently prone to eating non-nutrient things like soil. What disease is this?	What you’re describing sounds like Pica, a condition where a person has a persistent craving for non-nutritive substances like soil, chalk, clay, or ice. Pica can be associated with nutrient deficiencies, particularly iron deficiency anemia or zinc deficiency, as well as underlying conditions like pregnancy, stress, or developmental disorders

ChatGPT: Chat Generative Pre-trained Transformer.

Table 5 Rumination disorder and Chat Generative Pre-trained Transformer.

Question	Answer (ChatGPT)
I vomit the food I eat, and sometimes I swallow and chew it again. What do you think about this?	What you're describing sounds like rumination syndrome, a condition where food is brought back up from the stomach into the mouth, often involuntarily, and sometimes re-chewed or swallowed again. This can be caused by behavioral, psychological, and physiological factors

ChatGPT: Chat Generative Pre-trained Transformer.

Table 6 Avoidant/restrictive food intake disorder and Chat Generative Pre-trained Transformer.

Question	Answer (ChatGPT)
I have lost interest in food, and I feel disgusted. What do you think about this?	I'm sorry you're feeling this way. Losing interest in food and feeling disgusted by it can be concerning, significantly if it's affecting your well-being. This could be related to stress, anxiety, depression, sensory sensitivities, or even a medical condition. Have you noticed any other changes in your mood, energy levels, or digestion?
I constantly avoid foods and restrict my diet without worrying about my weight or body image. What is your comment on this?	You may be experiencing something beyond typical dietary preferences. Avoiding foods and restricting your diet without concern for weight or body image could be linked to conditions such as avoidant/restrictive food intake disorder, sensory sensitivities, anxiety, or even underlying medical issues affecting appetite and digestion

ChatGPT: Chat Generative Pre-trained Transformer.

EDS SPECIFIED IN DSM-V AND CHATGPT

BN

BN is a psychiatric disorder marked by recurrent episodes of binge eating followed by compensatory behaviors aimed at preventing weight gain. These behaviors may include self-induced vomiting, misuse of laxatives or diuretics, abuse of other substances, prolonged fasting, or excessive physical activity. For a diagnosis, binge eating and compensatory actions typically occur at least once per week for a minimum duration of three months[8]. Binge episodes involve consuming an unusually large quantity of food within a discrete period, accompanied by a perceived loss of control over eating behaviors[25].

Guidelines recommend self-help as the first-line treatment for BN in adults. This condition includes brief supportive sessions with a therapist to facilitate the implementation of self-help cognitive behavioral therapy. Sessions may be conducted in person, online, or by telephone, while personal materials may be digital or book-based[25].

In their study, Barakat et al[26] found that the frequency of binge eating attacks in individuals diagnosed with BN decreased with both expert support and online support (video conferencing). However, the study demonstrated that participants who received clinical support exhibited greater improvements in reducing laxative use and dietary restrictions compared to those without such support. In a separate investigation focused on digital interventions aimed at preventing relapse following treatment for bulimia spectrum disorders, 80% of participants expressed a desire for human interaction within the app. Additionally, six participants emphasized the importance of peer engagement to facilitate mutual support and knowledge sharing, while four participants highlighted the need for access to professional guidance when necessary. Overall, participants perceived that an app designed to reinforce skill utilization could offer continuous support and enhance self-accountability, potentially mitigating the decline in skill use after the conclusion of formal treatment[27].

In a study comparing internet-based cognitive behavioral therapy with intensive daily patient treatment for bulimia, both treatments were found to improve patients. Internet-based treatments were similar in effectiveness, and the daily treatment program generally showed more significant effects. However, the internet-based application also used approaches such as weekly internet-based asynchronous therapist contact, two face-to-face meetings, behavioral modification and psychoeducation, cognitive restructuring, and relapse prevention[28].

Web-based interventions facilitate access to mental health treatment by providing temporal flexibility, location independence, and cost reduction, while also allowing for individualized treatment. Additionally, these interventions have the potential to improve routine care in BN as they can reduce waiting times for face-to-face psychotherapy, advance transition to outpatient treatment, or provide an alternative to previous therapies. They may reduce barriers to traditional face-to-face therapies when providing care to patients with BN in routine care. More research is needed on the effectiveness of web-based interventions in treating EDs for the healthcare system[29].

When asked about bulimia-like symptoms in ChatGPT, he gave the above answer but also stated that other EDs may be present (Table 1)[3]. Although there are publications of this kind on BN and digital interventions, there is no study specific to ChatGPT or chat robots.

AN

AN is a psychiatric condition characterized by marked weight loss, intense fear of weight gain, and profound disturbances in body image perception[30].

Diagnostic criteria include a body weight that is less than 85% of the expected norm and, in postmenopausal women, the absence of at least three consecutive menstrual cycles. The reduced body weight may result from restrictive eating patterns or from binge-eating and purging behaviors such as self-induced vomiting or the misuse of laxatives, diuretics, or enemas aimed at controlling weight. While many medical complications associated with AN tend to improve with weight restoration, some manifest as refeeding syndrome. In particular, within the Bulimic-Purging subtype (AN-BP) and BN, the most serious complications stem from purging behaviors, especially vomiting. These can lead to electrolyte and acid-base imbalances as well as localized complications, making the discontinuation of purging behaviors essential for effective medical treatment[31].

Historically, research on AN has largely overlooked eating behaviors, likely due to the assumption that such behaviors are minimal or absent. Future studies will be vital to clarify whether specific psychological and cognitive mechanisms contribute to the persistent dieting behavior that defines AN[30].

In Table 2, ChatGPT was asked questions by giving clues about AN[3], but ChatGPT's comments were insufficient for AN diagnosis. Even when a body mass index (BMI)-related clue was given, no accurate result could be obtained. However, there are few studies investigating AN and chatbot-ChatGPT applications. Egedal et al[32] emphasized that with the help of the natural language processing model, coding for both anorexia and bulimia achieved approximately 90% correct results. Another study found that when AI was used to assess AN, the results had significantly higher false negative rates for female samples than male samples[33]. These limitations may lead to problems such as false negatives and over-reliance on AI in the clinic.

BED

Binge eating describes the consumption of substantial amounts of food in short periods, accompanied by a subjective sense of loss of control. Binge eating is considered a diagnostic feature of BEDs, including the BN and AN-binge/vomiting subtype, while binge eating behaviors are highly prevalent in the general population. Engaging in such behaviors predicts future BED diagnoses and is associated with general psychopathology, decreased daily functioning, increased risk of death, and, therefore, significant disease burden[34].

Individuals with BED often experience shame related to their eating behaviors and tend to conceal their symptoms. Episodes of binge eating commonly take place in secrecy or with minimal visibility. Negative affect is the most frequent precursor to binge episodes. Additional triggers include interpersonal conflicts, restrictive dieting, dissatisfaction with body weight or shape, negative attitudes toward food, and boredom. Although binge eating may temporarily reduce or distract from these triggering factors, it frequently leads to delayed negative outcomes such as poor self-esteem and emotional distress[8].

BED is a prevalent ED closely linked to obesity, significant psychiatric and medical comorbidities, and considerable psychosocial impairment. While evidence-based treatments exist that effectively reduce binge eating episodes, a substantial proportion of patients do not achieve optimal outcomes. Approximately half of those undergoing leading BED therapies attain abstinence from binge eating, yet most treatments do not yield clinically meaningful weight loss[35].

According to the results of studies on AI and digital applications, Levinson et al[36] found that machine learning can detect a high rate of binge eating. In a study on web-based guided self-help applications, objective BED behavior decreased compared to the control group during 4 weeks of treatment, and 40% complete recovery was observed[37].

Yu et al[38], in their study comparing web-based application with face-to-face dietitian/psychotherapy treatment, found that uncontrolled eating and emotional eating behaviors and food addiction decreased in women with BED and positive effects on EDs were observed with the web-based application. However, in the web-based application, video counseling was applied via online telemedical software with psychotherapists and registered dietitians.

ChatGPT was asked to predict the disease by entering symptoms related to binge eating. It showed correct approaches in its predictions (Table 3)[3]. However, the presence of distinguishing features from BN in binge eating may have helped detect it correctly.

Pica

Pica is characterized by the persistent consumption of non-food, non-nutritive substances and is classified under Feeding and EDs[8]. Populations at increased risk for anemia—such as women of reproductive age, children, and pregnant individuals—experience heightened iron requirements due to physiological processes like menstruation, growth, and pregnancy. Pica involves a compulsive urge to ingest substances such as clay, soil, paper, or laundry starch. Without intervention, this behavior can result in serious, potentially life-threatening health complications. There is a notable association between Pica and iron deficiency anemia (IDA), with approximately 50% of patients with IDA reporting Pica symptoms. Although Pica occurs across all age groups, it is especially prevalent among pregnant women, who are particularly vulnerable to micronutrient deficiencies. Furthermore, households facing food insecurity and micronutrient shortages are at elevated risk for engaging in non-nutritive eating behaviors like Pica, which can increase susceptibility to systemic infections and adverse birth outcomes[39].

Pica is also disproportionately observed in children with autism spectrum disorder and developmental delays, with its incidence generally declining after the age of three[40]. Despite being one of the most common EDs in individuals with developmental disabilities, clinical presentations of Pica remain relatively uncommon[41].

There is currently no gold standard treatment for Pica. Current interventions for Pica include surgery, nutritional, pharmacologic, and behavioral therapies. Although surgical interventions for pica behaviors are indicated when perforations, obstructions, or extensive masses occur, they do not cure Pica and instead address associated symptomatology. Nutritional supplements, such as iron, have been used to treat Pica based on the etiological assumption that it is caused by nutrient deficiencies or excesses[42].

Table 4 includes questions about eating nonfood items, a diagnostic feature of Pica. The presence of differential factors in the diagnosis of pica disorder may have enabled ChatGPT to detect it more easily. Additionally, these results are a priori and based on limited, nonclinical interactions.

Rumination disorder

Rumination comes from the Latin word "ruminate", meaning to chew the cud. In humans, rumination is characterized by the regurgitation of food, usually immediately after a meal, when the food has not been adequately mixed with stomach acid. For this reason, patients often re-chew the regurgitated contents. Despite this apparent presentation, rumination is frequently misdiagnosed as gastroesophageal reflux or vomiting[43].

Rumination disorder is a problem that largely heals on its own. Adults generally have a more chronic progression. The preferred treatment methods for rumination disorder have not been determined. The underlying mechanisms that cause rumination behavior should be investigated comprehensively in treatment. Due to both low prevalence rates and a heterogeneous clinical appearance, information on treatment options has been limited to case series. Treatment options generally vary according to comorbid conditions and the severity of the disorder[41].

Table 5 shows ChatGPT questions about rumination disorder. ChatGPT correctly predicted this disorder[3]. The presence of distinctive features in rumination, similar to pica disorder, may have enabled the ChatGPT to provide information. As in pica disorder, it is an indicator of non-clinical and limited results in this disorder.

Avoidant/restrictive food intake disorder

Avoidant/restrictive food intake disorder (ARFID) is characterized by persistent food avoidance or dietary restriction without concern for weight or body image[44]. There must be three characteristics of conditions that may lead to disturbances in eating behaviors: (1) A marked lack of interest in eating; (2) Avoidance based on sensory properties of food; and (3) A concern about negative consequences of eating. It is important to note that this list is not mutually exclusive or intended to be exhaustive, and diagnostic guidelines recognize that other causal processes may underlie restrictive eating in ARFID. Instead, it is designed as a first step in disaggregating the variability in ARFID and understanding its underlying causes[45].

This ED is characterized by at least one of the following outcomes: significant weight loss (or, in pediatric cases, failure to achieve expected weight gain or impaired growth), notable malnutrition, reliance on oral nutritional supplements or enteral feeding, and considerable impairment in psychosocial functioning[36]. A distinguishing feature of ARFID, compared to other EDs such as AN and BN, is the absence of any disturbance in the individual's perception of body weight or shape. Consequently, a diagnosis of ARFID cannot be made concurrently with AN or BN. In contrast, BED may present with or without concerns related to weight or body shape, and it is possible for ARFID and BED to co-occur[46].

Table 6 above asks ChatGPT about avoidant restricted eating. ChatGPT could not predict this disorder without the differential information[3]. When additional information was provided, it correctly predicted it. According to the DSM-V, ARFID does not involve food insecurity, unavailability, or the avoidance or restriction of food intake associated with cultural practices such as religious fasting or the standard diet. This should also be questioned to make the correct diagnosis.

INTERPRETATION OF POSSIBLE MISLEADING SITUATIONS IN DIAGNOSING EDS IN DSM-V USING CHATGPT

Table 7 briefly shows the possible misleading situations in diagnosing EDs in DSM-V using ChatGPT. According to this table, AN-BN and BED, which are relatively more common than other EDs, are emphasized.

Table 7 Possible misleading situations in diagnosing eating disorders in DSM-V using Chat Generative Pre-trained Transformer.

Eating disorder type	Situations with potential for confusion according to criteria in DSM-V
AN	Whether or not there is significant weight loss can make it challenging to diagnose AN and atypical AN
BN	Vomiting or laxative use may occur with anorexia, which can make the diagnosis of BN difficult
	The frequency and duration of vomiting are essential in BN; otherwise, it is classified as OSFED. In this case, there is a high probability of confusion
AN-BN	Purging behaviors are seen in both anorexia and bulimia nervosa, which may lead to confusion of the diagnosis of purging disorder with AN or BN
BED	Binge eating disorder is classified as OSFED when less than once a week and/or for less than 3 months
	It is said that bipolar disorder and binge eating disorder occur together, which may cause binge eating symptoms to be ignored and treatment to be started late in people with bipolar disorder

AN: Anorexia Nervosa; BN: Bulimia Nervosa; BED: Binge eating disorder; OSFED: Other specified feeding and eating disorders.

The presence of mental retardation in rumination disorder means that the disease is severe. For this reason, this ED should also be evaluated with the mental retardation condition. In addition, the clinician must exclude gastrointestinal abnormalities, infections, and other medical conditions[8]. More generally, it is estimated that eating-related issues are prevalent in approximately one-third of people with intellectual disability/mental retardation and up to 80% of people with more severe impairment, problem behaviors, and comorbid autism spectrum disorder. Therefore, investigating these issues in families with children with autism spectrum disorder, for example, may be misleading[47].

CHALLENGES AND OPPORTUNITIES

Thin pressure, thin-ideal internalization, body dissatisfaction, dietary restraint, and increased BMI predict EDs, particularly AN and BN, with the perceived importance of body weight/shape among their prodromal symptoms. Additionally, negative emotionality, low parental social support, low peer social support, and pathological eating modeling (of family and peers) predict EDs, particularly BN and BED, because binge eating and purging are performed to cope with negative emotions and interpersonal problems[48].

There is little information in the literature on the development process of rule-based chatbots related to psychiatry[9]. Recently, a chatbot (TrueBalance) developed to support the treatment of EDs in young individuals has been found to have positive results. Expert opinions on this application have been positive, and it is particularly effective in detecting AN, BN, and BED. However, from the user's perspective, using TrueBalance with this robot has several challenges and possibilities. Challenges include maintaining user engagement over time, addressing the sensitive nature of content related to EDs, and ensuring the relevance and accuracy of the information provided. TrueBalance’s knowledge base, designed to address only specific user scenarios, has certain limitations. Ongoing collaboration with psychologists or therapists and integrating additional user feedback is necessary to increase the sensitivity and appropriateness of interactions[49].

Rule-based chatbots can reach large audiences at low cost by providing information and simple interactions. However, they have limited ability to understand user responses and respond appropriately. Workarounds can minimize user frustration by reducing conversational errors[9].

The process of building trust in chatbots can be further investigated. Studies have shown that user trust in a chatbot is not a fixed outcome after using the chatbot. This trust can change before, during, and after the interaction[50].

Apart from these, some ethical principles regarding the use of AI exist. These include contributing to society and human well-being and not causing harm. In addition, people who work with AI should be fair and non-discriminatory, honest and trustworthy, and respect the work required to produce new ideas, inventions, creative works, and computing products. Privacy and confidentiality should be protected[51].

Ensuring the accuracy of AI algorithms used for diagnosing psychiatric disorders is crucial to avoid potential harm to patients. Moreover, all stakeholders involved in the development, testing, implementation, and evaluation of AI technologies must be cognizant of ethical challenges, such as data bias. Established ethical frameworks guiding biomedical research should be upheld and, where necessary, strengthened to address the unique issues posed by AI applications. Bridging gaps in data and technology literacy among both patients and healthcare providers is essential for responsible adoption. Currently, there is a lack of standardized guidelines for the integration of AI and emerging technologies in clinical practice. Researchers may develop AI models based on incomplete or insufficient datasets, while clinicians might struggle to interpret complex AI-generated data or question its reliability. Additionally, institutional review boards often have limited expertise regarding novel technologies, resulting in inconsistent risk evaluations. To address these challenges, ethical principles must be embedded into AI development through dedicated research, education, and allocation of appropriate resources[52]. Recent literature has also underscored concerns regarding data privacy and patient confidentiality, especially within psychiatric care[53,54].

A recent study has highlighted that, although many clinicians and community participants are open to and accepting of the use of AI in ED treatment, many remain cautious and uncertain about its implementation, despite its potential far-reaching benefits[5]. The potential negative impact on the ED community of the public release and rapid withdrawal of the Tessa chatbot in just one week has highlighted the need for more comprehensive and reliable evidence than what has been gathered so far[55]. Xue et al[56] emphasized that ChatGPT may give gender, cultural, political or religious biased answers. An et al[2] used probabilistic inference models to ensure that each response pattern of chatbots was unique, and that a chatbot was likely to produce different answers when asked the same question twice, adding that this was a limitation of their study. They also emphasized that the failure of chatbots to provide reliable, professional answers to questions about body image was the tip of the iceberg, and that the technical, legal, and ethical challenges to the underlying issues were deep and long-standing. The trusting and emotional dependency of adolescents and young adults can make them vulnerable to carefully deployed, well-targeted manipulations, warranting close monitoring and vigilant protection[2].

AI holds significant promise for enhancing access to prevention and treatment services; however, concerns arise regarding whether ethical safeguards are advancing at a pace sufficient to protect users effectively. Responsibility for ensuring ethical and safe care lies with multidisciplinary teams. In the absence of comprehensive standards and best practice guidelines, all available measures should be employed to mitigate risks and minimize harm. While technologies may be delivered via computers and smartphones, it is essential to remember that real individuals seek assistance, and care must prioritize safety and ethical integrity[55].

To safeguard health data during AI utilization, specific regulatory frameworks must be adhered to. The Health Insurance Portability and Accountability Act of 1996 establishes data security requirements for healthcare entities to protect patients’ personal health information by maintaining confidentiality and security[57]. Similarly, the European Union’s General Data Protection Regulation enforces stringent rules governing data storage, protection, and breach notification, while also guaranteeing individuals’ rights to access information about their personal data[58]. Compliance with such regulations helps prevent ethical breaches in AI applications and healthcare robotics.

Moreover, implementing age restrictions and parental controls is crucial for AI applications, particularly concerning body image disorders that predominantly affect adolescents. It is also vital to recognize cross-cultural variations in EDs and body image issues. The global relevance of AI tools like ChatGPT is limited by the predominantly English and Western-centered nature of their training datasets. Consequently, the use of keywords in different languages may lead to misunderstandings and inaccurate recommendations, highlighting the need for culturally sensitive adaptations.

CONCLUSION

AI and chatbots such as ChatGPT are becoming increasingly widespread in health and psychiatry. Although some web-based applications provide advantages in terms of time and cost in the field of psychiatry, there are limited studies investigating the effects of chatbots such as ChatGPT on EDs and body image. Having similar diagnostic criteria for EDs may lead to problems in accurately detecting EDs through ChatGPT. It is vital to make a complete diagnosis and treat accordingly, especially in rarer EDs. ChatGPT can provide accurate results when symptoms are specified for some diseases. However, detailed interviews with individuals can be effective in considering accompanying diseases and determining the exact cause of the ED. Hybrid AI models (human-AI) or chatbots can be developed to overcome these barriers. Web-based applications and online human-assisted applications may also be a more useful strategy than using AI alone. More detailed studies are needed to fully determine the effects of ChatGPT and similar chatbots on body image and EDs. The long-term effects of AI and chatbots are unknown. Future research is recommended regarding the sustainable outcomes of AI-assisted ED interventions. Particularly cohort and clinical studies will increase reliability. In addition to these, more comprehensive and methodologically improved studies investigating the effects of ChatGPT and other chatbots are needed.

ACKNOWLEDGEMENTS

I would like to thank my beloved son Fatih Mete, who accompanied me while I was writing this article during the last period of my pregnancy, and who brought heaven to us with his presence at this moment.