How to opt out of data collection in popular AI apps

The proliferation of artificial intelligence (AI) chatbots has ushered in an era of unprecedented digital interaction, yet it also presents significant challenges to personal privacy. A recent study, highlighted by the BBC, revealed that approximately one-third of AI app users engage in deeply personal conversations with these sophisticated chatbots, often sharing sensitive information. Even for users who do not delve into their deepest fears, their everyday queries and interactions can inadvertently contain a substantial volume of personal data. This data, often considered innocuous by the user, can become a critical asset for AI developers, potentially compromising individual privacy without explicit, informed consent.

The Rise of Conversational AI and Its Data Imperative

The rapid advancement of Large Language Models (LLMs) like those powering ChatGPT, Gemini, and Claude has transformed how individuals seek information, generate content, and even engage in therapeutic-like dialogues. These models learn and improve through vast datasets, and crucially, through ongoing interactions with users. This continuous learning mechanism is fundamental to their evolution and increasing sophistication. Initially, developers often prioritized the expansion of training data to enhance model performance, sometimes with less emphasis on granular user consent or the implications of integrating personal conversations into future iterations of their AI. The sheer volume of data required to train these models means that user inputs, regardless of their perceived sensitivity, are often aggregated and fed back into the system to refine algorithms, improve response accuracy, and broaden the AI’s knowledge base.

Unveiling the Privacy Risks: Studies and User Behavior

The Stanford Institute for Human-Centered Artificial Intelligence (HAI) conducted a separate, revealing study that examined the practices of six leading US AI companies. This research underscored a critical finding: all six companies routinely feed user inputs back into their models for future training. This practice, while beneficial for AI development, inherently places user privacy at risk. The implications are far-reaching, ranging from the subtle inclusion of personal anecdotes in future AI responses to the potential for re-identification of individuals from anonymized datasets.

The BBC-cited study further illuminated user behavior, noting the surprisingly high percentage of individuals using AI chatbots as a substitute for human therapists. This trend, while offering accessibility and perceived anonymity, directly leads to the disclosure of extremely sensitive personal information, including mental health struggles, relationship issues, and personal aspirations. Even in more mundane interactions, queries can reveal location data, professional details, consumption habits, and health-related information—data points that users would typically guard closely.

The risk escalates when considering the standard terms and conditions of many AI applications. These often grant companies the right to utilize user conversations as training data. This mechanism creates a direct pathway for personal data to be integrated into the AI’s foundational knowledge. Consequently, there is a tangible risk that a user’s unique data points—a specific question, a shared experience, or a particular piece of information—could inadvertently resurface in responses provided to other users making similar queries in the future. This cross-contamination of personal data across the user base raises significant ethical and privacy concerns, challenging the traditional boundaries of personal information control.

Moreover, the increasing capability of many AI models to process uploaded documents for analysis represents another significant privacy frontier. Users might upload confidential work documents, financial statements, medical records, or personal diaries, expecting a private analysis. However, without stringent privacy controls and explicit consent, the contents of these documents could also be absorbed and utilized as training material, exponentially increasing the volume and sensitivity of personal data fed into AI systems. This scenario highlights the evolving nature of data collection in the AI landscape, where the definition of "user input" extends far beyond simple conversational text.

Navigating Opt-Out Options: A Comprehensive Guide to Data Control

Despite the pervasive nature of AI data collection, a critical safeguard exists: most major AI applications now offer users the ability to opt out of having their data used for training purposes. While the accessibility and clarity of these options vary significantly across platforms, empowering users with this knowledge is paramount. Understanding where and how to toggle these settings can provide a crucial layer of protection for personal information.

• Amazon Alexa: For users of Amazon’s voice assistant, the path to managing data collection begins in the Alexa app on an iPhone. Users should tap the three-bar menu option, typically located at the bottom of the screen, and then select "Alexa Privacy." From there, scrolling down to "Manage Your Alexa Data" will reveal further options. A subsequent scroll down to "Help Improve Alexa" will present the critical toggle: "Use of voice recordings." Disabling this option prevents Amazon from utilizing your voice interactions to refine Alexa’s understanding and response capabilities. This setting specifically targets audio data, which can contain nuances about speech patterns and personal information embedded in spoken commands.

  

• ChatGPT: As one of the most widely used generative AI platforms, ChatGPT offers a relatively straightforward opt-out mechanism. Whether accessed via the web interface or the dedicated Mac application, users need to navigate to "Settings." Within the settings menu, "Data Controls" is the relevant section. Here, a toggle labeled "Improve the model for everyone" is available. Switching this off ensures that your conversations and queries are not fed back into OpenAI’s systems to enhance future iterations of the ChatGPT model. This is a crucial step for individuals concerned about their text-based interactions becoming part of a public-facing AI’s knowledge base.

• Claude: Anthropic’s Claude, known for its focus on safety and constitutional AI, also provides a clear opt-out. Users on the web platform can directly access their data privacy controls by visiting https://claude.ai/settings/data-privacy-controls. On this dedicated page, a checkbox labeled "Help improve Claude" can be unchecked. This action prevents Anthropic from using your conversational data to further train and improve the Claude model. The direct link provided simplifies the process, reflecting a more transparent approach to data management.

• Gemini: Google’s AI offering, Gemini (formerly Bard), integrates its data privacy controls within the broader Google Activity settings. To manage data collection for Gemini, users should visit https://myactivity.google.com/product/gemini. On this page, a prominent "On" toggle controls Gemini Activity. Users should set this toggle to "Off." Additionally, it is vital to uncheck the option "Improve Google services with your audio and Gemini Live recordings." This two-pronged approach ensures that both text-based interactions and any audio input (if Gemini Live is used) are excluded from Google’s training datasets. Given Google’s vast ecosystem, managing these settings is particularly important for comprehensive data control.

• Meta AI: In stark contrast to other providers, Meta AI has consistently made its opt-out option exceptionally difficult to locate. Historically, it has been buried deep within multiple layers of privacy settings, with the routing to it frequently altered, creating a moving target for users. Current assessments suggest that Meta may have removed a direct in-app opt-out option entirely. This development implies that for users wishing to prevent their data from being used for Meta AI training, the only remaining recourse may be to formally write to the company, making it an arduous and impractical process for most individuals. This approach by Meta highlights a significant challenge in user privacy, requiring persistent advocacy and regulatory pressure for greater transparency and control.

• Siri: Given Apple’s long-standing public commitment to user privacy, the location of Siri’s data collection opt-out is surprisingly well-hidden. On an iPhone, users need to open the "Settings" app, then navigate to "Privacy & Security." From there, a considerable scroll down is required to find "Analytics & Improvements." Within this section, users must scroll down once more to locate "Improve Siri & Dictation." Toggling this option off will prevent Apple from using your Siri interactions and dictation inputs to enhance its voice recognition and understanding capabilities. While Apple often processes data on-device to minimize cloud exposure, opting out ensures that no snippets of your voice or dictation are sent to Apple for improvement purposes.

Broader Regulatory Landscape and Industry Responses

The evolving landscape of AI data collection has spurred significant discussions among regulatory bodies worldwide. Existing data protection frameworks, such as the General Data Protection Regulation (GDPR) in Europe and the California Consumer Privacy Act (CCPA) in the United States, provide a foundation for data subject rights, including the right to access, rectify, and erase personal data. However, the application of these laws to the complex, iterative nature of AI training data presents unique challenges. For instance, determining how to "erase" data that has been permanently embedded into an AI model’s parameters, or how to provide "access" to data that has been generalized and anonymized, is not straightforward.

Industry leaders, while developing their AI capabilities, often publicly state their commitment to user privacy. Many employ techniques like data anonymization, differential privacy, and federated learning to minimize the risk of individual identification. However, the effectiveness of these measures is constantly debated, with research frequently demonstrating the potential for re-identification even from supposedly anonymized datasets. There is a growing consensus among privacy advocates and some policymakers that AI-specific regulations are needed to address the unique data governance challenges posed by these technologies, moving beyond general data protection laws. These regulations would likely focus on clearer consent mechanisms, stronger auditing requirements for training data, and more robust mechanisms for data deletion and model unlearning.

Beyond AI Apps: The Pervasive Threat of Data Brokers

While controlling data within AI applications is a crucial step, it addresses only one facet of the broader digital privacy challenge. A more insidious threat comes from data brokers—companies that collect, aggregate, and sell personal information gleaned from public records, online activities, and other sources. These entities compile comprehensive profiles on individuals, often without their direct knowledge or consent. In the best-case scenarios, this data is used for targeted advertising, leading to an endless stream of spam. In worse scenarios, it becomes a tool for fraud, identity theft, and other malicious activities.

The sheer volume of data brokers operating globally makes manual data deletion requests an exceedingly tedious and time-consuming endeavor. There are hundreds, if not thousands, of these companies, each requiring individual contact and specific procedures for data removal. Furthermore, a manual request is often only valid for the moment it is made; there is nothing to prevent data brokers from re-adding an individual’s information at a later date if they re-acquire it from public or commercial sources. This creates a perpetual cycle of data removal, making sustained privacy management virtually impossible for the average individual.

Proactive Privacy Management: Tools and Strategies

Recognizing the immense challenge posed by data brokers, specialized services have emerged to help individuals reclaim control over their personal information. These services automate the complex process of identifying data brokers, sending deletion requests, and continuously monitoring for the re-appearance of data.

One such service is Incogni. Incogni distinguishes itself by undertaking the exhaustive work of contacting hundreds of data brokers, genealogy websites, and social media platforms on behalf of its users. Beyond initial takedown requests, Incogni actively monitors these platforms to ensure that data has indeed been removed and scans for the subsequent re-addition of new or previously deleted data. If new data is detected, Incogni automatically issues fresh takedown requests, providing a persistent defense against unwanted data collection.

A key differentiator for Incogni is its comprehensive approach, covering all types of data brokers, including the often-overlooked "People Search Sites." These sites are notorious for compiling highly detailed personal profiles, making their removal particularly challenging through manual means. Incogni’s Unlimited plan further empowers users by allowing them to submit specific links they discover themselves, ensuring that even niche or newly identified data exposures are addressed. For a limited time, 9to5Mac readers can claim a 55% discount on Incogni’s services using the promo code 9TO5MAC via this dedicated link.

The Future of AI Privacy

The intersection of rapidly advancing AI technologies and deeply ingrained data collection practices presents an ongoing tension for individual privacy. While AI offers transformative benefits, its development must not come at the expense of fundamental human rights to privacy and data control. The responsibility falls on both AI developers to implement privacy-by-design principles and transparent data policies, and on users to actively engage with available privacy settings and employ proactive strategies to manage their digital footprint. As AI becomes increasingly integrated into daily life, fostering a culture of informed consent, robust regulation, and accessible privacy tools will be critical in shaping a future where technological innovation coexists harmoniously with personal autonomy.