Your Tweets Are Training Data: The Personal Data Problem AI Created Without Telling You

Most software developers think a lot about the data their apps collect. Very few think about the data they create themselves. For over ten years, people have been writing posts online. Now, writing has become one of the biggest sources of data used to train Artificial Intelligence (AI). \ Twitter/X is not just a place to talk. It is a main ingredient used to build Large Language Models (LLMs). Most people who wrote these posts never gave permission, never knew it was happening, and cannot take their words back. \ This is the problem of "personal data decay." It is an urgent issue that is much more important than most people realize. What "Public" Actually Meant for AI Training When researchers needed massive amounts of human writing to train early AI models, Twitter was the best tool available. The text was casual, full of feelings, and huge in volume. \ Early academic projects turned Twitter into the gold standard for understanding how people talk online. By the time big tech companies needed billions of words to train modern AI, Twitter data was already mixed into the foundation of the internet. \ According to a Mozilla Foundation study analyzing 47 AI models published between 2019 and 2023, at least 64% of those models were trained on data from Common Crawl. The same study found that Common Crawl made up more than 80% of the data used to train GPT-3. Because Common Crawl copies and saves public websites, years of public tweets were taken into these AI systems at a scale no one expected. \ Also, research published on ResearchGate about digital consent shows that this has changed the rules of personal data. People's personal posts have become involuntary donations to AI systems without their knowledge. \ Content originally shared for others to read is now being used for purposes many users never expected. Understanding Data Decay In data management, "data decay" happens when information becomes less accurate or less useful over time. \ For example, a customer database from five years ago may contain outdated addresses, job titles, or contact details. The information was correct when it was collected, but it no longer reflects reality. \ The same thing happens to your social media posts. \ A tweet from 2014 showed your thoughts, mood, and life at that exact moment. It was accurate for who you were back then. The problem is that the AI model does not know that time has passed. To an AI model, there is no timestamp on the meaning of your words. Your old writing patterns and old opinions are permanently baked into the AI's memory with no expiration date. \ Researchers call this "temporal data misrepresentation." The growing international discussion around data dysphoria and AI governance shows this worry, as government regulators try to fight back against data practices that ignore the individual. \ The challenge is that AI systems do not naturally understand this change. Older content can continue influencing models even when it no longer represents the person who originally wrote it. Why Deleting Posts Doesn't Fully Solve the Problem This is where the technical reality gets tough. \ Deleting a tweet from Twitter/X removes it from the live website. It does not remove it from an AI model that was already trained on it. It does not remove it from internet archives, copycat websites, or companies that saved it in the past. \ Scientists are trying to fix this with a new method called machine unlearning. This is a way to make an AI forget specific data points without having to rebuild the whole AI from scratch. However, the truth is that making an AI forget information at a large scale is currently impossible. AI models do not store data like a standard file cabinet; the information is completely blended and expensive to remove. \ This creates a problem for privacy laws like GDPR Article 17, which gives people in the EU the right to have their personal data deleted. \ The challenge is that these laws were written before modern AI systems existed. While a company can delete personal data from a website or database, removing that same data from an AI model is much more difficult. \ According to PrivacyEngine , GDPR enforcement had led to more than €2.8 billion in fines by 2023, mostly for issues involving user consent and data security. However, there is still no clear industry-wide solution for removing a person's data from AI models that have already been trained on it. What the Legal Battles Mean Twitter/X has started fighting back against the large-scale scraping that helped build many AI training datasets. \ In several legal cases , X Corp has argued that automated mass scraping of its platform should be treated as unauthorized access. In other words, the company is trying to regain control over user-generated content that was once widely collected and reused by third parties. \ These legal battles reflect a bigger change in how platforms view their data. Content that was once seen as freely available online is now considered a valuable business asset, especially as AI companies use massive amounts of data to train commercial models. \ The irony is that much of this data was collected years before platforms fully recognized its value. While companies are now trying to limit future scraping, that does little to affect data that has already been gathered and used in existing AI systems. \ The broader legal debate centers on a simple question: if content is public online, can companies scrape it and use it to train AI systems without the permission of the people who created it? The Best Response: Control Your Source Data Right now, the most practical form of control is managing the source data itself. \ If content is not publicly available, it cannot be collected for future training datasets. \ For people with years of public posts, reviewing old content can be a useful form of digital housekeeping. Posts that no longer reflect your views, professional standards, or personal identity may not be something you want permanently attached to your online presence. \ A practical approach includes: Download your archive first. \n Before deleting anything, download a copy of your Twitter/X archive. You can find it under Settings → Your Account → Download an archive of your data. Think of it as a personal backup of everything you've posted over the years. Store the file somewhere safe, as it contains a complete record of your activity on the platform. \ Review your posts by time period, not one by one. \n Most people look for specific posts they regret. A better approach is to review your account year by year. Ask yourself whether content from a particular period still reflects who you are today. If your opinions, professional focus, or communication style have changed significantly, that period may be worth removing as a whole. \ Use bulk deletion tools. \n If you've been posting for years, deleting tweets manually can take an enormous amount of time. Tools like TweetEraser let you remove posts in bulk based on date ranges, keywords, content type, or other filters. They can also help you manage replies and retweets separately, making it easier to clean up large amounts of old content without reviewing every post individually. \ Create rules for future posts. \n Some people automatically delete tweets after a set period, such as 90 days, unless they choose to keep them. This helps prevent years-old content from piling up and reduces the need for a major cleanup later. A simple policy like this can make it much easier to manage your public online presence over time. The Signal and Noise Problem There is an interesting parallel between AI models and people. \ For AI models, outdated training data can become a problem. Information that was accurate years ago may no longer be true today, which can lead to misleading answers. That is why AI developers spend so much time improving data quality and keeping information up to date. \ The same thing happens with personal social media histories. \ A decade of tweets often reflects several different stages of a person's life. Your views, interests, career, and communication style may have changed significantly over the years. Yet, someone reading your old posts — whether a person or an AI system — has no easy way to tell which posts still represent who you are today. \ Reviewing and cleaning up old content is not about rewriting history. It is about making sure your public profile accurately reflects the person you are now. Like any good data cleanup process, the goal is simple: remove what is outdated, keep what is still relevant, and make sure the information that remains tells the right story. Conclusion The relationship between personal data, AI training, and privacy rights is becoming one of the biggest challenges in modern technology. Yet, most people with years of public posts online do not realize they are part of the conversation. \ For developers, creators, and everyday users alike, the reality is simple: your public posts may already have been used to help train AI systems. The question now is whether your online history still reflects who you are today and how much of it you want to remain publicly available in the future. \ Managing your digital footprint is not about hiding the past. It is about deciding what information should continue to represent you going forward. \ Like any cleanup task, it is much easier to stay on top of it now than to deal with years of accumulated content later.