“Based on everything you know about me, tell me something about myself that I may not realise.”
Questions about ourselves have always attracted attention online. Long before conversational AI, social media quizzes promised to tell us “What bird are you?”, “Which character matches your personality?” or “What does your choice say about you?” We answered a fixed set of questions, received a result and usually treated it as fun.
Trending AI prompts may look similar, but there is an important difference. When we ask AI, “Based on everything you know about me…”, the response may draw on more than the prompt in front of us. Depending on the service and its features, the response may also be informed by context from our previous interactions.
The response can feel surprisingly personal, leaving us wondering, “How does AI know so much about me?” But perhaps there is another question we should ask first: “How much have I already told it?”
We No Longer Prompt AI Like a Search Engine
A traditional internet search often requires only a few words: “weather tomorrow,” “restaurants near me” or “JavaScript date format.” Talking to AI is different.
We explain situations and provide background. We paste emails and ask for better wording, upload documents for summaries, discuss workplace problems, brainstorm business ideas, plan travel and think through difficult decisions. Sometimes we talk about relationships, frustrations, ambitions and fears.
This happens partly because context makes AI useful. The more relevant information it has about a situation, the better positioned it is to provide a useful response. One conversation may reveal very little about us, but hundreds of conversations are different.
Imagine someone who regularly uses an AI assistant. Across separate chats, they mention preparing for an interview, ask for help replying to their manager, discuss a project deadline, plan a family holiday and repeatedly ask for help improving their writing. Each conversation has an ordinary purpose. Together, they may reveal a profession, family circumstances, interests, responsibilities, communication preferences and current priorities.
There was never a moment when the person deliberately created a profile and said, “Here is information about my life.” They simply had conversations.
Why Sharing with AI Can Feel Different
People naturally adjust what they say depending on who is listening. We may hesitate before discussing a personal fear with a colleague or describing a workplace problem to someone we barely know. Human conversations come with reactions, expectations and possible consequences.
AI changes some of those conditions. It does not visibly become impatient when we explain something again or appear shocked by an awkward question. We can rewrite what we want to say, abandon the conversation or begin again. For some people, that can make the interaction feel easier or less socially exposing.
Research into online communication has long examined how people sometimes disclose differently through technology than they do face-to-face. Conversational AI introduces another variation because the interaction can feel highly personal even though it is taking place through a technology service.
This creates an important distinction: a conversation can feel private without necessarily being private in the way we imagine.
When AI remembers context, adapts to our preferred style or refers to something discussed earlier, the interaction can also become more familiar. That sense of conversational ease may influence what some people are willing to share, allowing details to accumulate over time.
The Personal Information That Does Not Look Personal
When we hear personal information, we usually think about obvious identifiers such as our name, phone number, email address, home address, passwords or financial information. Those certainly matter, but conversations can reveal something beyond identifiers.
Consider someone saying, “I have an important meeting with my manager tomorrow,” “My daughter is starting college next year,” or “I’ve rewritten this resignation email five times.” None contains a password or bank account number, yet each reveals something about the person behind the screen.
Across many conversations, AI may receive information about our routines, relationships, preferences, responsibilities, recurring concerns, professional circumstances and ways of making decisions.
This is why the phrase “everything you know about me” deserves more attention than it usually receives. The information may never have arrived as a profile; it may have arrived as fragments.
What We Share Is Not the Same as What AI Infers
Trending prompts introduce another layer. AI not only repeats information we have provided; it can also generate interpretations from it.
Suppose someone asks, “Based on our conversations, what is one weakness I don’t recognise in myself?” AI might respond: “You appear highly independent, but you may seek reassurance when decisions carry emotional consequences.”
The answer may feel remarkably insightful. But where did that conclusion come from?
Perhaps several conversations genuinely support it. Perhaps the model identified a recurring pattern. Perhaps it over-interpreted a small number of examples. Or perhaps it produced a plausible description that happens to resonate with the person reading it.
Research suggests language models can identify some personality-related signals from written language, but that does not mean every generated interpretation is accurate or that AI has discovered a hidden truth about an individual. Once a description feels accurate, we may also begin noticing experiences that support it while giving less attention to those that do not.
Inference is not knowledge. A convincing description of us is not automatically an accurate description of who we are.
AI may identify patterns in the information available to it. That is different from having privileged access to who we fundamentally are.
When an AI Response Becomes a Public Conclusion
There is another side to trending prompts that begins after the answer is generated. People post screenshots, discuss responses in videos and react with statements such as “This is exactly me,” “AI understood something I never realised,” or “Everyone should try this prompt.”
At that point, a private AI interaction becomes public content.
Imagine an influencer trying a trending prompt, finding the answer compelling and sharing it with thousands of followers. People who trust that person may become curious enough to try the same prompt themselves.
Greater care may be needed when the response is shared by someone the audience already regards as knowledgeable about human behaviour, counselling, mental well-being or a related field. Their professional authority may affect how seriously some viewers interpret what is being shared. But that authority belongs to the person — not automatically to the AI-generated conclusion.
An AI-generated interpretation does not become a professional assessment simply because a professional shares it.
Someone watching may think, “If AI discovered something meaningful about that person, perhaps it can do the same for me.” They copy the prompt, receive a personalised response and find that part of it feels accurate. Because one part resonates, the broader interpretation may begin to feel more credible.
A trending prompt can therefore travel beyond its original wording: AI generates an interpretation, someone gives it public visibility, others attach meaning to it, and more people repeat the interaction. The response itself has not become more reliable simply because it has become popular.
There is an important difference between using an AI response as something to reflect on and treating it as something established about who we are.
From Personal Data to Personal Meaning
Most conversations about AI privacy focus on data: what information did we provide, was it stored, how long was it retained, and could it be used to improve the model?
Those are important questions, but trending prompts introduce another one. They often ask AI to turn personal information into personal meaning: “Who am I?”, “What is my biggest weakness?”, “What pattern am I repeating?” or “What don’t I understand about myself?”
The question is therefore no longer only how much information AI has about us, but also how much authority we give AI to interpret that information. When an interpretation is shared publicly by someone influential, we must also consider how much authority others may give it.
These are three different stages: what we share, what AI infers, and what significance we give the inference. Understanding the difference between them may become an increasingly important part of interacting with AI.
The Personalisation Trade-Off
Personalisation itself is not the problem. It is one of the reasons conversational AI can be genuinely useful.
If an assistant understands our level of experience in a particular field, it can avoid explanations we may not need. If it understands an ongoing project, we do not have to repeat the background every time. If it knows how we prefer something to be written, it can adapt its suggestions.
We provide context and receive a more relevant response. If that relevance makes the interaction feel more useful or personally attuned, we may become more comfortable providing context again. Across repeated conversations, information can accumulate without any single interaction feeling particularly revealing.
The trade-off is simple but easy to overlook: personalisation needs context, and context often contains information about us.
But Is All of This Information Stored?
There is no universal answer because AI services handle information differently. More importantly, several ideas are often treated as though they mean the same thing when they do not. Even the word “stored” can mean different things depending on what we are actually asking about.
Conversation history: Can we see an old conversation in our account?
AI memory: Can AI use certain information about us in future conversations?
Data retention: Does the company keep some or all of the information on its systems, and for how long?
Model improvement or training: Can the information we provide be used to improve AI models?
These distinctions matter. For example, deleting something from our chat history does not automatically tell us everything about how the service handles data retention. Similarly, AI remembering a preference is not the same as saying that the entire conversation was used for model training.
Understanding these differences gives us a clearer picture of what may happen to our information after we share it. Instead of asking only, “Does AI store my prompts?”, we can ask more specific questions about what is visible in our history, what AI may remember, what the service retains and whether the information may be used for model improvement.
For example, ChatGPT provides controls relating to memory, chat history, model improvement and Temporary Chat. OpenAI states that Temporary Chats do not appear in history, do not create memories and are not used to improve its models, although copies may be retained for a limited period for safety purposes. Google provides its own controls for Gemini activity, retention and model improvement. Other AI providers have different policies, and business or enterprise services may operate differently from consumer accounts.
These policies can also change, which is why understanding the distinction between these concepts may be more useful than memorising a particular company’s current retention period.
Rather than memorising a particular company’s retention period, it is more useful to ask four questions:
- What can AI use during this conversation?
- What can it remember for future conversations?
- What information does the service retain?
- Could that information be used for model improvement?
Knowing the difference gives us a clearer picture than simply asking, “Does AI store my prompts?”
A Trending Prompt Is Still a Request for Information
A prompt can feel harmless when thousands of people are trying it, but popularity tells us nothing about how much information it may encourage us to provide.
Some trends ask AI to analyse existing conversational context. Others invite people to answer increasingly personal questions or upload photographs, screenshots, résumés, conversations or documents for deeper analysis. The more personal the resulting answer becomes, the more impressive the experience may feel.
Before trying a trending prompt, perhaps the first question should not be “What will AI tell me?” but “What will I need to tell AI for this prompt to work?”
Not Every Detail Improves the Answer
Useful context and unnecessary identification are not the same thing. If we want help understanding a workplace disagreement, AI may need to know what happened, but it probably does not need the real names of everyone involved. If we want an email improved, the message may be necessary, while the recipient’s phone number, email address and signature may not be. Confidential information may sometimes be removed from a document before it is uploaded.
The objective is not to become afraid of sharing anything with AI. That would remove much of what makes conversational systems useful. A better habit is: give AI the context it needs, not automatically all the context we have.
Sometimes the Information Is Not Ours
There is one more boundary that is easy to overlook. When we paste a conversation with a colleague, upload someone’s résumé, share customer details or describe another person’s private situation, some of the information inside our prompt belongs to someone else.
We may be comfortable discussing our own lives with AI, but the other person has not necessarily made the same choice. The question then changes from “Am I comfortable sharing this?” to “Is this mine to share?”
As AI becomes part of everyday work and communication, that distinction may become increasingly important.
Before We Press Enter
AI becomes more useful when it understands context. That is not a flaw; it is one of the reasons conversational systems are powerful. The challenge is that context accumulates quietly.
We rarely experience it as “giving AI personal data.” We experience it as asking for help with an email, thinking through a decision, solving a problem, discussing an idea or trying a prompt everyone is sharing online. Then one day we type, “Based on everything you know about me…” and the answer surprises us.
Perhaps the surprise itself tells us something important about Human-AI Interaction.
Learning to use AI well cannot only mean learning how to write better prompts. It may also mean understanding what we share, recognising the difference between an AI inference and a fact about ourselves, and deciding how much significance we should give to an answer simply because it feels personal.
So the next time we encounter a trending prompt asking AI to tell us something we do not know about ourselves, there may be one question worth asking before pressing Enter:
“What have I shared for you to know this much about me?”