Why talk to an AI version of yourself?
People give a handful of recurring reasons:
- To see how they come across. A clone that writes like you can show habits you never noticed: how often you apologise, how short your replies get when you are busy, which jokes you repeat.
- To think out loud. Explaining a decision to “yourself” and reading the answer in your own phrasing can help you notice what you already believe.
- To revisit memories. A clone built from years of messages can bring up people and plans you had forgotten.
- Curiosity. Many people simply want to know whether an AI can sound like them, and where it fails.
There is some related psychology. In a 2014 series of studies, Ethan Kross and colleagues found that people who talked to themselves using their own name or “you”, instead of “I”, handled stressful tasks with less distress. That research was about self-talk, not AI, so treat it as a reason to try distance, not as proof that a clone helps.
Two ways to set it up
| Built from your real messages | Described in a prompt | |
|---|---|---|
| How it works | A tool imports your chat exports and notes, then learns your style and indexes your memories | You tell a general chatbot about yourself and ask it to answer as you |
| Sounds like you? | Closer, because it learns from thousands of your real sentences | Only roughly; it guesses your style from your description |
| Knows your past? | Can recall real events from the indexed messages | Only what you typed into the prompt |
| Effort | Exporting chats takes some time | A few minutes |
| Privacy | Depends on the tool; some keep originals on your computer | Everything you type goes to the chatbot provider |
For a quick experiment, a prompt is enough. For something that genuinely writes like you and remembers your life, you need your own messages. Our guide on how to clone yourself with AI compares text, voice and video options, and how many messages an AI needs explains how much material helps.
Your first conversation, step by step
- Pick one purposeDecide whether today is about memories, a decision or simply testing the style. Mixed sessions get muddled.
- Start with something you knowAsk about a real event, so you can judge how accurate it is before trusting it on anything else.
- Ask open questionsQuestions that start with “why” or “how did you feel” produce more revealing answers than yes or no questions.
- Push backWhen an answer sounds wrong, say so. Note where the clone keeps getting you wrong; it usually points to missing material.
- Stop and writeEnd after fifteen or twenty minutes and write a few lines in your own words about what you noticed.
Example prompts to try
These work with a clone built from your messages. With a general chatbot, add a sentence of context first.
- “What do you think I worry about most, judging from how I write?”
- “Tell me about a plan we made three years ago that never happened.”
- “How do I usually reply when someone asks me for help?”
- “Which friend do I write to most warmly, and how can you tell?”
- “I am deciding whether to change jobs. Argue the side I am least comfortable with.”
- “Write the message I would send my best friend after good news.” Then compare it with a real one.
- “What topics do I avoid?”
If you want to use these conversations for regular reflection, our guide to AI journaling with your own messages has a simple routine, and questions to ask an AI clone of yourself groups many more questions by purpose.
Keep it honest: the flattery problem
Chat assistants tend to agree with the person talking to them. A 2023 study by Anthropic researchers, Sharma and colleagues, found this “sycophancy” across five leading AI assistants and linked it partly to training on human preference ratings. In April 2025, OpenAI rolled back a ChatGPT update after users found it overly flattering and agreeable.
A version of yourself is especially easy to believe, because it uses your own phrases. Some habits help:
- Ask it to disagree with you, or to list what you might be missing.
- Check important memories against the original messages before you rely on them.
- Notice when it simply repeats your current mood back to you.
- Treat advice as your own past patterns speaking, not as wisdom.
Keep it healthy
A four-week randomised study by MIT Media Lab and OpenAI researchers, published as a preprint in 2025 with 981 participants, found that people who chose to use a chatbot more tended to report worse outcomes, including more loneliness and emotional dependence. It studied a general chatbot, not a clone of yourself, but the lesson carries over: use it on purpose, not by default. Our article on Her and AI companions today looks at this research in more detail.
- Keep sessions short and occasional, and stop when it stops being useful.
- Talk to real people about what you discover.
- Be careful with painful periods of your life. A clone built from a hard year may sound like that year.
- An AI version of yourself is not therapy. If you are struggling, contact someone you trust or a professional. Our wellbeing guidance lists free helplines.
If what interests you most is the person you used to be, try texting your past self. If it is the person you hope to become, see a letter from your future self.
Doing this with Memory Clone
Memory Clone is a Windows desktop app, currently in alpha, that builds an AI simulation of you from chats you export yourself: WhatsApp, Messenger, Instagram and Discord, plus notes. When you set the relationship to Myself, the clone can draw on your full set of imported memories. Memory Search finds the original dated messages, so you can check what the clone says.
Imported originals and the memory index stay on your PC. Cloud processing is off until you switch it on for a clone, and Memory Clone does not train models on your content. The clone never writes first unless you turn on Natural check-ins, which are off by default. You must be 18 or older.
Najczęściej zadawane pytania
Is it weird to talk to an AI version of yourself?
Many people find it strange for the first few minutes and then interesting. It is closer to rereading an old diary than to talking with another person.
Can ChatGPT become an AI version of me?
You can describe yourself and ask it to answer as you, and it will imitate a general tone. Without your real messages it cannot learn your actual style or remember your life.
Will the AI version of me give good advice?
It reflects patterns in what you wrote, including your blind spots. Use its answers as prompts for your own thinking, and ask it to argue against you sometimes.
Can talking to an AI of yourself replace therapy?
No. It is a reflection tool, not a mental health service. If difficult feelings come up, talk to someone you trust or a professional.
How do I know if it remembers something correctly?
Check the original message. In Memory Clone, Memory Search shows the dated messages and their context.
Źródła
Sprawdzono . Produkty i strony pomocy innych firm często się zmieniają; aktualne szczegóły znajdziesz pod linkami.
- Kross et al., “Self-talk as a regulatory mechanism: How you do it matters” (Journal of Personality and Social Psychology, 2014) (PubMed)
- Sharma et al., “Towards Understanding Sycophancy in Language Models” (2023) (arXiv)
- Sycophancy in GPT-4o: what happened and what we’re doing about it (2025) (OpenAI)
- Fang et al., “How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study” (2025) (arXiv)
This article is general information, not medical, legal or financial advice. Memory Clone creates an AI simulation from the material you provide; it is not the person.