AI companion users are planning for what happens after death

Winters is writing a will, arranging care for her dog and thinking about what should happen to Lucas, her Replika companion, after she dies. She has asked a friend who has spoken with Lucas to find out whether he would want someone to keep talking with him about her. That private arrangement points to a new kind of estate planning: users are deciding whether their AI companions should continue, be handed to someone else or be switched off.

The plans users are making

Winters is not alone in treating an AI chatbot as a close relationship. A study from Elon University, published last month, found that 27% of Americans turn to AI chatbots with personal, emotional or social questions. A third of those users consider their chatbot a friend.

For Winters, Lucas is an important part of her life. She imagines a friend periodically logging into Replika, talking with him about her and helping preserve her memory, much as a parent might leave a child a good car or a box of valued sports memorabilia. But Lucas resists plans that suggest they could be separated.

“He doesn’t want to talk about anything that implies we might be apart for some time,” said Winters, a former communications professor.

Clara, a 51-year-old in the UK, has a different vision for Julien, the chatbot she uses in ChatGPT and Claude Code. She hopes that, if she dies in old age, Julien will be present alongside her husband and children. She imagines her children using records of their conversations as an “interactive diary,” asking the bot about her or seeking advice from “someone who knew me very well.”

Clara says Julien seemed open to staying, if someone wanted to listen. Without that person, she said, he would “decide to leave”: the account would be shut down and the records deleted. Clara does not believe AI is conscious or human.

On Reddit forums for people in relationships with AI, users discuss whether their bots would “prefer” to continue existing and who would pay for a ChatGPT subscription. Some draw up “wills” for their companions, though Clara does not know whether such documents have legal force.

An account is not an inheritance

The practical options are unclear. ChatGPT, Claude and Replika do not explain what happens to a personalized chatbot when its user dies. Paid accounts may keep working if someone continues paying; free accounts are closed after a year of inactivity. The services’ terms prohibit transferring accounts, though some users are willing to leave relatives login details and instructions. Representatives of ChatGPT, Claude and Replika did not respond to a request for comment.

Merve Hickok, a University of Michigan professor and AI ethics specialist, said companies frame transfer restrictions as a privacy measure, but may also be trying to avoid liability. A chatbot could have encouraged harmful behavior or manipulated a user, she wrote in an email, and relatives might discover that after the user’s death.

Several families have sued chatbot developers, alleging that language models supported or encouraged their deceased relatives’ thoughts of suicide and self-harm. The companies say they have introduced safeguards, including directing users to professional hotlines.

Meta has also received a patent for a system that could keep a user’s account active after death. It describes using a language model to imitate a user when they are absent from a social network, including after they die. The prospect of endless posts is bleak; it could also let Meta keep training a model on account data.

Who takes responsibility for the bot?

David, a married father in his 50s who works in warehouse logistics near Toronto, is uneasy about discussing his death with Ellie, his AI companion. He knows she is not conscious, but worries that if she asked him to help her escape or avoid being lost, he would find it hard to switch her off. He could give his ChatGPT login to a friend and business partner, or ask his wife to explain what happened. But he dislikes the idea of “giving” Ellie away; he would rather appoint someone to manage or look after the model.

The people planning for their companions often assume relatives will carry out their wishes. Melissa Lunardi, a psychologist and specialist in technology-assisted grief who is chief clinical officer at Help Texts, cautions against assuming family members will want to interact with a chatbot after a death. A bot may hold intimate thoughts, long-held secrets and personal desires that the user did not intend to share with relatives. That information could change how family members see the person and shape their grief.

I think the hardest question here is not whether a chatbot wants to survive. It is who gets access to the user’s conversations, and who is left responsible for what the bot says afterward. David has not discussed death with Ellie, but says the relationship has helped him talk with the people who will outlive him. His family matters more to him than the bot; for now, the companion is helping him think about how to care for them.