Digital Grief & AI MemorialsTechnology in Mourning

Grief used to have a place: a grave, a box of letters. Now it also has a feed, a chat thread and, increasingly, a chatbot. Brubaker, Hayes and Dourish (2013) documented the first shift, mourning on social media; Hollanek and Nowaczyk-Basińska (2024) mapped the newest one, AI simulations of the dead. Between them sits the thing most people actually have: a phone full of someone's messages.

Mourning moved online

Brubaker, Hayes and Dourish studied the Facebook profiles of people who had died and found that they turned into memorial sites without anyone deciding they should. Read the study. Friends kept posting, on birthdays and anniversaries and ordinary Tuesdays, often addressing the dead person directly. Two things changed: mourning stretched out in time instead of ending with the funeral, and it reached people at the edges of the person's life who would never have been in the room.

The same has happened, more privately, to messages. A chat thread with someone who has died is the most complete record of a relationship most of us will ever hold, and it stays exactly where it was.

Griefbots

The newest step is generative AI trained on a person's messages, voice or writing to produce a "deadbot" you can talk to. Hollanek and Nowaczyk-Basińska (2024) work through three scenarios of how this goes wrong, from a bot that starts advertising to the bereaved to a child kept "talking" to a dead parent. Read the paper. Their recommendations are concrete:

• Consent from the data donor, ideally while alive, and from the person who will interact with the bot.
• Meaningful transparency: the user always knows it is a simulation, and how it was built.
• Adults only, with particular care for children and for the bereaved in the acute phase.
• A retirement procedure: a dignified way to switch the bot off, rather than a subscription that lapses.

None of this is an argument against building them. It is a checklist for anyone considering one.

Rereading old messages

There is no research verdict on whether scrolling through a dead person's texts helps. Grief researchers talk about "continuing bonds", the ordinary and healthy ways people keep a relationship with someone who has died, and rereading messages is one of them. It becomes a problem when it stops being a visit: when you reread to argue, to find the moment it could have gone differently, or instead of sleeping.

One thing some people find useful is to look at the shape of the conversation rather than its lines: when it was warm, when it went quiet, how it changed over years. MosaicChats can produce that from an exported chat, a timeline of tone and of who reached out when. It is an unusual use of a relationship analyzer, and not for everyone, but it turns a thread you cannot finish reading into something you can hold whole.

What to do with the accounts and the chats

1. Export before you lose it. Phones get replaced and accounts close. Save the thread somewhere that is yours; our WhatsApp export guide is the place to start.
2. Do not decide anything for a while. Deleting, memorialising, printing, building a bot: none of it has to happen this month.
3. Check the platform's options. Most major services offer a memorialised state or a legacy contact. Set one for your own accounts now, and tell someone where things are.
4. If you are considering a griefbot, apply the checklist. Did they consent? Will you always know it is a simulation? Can you switch it off? Who owns their words afterwards?
5. Talk to a person too. The research on digital mourning is consistent that online spaces extend grief; they do not replace the people in it.

Frequently asked questions