Your Son Is Not the Person Who Will Open Your Persona

Here are the numbers, because the sentiment on this subject is thick and the evidence is thin.

• Most families will value an AI version of you and open one rarely. Stated value runs very high. In a University of Toronto survey of 41 bereaved adults presented at CHI 2010, 82% agreed that reminders of the dead were important, and 78% said they treasured mementoes.

• Wanted frequency runs low. In the same survey, 73% said they already had enough time to reminisce, 36% wanted more opportunities to do so, and 14% thought they should be thinking about their dead relative more often than they already did.

• Voice is the strongest material and the scarcest. Asked what they most valued when remembering, respondents put photographs top at 92% and voice recordings bottom at 29%, level with music. The researchers put that gap down to supply rather than worth: voice recordings, unlike photographs, are rarely made for remembrance, so far fewer people have one to name.

• The relationship carries on after the death. 53% said they had tried to keep a connection to their dead relative alive, which is a stronger claim than remembering somebody.

• Doing nothing loses the record. Physical objects were usually inherited by a family member, while non-physical digital assets commonly went missing or were destroyed outright. 65% of respondents had never thought about how their own devices should be handled after their death, 80% had made no plans, and 56% said they were concerned about the question anyway.

• No usage data exists for AI Personas. The category is too young for a dataset, so anyone showing you a usage graph is showing you a number they invented or a number measured over eighteen months on people who self-selected. Everything above is adjacent evidence about photographs, voicemails and inherited devices, drawn from a 41-person, largely North American sample. Treat the percentages as a shape, not a forecast.

High value. Low frequency. Nobody in those numbers is failing at grief. That gap is what holding something precious looks like when you are also busy being alive (Massimi and Baecker, CHI 2010).

Your house is already full of objects that behave exactly this way. A box of letters in the loft that gets opened twice a decade. A wedding video nobody has put on in years and nobody would let you throw out. A parent's handwriting on the inside cover of a recipe book, which no one reads, and which somebody would be in the car within the hour to rescue from a house fire.

None of those are judged on engagement. You would find the question offensive if someone asked how many times per quarter you open the letters before deciding whether to keep the letters.

So when you say nobody would use this, check which claim you are making. If you mean they will not talk to your Persona every week, you are almost certainly right, and weekly was never the deal. If you mean your Persona will be worthless to them, that is a far bigger claim, and the evidence runs the other way.

There is one careful account of the rhythm from the inside. In the CHI 2020 ReFind study, a research team built a handheld object that draws on an archive of 63 tagged photographs of the first author's dead mother, and the first author lived with the piece for ten weeks and kept a diary of every interaction. She looked at the object every day. She sent images to it only twelve times across those ten weeks, seven of those simply as a way of checking in, and she writes that she did not want to force her use of the object and so sent images only when she felt a genuine personal motivation. One person, and she says so plainly. Constant low-grade presence, real contact now and then, never on a schedule.

People keep recordings of a dead person's voice even when the recording is banal and was never meant to be kept. The recording does not have to be good. The recording has to exist.

The Toronto study turned this up almost as a side effect. Photographs, the researchers noted, are taken on purpose and are explicitly meant for remembering. Voice recordings are not. They exist incidentally, as voicemails and functional messages, and people came across them without meaning to. The paper reports that these recordings conjured up memories very strongly and took participants by surprise with how powerfully they did so.

One participant described her husband keeping his own father's answering machine message going, live on the line, for a long time. She said they kept the message "until it got too creepy". A functional greeting stayed live on a phone line well past the point of comfort, in a house where somebody had already named the discomfort out loud, and the study does not record why nobody deleted the message sooner.

Another participant in the same study said the thing she missed most was not having a recording of her mother's voice at all.

Photographs top at 92%, voice recordings bottom at 29%, and the researchers' own reading of the gap is about supply. Voice is the most powerful material in that paper and the scarcest, for the same reason. Voice only ever got captured by accident.

Most families have three or four generations documented in photographs and not one voice among them.

So the comparison your worry is running is the wrong one. You are setting a Persona built from your own answers, in your own voice, against a hypothetical family who talk to a dead parent every day. The real comparison is against what the Toronto participants actually described keeping: a short, accidental, entirely unremarkable recording that no one in the house could bring themselves to delete.

Very little of this is really about logins.

The worry underneath is that your children do not need you as much as you would like them to, and that a quiet dashboard would be the proof. "They wouldn't use it" is a much easier sentence to say than "I am not sure I am as central to their lives as they are to mine." I have caught myself reaching for the easier sentence too.

Two things about that fear. The first is that it is partly true, and it is supposed to be. An adult child who can go three weeks without ringing you is an adult child who is all right. That is the outcome you spent decades working towards. Low usage while you are alive is not a verdict on your worth. It is a report on your success, and every functional adult child files that report.

The second is that the situation changes. Bereavement researchers have a name for the pattern in that 53% who said they had tried to keep a connection to their dead relative alive: continuing bonds, meaning people do not detach from the dead and file them away, they carry the relationship on in some altered form. The ReFind team narrow the idea further to ongoingness, the parts of that bond that are live and connected to what is happening in a person's life right now.

Which is why the man at the kitchen table is the wrong man to picture. Your son at fifty-four, three months after your funeral, working out how to talk to his own child about something you once said to him, is a different person, and he is the one holding the login. If you want a sense of the questions that arrive at that age rather than this one, what your child will want to ask you at thirty sets out the ones that turn up late. The audience for this is a version of your family that does not exist yet, and you are asking the current version to vote on their behalf.

You cannot make anyone log in. Two inputs plausibly move the odds, and one popular test moves nothing at all.

Specifics, not general truths. A Persona built on the sentence "I always believed in working hard" is not worth opening. A Persona that knows the name of the foreman at the yard, what the family argument was actually about, why you left the second job, and exactly how you handle the stew, is worth opening, because there is no other route to any of that. Specificity is the whole difference between a greetings card and an answer nobody else alive can give. If you think you hold no such specifics, the things only you know how to do is the page for that, because the practical knowledge people assume is ordinary is usually the material nobody else can supply.

An occasion. People do not open things for no reason. They open them when a reason arrives: a first child, a diagnosis, an anniversary, a stupid argument, a house move, a recipe nobody wrote down, a question about what you were like at their age. Occasions cannot be manufactured. What you can do is make sure the answer already exists when one turns up. Families who do not talk much have the widest gap between what is known and what has been written down, and in those families the questions are more likely to arrive after the funeral. That last part is reasoning rather than research, and it is the situation this work is built for.

And the test to ignore: asking them in advance whether they would use this. Asking costs nothing and the answer tells you very little, because you are asking a twenty-six-year-old to predict the behaviour of a fifty-four-year-old they have not met. Ask anyway if it settles your nerves. Do not let a shrug close the question.

A three-minute test you can run this week with your phone

Open the voice memo app on your phone. Record three minutes answering one question only you can answer. Not "what advice would you give". Something with a proper noun in the answer. What the street was called and who lived at the far end of the road. What you actually thought in the week after your first child was born, including the parts you did not say at the time. How you deal with the one relative everybody finds difficult. What you would say if they rang you at two in the morning about a job.

Then email the file to yourself with a subject line the person would actually search for.

That is the whole test. Three minutes, no money, and it will tell you more than another fortnight of thinking. If it goes well and you want more prompts of the same kind, the method for interviewing yourself properly is written up separately, built on the same rule: questions with proper nouns in the answers.

Start building your legacy Free build, 50 memories, no card required.

Some people do not want this and never will. That is a legitimate position rather than a failure of imagination on their part, and any honest version of this product has to let them decline. I would rather concede that at full volume than argue anybody out of a reasonable objection.

The ethics literature is blunt about the version that goes wrong. In a 2024 Philosophy & Technology paper, two Cambridge researchers set out a speculative scenario, which they state plainly is design fiction rather than a real case, in which a man in palliative care designates his two adult children as the recipients of his deadbot without seeking their permission. One of them wants no part in the thing and is buried in reminders and notifications. The other tries to have the service deactivated and is refused, because it was the father, not the children, who had prepaid a twenty-year subscription.

Their recommendation is a design principle they call mutual consent: the people on the receiving end should give explicit consent before they are ever introduced to the service, and should be able to decline at any point afterwards.

Tell your people. Ask them first. A Persona sprung on somebody unannounced is the one version of this with a serious argument against, and any company that ships the surprise version will deserve the backlash coming for them.

Telling them also hands you something useful. Their reaction now, their questions, and often a list of things they would want asked that you would never have thought of on your own.

Suppose you are right. Your children open your Persona once out of curiosity, find the whole business strange, and never go back. That cost you a handful of evenings and an afternoon of feeling silly. You have lost more than that on a course you did not finish.

Now suppose you are wrong. Your daughter is fifty-one. You have been dead nine years. Something happens that she wants to tell you about, or ask you about, and there is a version of you built from your own answers in your own voice that will take the question and answer the way you actually answered things.

Six wasted evenings sit on one side of that. On the other sits the only route to something that otherwise does not exist at all.

The default is not neutral either. Concerned and unprepared is the standard position: 65% had never thought about how their devices should be handled, 80% had made no plans, 56% were worried about the question regardless. Physical objects got inherited. Digital ones went missing or were destroyed. That is not a story about carelessness. That is how a voice goes missing.

<blockquote class="quotable"> Frequency is the wrong measure for anything kept from a person who has died. A box of letters in the loft gets opened twice a decade and would still be carried out of a house fire. An AI version of a parent gets judged the same way. What counts is whether the answer is there on the day somebody finally asks. </blockquote>

The mechanics at Afterlife.ai™ assume usage will be occasional and that consent runs in both directions.

A guide called Idy interviews you, rather than handing you a form or a blank recording screen. Idy asks a question, listens to the answer, and follows the answer, which is what pulls out the foreman at the yard instead of "I always believed in working hard". Those specifics are what get opened years later.

Your answers become memories. Your Persona answers from those memories, in your cloned voice. The rule the system enforces is that concrete facts about your life have to match something you actually recorded, so your Persona never invents what you did not say.

You nominate the people who get access. That nominated list is built while you are alive, which makes the building of the list the moment you tell them, and the moment they can say no.

Nothing releases while you are alive. Access opens only after Executor Lock™, a seven-day, evidence-verified process. A claim that somebody has died has to survive a week of scrutiny, backed by documents, before anything opens. At the Lock, your Persona is snapshotted whole: nothing pruned, nothing down-weighted. The version your family reaches in 2049 is the version that existed on the day. If you want the step-by-step of the building itself, how to create an AI version of yourself walks through the process, and if accuracy is your real question, what happens when your Persona gets something wrong covers how errors are handled.

Now the limits, from the people who built the thing. We cannot make anyone open your Persona, and we have deliberately built nothing that nags them into doing so, because a grieving person being prompted by a dead parent's account is the exact failure this page argues against. Your nominated list is only as accurate as the contact details you keep current, so a list made once and never revisited is the likeliest way for all of this to reach nobody. And your Persona only knows the ground you covered in the interviews, so a question you never answered will not produce an invented answer, and will not produce a real one either. No feature fixes any of that. Answering more questions does, and so does checking the nominated list once a year.

You will not settle this argument by thinking harder about your son. Build 50 memories, free, with no card taken, and let the answers tell you whether there is anything here worth keeping. That free build does not expire, so you can leave the questions half answered for a year and come back when an occasion arrives. If you later want to know what the paid plans add, you can compare what each plan includes, though nothing on this page depends on that.

Start building your legacy Free build, 50 memories, no card required.

Is there any data on how often families use these things?

No, and be suspicious of anyone who says otherwise. The product category is too young for durable usage data, and the numbers that do circulate come from short windows and self-selected users. The honest evidence is adjacent: research on what bereaved people do with photographs, voicemails and inherited devices, where high stated value sits alongside low reported frequency of use.

What if my kids find the whole idea creepy?

Some people do, and that is a real answer rather than a reflex to be argued out of. The 2024 Cambridge ethics paper on this technology recommends a principle of mutual consent: the people who would receive a Persona should agree before being introduced to one, and should be able to decline afterwards. Tell your family what you are building. If somebody says no, they should be able to keep saying no.

Should I tell them I am doing this, or leave it as a surprise?

Tell them. A surprise is the one version the ethics literature actively warns against. Telling them also gives you their reaction now, their questions, and often a list of things they would want asked that you would never have thought of yourself.

What does this cost, and is it worth doing if they only open my Persona twice?

You can build from 50 memories for nothing, with no card taken, and that free build does not expire, so the interview costs you time rather than money. On the second half of the question: weigh the cost against the outcome where they do want this, rather than against the outcome where they shrug. A few evenings, set against what may be the only recording of your voice anyone ever made on purpose, is not a close call.

Who owns my Persona, and can the whole thing be deleted?

You own your Persona while you are alive. Deleting your account removes everything held under the account in one cascade: the memories, the recordings, the uploaded media and the login itself. One honest caveat, because this is the kind of thing people discover at the worst moment: if you have already acted as an executor in somebody else's Executor Lock, your own deletion is blocked until that obligation is discharged. After Executor Lock, the nominated people have access under the terms you set. Nobody buys, sells or licenses your voice. If a nominated person wants no part in this, nothing obliges them to take part.

What happens if the company shuts down?

No company can promise to exist in 2049, and any that does is overpromising. What matters is what you can get out and how much notice you get, so ask that of any provider before you start, including this one. Ask what the export contains and in what format, because a legacy you cannot take with you is a legacy that depends on a business surviving.

• I am not important enough to leave anything behind. If the deeper doubt is whether your life warrants any of this, start there rather than here.

• Isn't a Persona morbid?. For the reader whose family reaction is less "I wouldn't use that" and more "why are you thinking about dying". The social embarrassment version of this objection lives there too.

• Am I too young to build a Persona?. The timing objection, which usually arrives attached to this one.

• What would I even say to an AI of myself?. If you are past the objection and stuck on the blank page, this is the next step.

• The ethics of AI Personas of the dead. The difference between building your own while alive and someone reconstructing you afterwards, which is where most of the criticism of this technology is aimed.

Every figure here comes from peer-reviewed conference papers and journal articles read in full, not from press coverage of them: two CHI papers on bereavement and inherited objects, one autobiographical design study, and one 2024 ethics paper on the digital afterlife industry. All four are listed below with DOIs so you can check the numbers against the source.

What is not settled: nobody has usage data on AI Personas of family members, because the category is too young. The percentages here describe how bereaved people treat photographs, voicemails and inherited devices, from a 41-person, largely North American sample recruited through Craigslist, snowball sampling and newspaper adverts, of whom 40 were North American. That is adjacent evidence, not proof of how your own family would behave, and this page does not claim otherwise.

Found something here that is wrong or out of date? Tell us through the contact page and we will correct the page.

1. Massimi, M. and Baecker, R. M. (2010). A death in the family: opportunities for designing technologies for the bereaved. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI '10), 10 April 2010. University of Toronto. DOI 10.1145/1753326.1753600. Full text PDF: dgp.toronto.edu. Survey n=41 complete responses plus 10 follow-up interviews; all participants had lost a family member within the previous five years. Source of every percentage quoted on this page and of the answering machine account.

2. Wallace, J., Montague, K., Duncan, T., Carvalho, L., Koulidou, N., Mahoney, J., Morrissey, K., Craig, C., Groot, L., Lawson, S., Olivier, P., Trueman, J. and Fisher, H. (2020). ReFind: design, lived experience and ongoingness in bereavement. CHI '20. DOI 10.1145/3313831.3376531. Open access copy: Sheffield Hallam University Research Archive. Autobiographical design study, first-author account, n=1. Source of the ten-week period, the archive of 63 tagged photographs, the twelve-images and seven-check-ins detail, and the definition of ongoingness.

3. Hollanek, T. and Nowaczyk-Basinska, K. (2024). Griefbots, Deadbots, Postmortem Avatars: on Responsible Applications of Generative AI in the Digital Afterlife Industry. Philosophy & Technology, published 9 May 2024. Leverhulme Centre for the Future of Intelligence, University of Cambridge. DOI 10.1007/s13347-024-00744-w. Open access copy: Cambridge University Repository. Source of the principle of mutual consent and of the scenario described on this page, which is the third of the paper's three design fictions ("Stay") and which the authors state plainly is speculative rather than a report of a real family.

4. Odom, W., Harper, R., Sellen, A., Kirk, D. and Banks, R. (2010). Passing on and putting to rest: understanding bereavement in the context of interactive technologies. CHI '10, 10 April 2010. DOI 10.1145/1753326.1753601. Companion CHI 2010 study on inherited digital and physical objects. Background only; no figure on this page comes from this paper.

Last reviewed: 30 July 2026. This page carries no legal or platform-policy claims and sits on a 90-day recheck cycle.