You Are Not Short of Material. You Are Short of a Question.

Tulving and Pearlstone showed in 1966 that people routinely fail to produce material that arrives at once when a cue points at the right part of memory. They named the gap availability versus accessibility. The material is available. Whether you can reach any of the material depends entirely on what you are given to reach with.

Conway and Pleydell-Pearce went further in 2000. Autobiographical memories are not stored intact and retrieved whole. They are assembled at the moment you go looking, out of a knowledge base that control processes search by "successively shaping cues".

Which makes "tell me about your life" the worst instruction you can be given. There is no cue in it, so there is nothing to shape. The search returns the only thing a search with no cue can return, which is a summary of a summary: the north, not much money, hard-working father, different time.

Name a room instead and you get the room, the people in the room, and what somebody said out loud in the room.

What you say to an AI of yourself is one scene, at one date, in one room, and then the next one after that. How to answer a question about your own life so the memory actually arrives:

• Pick a date, not a period. Not "my twenties". A day you can name: a first day in a job, the day you moved into a particular flat, a birthday you can place. If no day comes, name a room you spent a lot of time in during one particular year.

• Rebuild the room before you say a word. Sixty seconds, eyes shut, nothing spoken. The light. The sound. Who else was in the building. What you were wearing. What you had eaten. Whether you were warm. This step is called context reinstatement, and it is the half everybody skips.

• Start recording and report everything. The boring parts, the parts you are unsure about, the parts that seem to belong to a different story. Say "I think" out loud rather than dropping a detail you cannot stand behind. Do not tidy while you talk. Editing while recalling shuts the search down.

• When you dry up, follow one detail you already said. You mentioned the coat. Where did the coat come from. You mentioned the neighbour. What was the neighbour's name. That is the only follow-up rule you need, and an interviewer is somebody who performs that rule for you.

Two minutes of work, none of it invented by me.

Steps two and three go together for a measured reason, and the measurement is unusually clean.

Milne and Bull ran the test in 2002 with 125 people: 34 undergraduates, 44 children aged eight to nine, 47 children aged five to six. Everyone watched a video of an accident. Forty-eight hours later each person was interviewed under one of six instructions: reinstate the context, change perspective, change the order, report everything, report everything combined with context reinstatement, or a control instruction to simply try again.

Every technique used on its own produced no more recall than being told to try again. Context reinstatement and report everything used together produced significantly more correct recall than any of them alone. Rebuilding the room without reporting everything does nothing. Reporting everything without rebuilding the room does nothing. The pair is the whole effect.

The full protocol built on those principles is the cognitive interview, developed for police witness interviews in the mid-1980s. Köhnken, Milne, Memon and Bull pooled 42 studies, 55 comparisons and nearly 2,500 interviewees in 1999. Correct details increased with a strong effect size of d = 0.87. Incorrect details also rose, at d = 0.28, considerably smaller. Accuracy rates held almost level: 85 per cent of reported details were correct under the cognitive interview against 82 per cent under standard interviews. People said a great deal more without getting appreciably looser. Memon, Meissner and Fraser replicated that a decade later across 46 published articles, reporting "a large and significant increase in correct details and a small increase in errors" with "no differences in the rate at which details are confabulated". One line in that paper matters most if you are reading this in your sixties or seventies: the method "appeared to benefit older adult witnesses even more than younger adults".

Now the concession, because the honest version of this argument needs one. All of that is eyewitness research about a staged crime. None of it is research about recording your own life. The retrieval machinery is the same in both cases, so the part about getting more detail out of a person carries over. Nobody has run a meta-analysis on whether your daughter finds the recording moving in twenty years.

Which means everything on this page about emotional weight is craft, not evidence, and I am going to label it as craft where it appears. My own industry has a bad habit of dressing taste in a citation, and any company in this space that quotes a memory paper to prove its product is meaningful is overselling, and deserves every public correction that follows.

Levine and colleagues gave younger and older adults a structured Autobiographical Interview and scored every single detail as either episodic or semantic. Younger adults leaned toward episodic detail: happenings, locations, perceptions, thoughts. Older adults favoured semantic detail not connected to any particular time and place, and the drift persisted even after interviewers probed for more.

That drift is the failure mode of almost every answer given to a broad question, at any age. It is what the four flat sentences at the top of this page are made of.

So one standard runs through everything below. A weak answer summarises a period. "We were always outside." "Dad worked a lot." Both true, both dead on the page. A strong answer puts you somewhere on a nameable day, with weather, furniture, names and something somebody said out loud. If you can hear a sentence in your own answer that somebody actually spoke, you are at the right level and you can stop worrying about whether the answer is any good.

Two rules sit underneath every question in this piece, so that you can write your own once you run out. Every question names a time, a place, a person or a sense. And every question can be answered in one scene rather than one summary.

There is no correct thing to say in a legacy recording. There is one scene, at one date, in one room, and then the next one after that.

Start here when nothing is coming, because smell and sound reach material that words never touch, and the difference has been measured directly.

Chu and Downes tested it head on in 2002. Using a double-cuing method, they compared memories cued by an actual odour against the same memories cued by the odour's name, by an irrelevant odour, and by a visual cue. Odour-cued memories were "reliably superior in the amount of detail yielded". The smell itself, not the word for the smell.

Willander and Larsson pushed further with 93 older adults, and the result is the strangest finding in this whole literature. Word and picture cues pulled memories peaking in early adulthood, roughly ages 11 to 20. Odour cues pulled memories from the first decade of life. Scent-cued memories also produced stronger feelings of being brought back in time, and they "had been thought of less often" than memories reached any other way. Rarely rehearsed, and therefore rarely smoothed.

Music does something similar. Belfi, Karlan and Tranel played 30 songs and showed 30 famous faces to the same participants. Music-evoked memories came back more vivid, with a greater proportion of internal detail and more perceptual detail than the faces produced.

So put the smell or the song in the room before you read the question out. Then ask what the house you grew up in smelled like first thing in the morning, and who was already awake. What song was playing the first time you drove somewhere alone, and where you were going. What your grandmother's kitchen sounded like when the room was full, whose voice carried, and who was not speaking to whom. Which smell tells you a season has changed, and how old you were the first time you noticed. What your hands smelled like at the end of a shift in the worst job you ever had.

Photographs work too, with one correction. Sellen and colleagues at Microsoft Research studied a wearable camera that captured a person's day automatically. The images did help people connect to their past, but the researchers separated genuinely remembering an event from merely knowing about one, and the two behaved differently over time. The photograph is the cue. The ten minutes before the photograph are the memory. So never describe what is in the frame. Describe the argument in the car on the way there, the wait, the joke somebody made just before the camera came up.

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

Chronological order is the format that dies in the 1970s and never reaches you at fifty. Left alone, people start at birth, work forward, get tired somewhere around the second job, and stop. The ones who eventually listen get a childhood in high resolution and an adulthood in one paragraph.

Start in the dense part instead. A systematic review of 68 studies confirmed the reminiscence bump: middle-aged and older people recall disproportionately more personal memories from roughly ages 10 to 30, widening to approximately 5 to 30 when word cues are used. Most first times sit in that stretch, and identity got assembled there, which makes it the richest seam per minute of recording you will ever have.

Then pin the search to a date. Where you were on your thirtieth birthday, hour by hour, starting with the ceiling you woke up under. What the last ordinary day looked like before something changed for good, deliberately mundane, what you ate and what you were annoyed about. What you did on the Tuesday after you got the news, not the day of, the Tuesday after, when the phone had stopped ringing. The first night in the first place you paid rent on, what you slept on, what you had not bought yet, what came through the wall. The morning of the day you met the person you ended up with, beginning before you knew anything was going to happen, including how close you came to not being there.

Or pin it to a routine, which is where families actually live. What a normal Thursday looked like the year you turned twenty-two, alarm to bedtime, prices and brand names welcome. What you ate most weeks for a whole year, and who cooked. What your father said when he answered the telephone, in his words, at his speed. What is a joke in your family that no outsider would understand, and where the joke came from. What was in your pockets at twenty-five.

Or pin it to what only you can do, because practical knowledge is the first thing to go. What you do at work that you could not put in a handover document. How you make the one thing you make well, said out loud for somebody who cannot see your hands, including the bit you never measure. What you know about your own family that has never been written down anywhere, names first, then who told you, then how sure you are. What advice you give people that you got from somebody who has since died. If that last group is the one that lands hardest, the things only you know how to do goes further into the judgement that leaves with one person, and capturing your expertise before you retire covers the same problem at work.

Leave the position questions until you are warm. Dan McAdams' Life Story Interview II, the Foley Center protocol at Northwestern in its 2007 revision, is structured around eight key scenes including the high point, the low point, the turning point and a wisdom event, and every one of them is asked as a single scene: what happened, when and where, who was involved, what you were thinking and feeling. In that spirit: what you believe now that you did not believe at twenty-five, and the specific thing that happened that made the old belief impossible to hold. What you have been publicly wrong about. What you would want said about you if you were not in the room, in one sentence, and who you would want saying that sentence. What you want your children to do that you never managed, your own attempt first. What question you would most want to be asked, and why nobody has asked.

None of this needs an account or a payment. Print the questions, hand the list to a friend and swap. StoryCorps publishes a free question list across seventeen categories, the Life Story Interview II protocol is free to read in full, and our own longer set of life story questions is open to anybody. For the structure of a whole hour alone rather than one scene at a time, how to interview yourself with nobody sitting opposite is the method these questions are ammunition for, and if you would rather write than speak, the case for journalling that asks you a question instead of facing you with a blank page sets out how to keep the prompt doing the work.

The hard questions produce more detail than any others, and they come with one requirement that is measured rather than tasteful.

Talarico and Rubin tested 54 students the day after 11 September 2001 and again 1, 6 or 32 weeks later. Consistency for the flashbulb memory declined at the same rate as consistency for an everyday memory. Only the confidence stayed high. Your most vivid story is not automatically your most reliable one, and the vividness is exactly what stops you checking.

So mark your uncertainty out loud as you go. "I think it was the Thursday" is worth far more to the people listening than a confident wrong date, because a confident wrong date is the thing a family will argue about at a funeral.

Then set the limit before you start rather than halfway through. The worst day you have had that you have never described out loud, and how much of that day you are willing to say. What you got wrong that you have never corrected with the person involved, named as an action rather than a feeling. When you knew something was over before you admitted so, one moment, placed, with what was said and what you said back. What you lost that nobody else understood as a loss, the thing itself described in full before you explain why the loss mattered. What you were frightened of at fourteen, and what you actually did about that. State the boundary in the recording, out loud: I will say this much and not the rest. Then answer inside the boundary in complete detail.

This is the standard software has to meet too, and I run a company that sells the software, so read the rest of this section with my interest in front of you.

Afterlife.ai™ performs the four moves for you. A guide called Idy asks one question at a time and reads what you actually said before choosing what to ask next. Mention a coat and the next question is about the coat. Let a name arrive and go quiet, and the name comes back later. Idy goes for the biggest thing first rather than starting at your birth and marching forward. Sessions are short and nothing asks you to prepare. What comes out of a session is memories, held in your account, editable and deletable by you at any time while you are alive, and you can record your own voice and have a voice clone built from those recordings so that what is eventually heard carries your voice rather than a stock narrator reading a transcript. If handing over voice samples is the part that gives you pause, what makes a voice clone safe or unsafe to make sets out the consent model, and protecting your voice from being cloned by somebody else covers the wider problem this sits inside.

Two limits, plainly, from building the product and watching people work through a session. The guide is good at following the detail you gave a minute ago and much weaker at judging which decade of your life matters most, so a list like the one above still beats leaving that choice to software. And a cloned voice carries whatever you gave the recorder: a noisy kitchen, a head cold, a phone held at arm's length, all of it stays audible in the result, and no amount of processing afterwards puts back the recording you did not make.

Personas built this way answer from what you actually said. Where a specific detail was never recorded, the standing instruction is to speak from what is there and let the silence stand rather than manufacture a fact, and to hedge in plain language ("I think", "my sense is") no more than once in a message. What you never said never gets invented for you.

Nothing is released to anybody by default either. You nominate the specific people who may ever have access, by name, and access opens only through Executor Lock™: a seven-day, evidence-verified process that has to be triggered and verified before a single nominated person can reach what you built. If you never nominate anybody, nothing is ever released. Your first 50 memories are free, no card, and the free build does not expire. Listening back in your cloned voice is metered on the free tier, and the current numbers live on the pricing page rather than in this article, because a figure typed into an article goes stale. You can delete individual memories whenever you like, or delete the whole account, which removes your memories, your recordings and what was built from them. Our privacy policy commits to removing personal data within 30 days, backups included, and states that personal data is not sold, rented or licensed and is not used to train AI models. One honest exception: an account holding active Executor Lock™ records has to have those resolved with support before the account itself can be deleted.

People sit down to answer the big questions and leave having recorded the asides.

What families come back to, in my experience, is almost never the thing that would have gone on the list. The way you said a particular word. The reason you never ate a certain food. The five seconds where you started laughing halfway through a sentence about your brother. The bit where you said "I don't know why I've never told anyone this" and then told them anyway.

None of that survives a plan, which is why the question has to come from outside you. You cannot see the answer coming, so you cannot arrange the answer in advance. That is the entire mechanism, and it is why being asked beats deciding what to say.

So answer the telephone question before you answer the belief question. The version of this that works is the one where somebody asks what your father said when he answered the phone, and you hear yourself do the voice, at his speed, with his words, and you sit there working out how many years it has been since anybody last heard that.

The blank you felt at the start of this was never a verdict on your life. It was a verdict on the question you were handed.

Pick three. Not thirty. Three answered properly beats a whole list skimmed and none of them started, and two minutes of talking gets you the first usable scene, with the recorder already in your pocket.

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

• How to interview yourself and run a full session alone is the long-form method behind the four moves, for anybody working without software.

• Why an ordinary life is still worth recording deals with the belief underneath a blank recording: that a life has to be notable before any of it is worth keeping.

• The questions your child will actually ask you at thirty is a reality check on which of these answers gets used later.

• What to do with your story when you turn sixty works through where to start when the whole span is finally in view.

• Recording your life story end to end covers formats, equipment and what to do with the files once you have them.

• See the plans if you want to know what sits beyond the free build.

The claims above about cueing, context reinstatement, report everything, sensory cues and the reminiscence bump come from the primary literature listed here: peer-reviewed journal articles and two meta-analyses, read directly rather than through summaries. The question lists referenced are the published protocols themselves, one academic and one public. What is not settled: every piece of the retrieval research was run on eyewitness memory or laboratory cues, and no study has tested whether any of these methods makes a recording more meaningful to the family who eventually hears the result. The paragraphs describing what Afterlife.ai™ does, including the two stated limits, are written from the product code and from watching people work through a session, and the statements on cost, deletion and ownership cite our own published privacy policy (source 15). No memory study on this page supports a claim about our software, and none is offered as though it does. If something here is wrong or out of date, tell us through the contact page and we will fix the error.

1. Tulving E, Pearlstone Z. "Availability versus accessibility of information in memory for words." Journal of Verbal Learning and Verbal Behavior, 5(4), 381-391, August 1966. https://doi.org/10.1016/S0022-5371%2866%2980048-8

2. Conway MA, Pleydell-Pearce CW. "The construction of autobiographical memories in the self-memory system." Psychological Review, 107(2), 261-288, 2000. Memories described as "transitory mental constructions within a self-memory system"; control processes work by "successively shaping cues". PMID 10789197. https://pubmed.ncbi.nlm.nih.gov/10789197/

3. Levine B, Svoboda E, Hay JF, Winocur G, Moscovitch M. "Aging and autobiographical memory: Dissociating episodic from semantic retrieval." Psychology and Aging, 17(4), 677-689, 2002. Younger adults biased toward episodic detail; older adults favoured semantic detail "not connected to a particular time and place", a pattern that "persisted after additional structured probing". https://doi.org/10.1037/0882-7974.17.4.677

4. Milne R, Bull R. "Back to basics: a componential analysis of the original cognitive interview mnemonics with three age groups." Applied Cognitive Psychology, 16(7), 743-753, 2002. 125 participants; six instruction groups; individual mnemonics produced "no more recall than the 'try again' control", while report everything plus context reinstatement "resulted in significantly more correct recall". https://doi.org/10.1002/acp.825

5. Köhnken G, Milne R, Memon A, Bull R. "The cognitive interview: A meta-analysis." Psychology, Crime & Law, 5(1-2), 3-27, 1999. 42 studies, 55 comparisons, nearly 2,500 interviewees. Correct details d = 0.87; incorrect details d = 0.28; accuracy 85 per cent versus 82 per cent. https://doi.org/10.1080/10683169908414991

6. Memon A, Meissner CA, Fraser J. "The Cognitive Interview: A meta-analytic review and study space analysis of the past 25 years." Psychology, Public Policy, and Law, 16(4), 340-372, November 2010. 57 articles (65 experiments) in the study space, 46 in the meta-analysis; "a large and significant increase in correct details and a small increase in errors"; "no differences in the rate at which details are confabulated"; benefits older adults more than younger adults. https://doi.org/10.1037/a0020518

7. Chu S, Downes JJ. "Proust nose best: odors are better cues of autobiographical memory." Memory & Cognition, 30(4), 511-518, 2002. Odour-cued memories "reliably superior in the amount of detail yielded". PMID 12184552. https://pubmed.ncbi.nlm.nih.gov/12184552/

8. Willander J, Larsson M. "Smell your way back to childhood: autobiographical odor memory." Psychonomic Bulletin & Review, 13(2), 240-244, April 2006. 93 older adults; odour-cued memories from the first decade of life, while "memories associated with verbal and visual cues peaked in early adulthood (11-20 years)"; odour memories "had been thought of less often". PMID 16892988. https://pubmed.ncbi.nlm.nih.gov/16892988/

9. Belfi AM, Karlan B, Tranel D. "Music evokes vivid autobiographical memories." Memory, 24(7), 979-989, 2016. 30 songs versus 30 famous faces; music-evoked memories more vivid, with a greater proportion of internal details and more perceptual details. https://doi.org/10.1080/09658211.2015.1061012

10. Munawar K, Kuhn SK, Haque S. "Understanding the reminiscence bump: A systematic review." PLOS ONE, 13(12): e0208595, 11 December 2018. 68 quantitative studies across 9 databases; bump at approximately 10-30 years for important memories and approximately 5-30 years under word cuing. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0208595

11. Talarico JM, Rubin DC. "Confidence, Not Consistency, Characterizes Flashbulb Memories." Psychological Science, 14(5), 455-461, September 2003. 54 Duke students recorded memories on 12 September 2001 and were retested at 1, 6 or 32 weeks; consistency for flashbulb and everyday memories did not differ and declined equally, while vividness and belief in accuracy declined only for everyday memories. https://doi.org/10.1111/1467-9280.02453

12. Sellen AJ, Fogg A, Aitken M, Hodges S, Rother C, Wood K. "Do life-logging technologies support memory for the past?" CHI 2007, 81-90. Images "facilitate people's ability to connect to their past", drawing a distinction between "remembering" the past and "knowing" about the past. https://doi.org/10.1145/1240624.1240636

13. McAdams DP. "The Life Story Interview II." Foley Center for the Study of Lives, Northwestern University, revised 2007. Section B lists eight key scenes, numbered 1 to 8, including high point, low point, turning point and wisdom event, each asked with "What happened, when and where, who was involved, and what were you thinking and feeling?". Protocol PDF, checked 30 July 2026. https://cpb-us-e1.wpmucdn.com/sites.northwestern.edu/dist/4/3901/files/2020/11/The-Life-Story-Interview-II-2007.pdf

14. StoryCorps. "Great Questions." Free question list, 17 categories, no account required, checked 30 July 2026. https://storycorps.org/participate/great-questions/

15. Afterlife.ai privacy policy, checked 30 July 2026. Source for the statements that personal data is not sold, rented or licensed, is not used to train AI models, and is removed within 30 days of account deletion including backups. https://www.afterlife.ai/privacy-policy

Last reviewed: 30 July 2026. This page carries no legal, regulatory or platform-policy claims, and its research citations are stable publications. It sits on a 90 day recheck cycle, primarily to confirm that the external question lists linked above are still free and still live.