You Are the Worst Interviewer Your Own Life Will Ever Have

You reason better about other people's lives than about your own, and the gap has been measured.

Grossmann and Kross ran three experiments with 693 participants in 2014 and found that people reasoned more wisely about someone else's relationship problem than their own: recognising the limits of what they knew, weighing other perspectives, allowing that things might change. The gap held in younger and older adults alike. What closed it was self-distancing. Instruct people to step back from their own situation rather than sit inside it and the asymmetry disappeared.

In a self-interview the person asking and the person answering are stuck in the same seat. That is position, not laziness, and practice does not fix a position. You can run the same session forty times and get forty smoother versions of the story you already tell.

Then there is what you leave out, and you leave it out precisely because it is obvious to you. Nickerson's 1999 review in Psychological Bulletin set out how consistently people impute their own knowledge to others: assuming what they know is known, assuming what is unremarkable to them is unremarkable full stop. So you skip your father's hands. You skip the smell of the corridor at the factory, and what a phone call cost in 1987, and the fact that nobody in the house ever answered the door. None of that feels like information. That is the only information anybody will ever want.

The underlying mechanism was named sixty years ago. Tulving and Pearlstone showed in 1966 that people could not recall words they had definitely stored, until somebody handed them the category name, at which point the words arrived. Availability versus accessibility. The words were never lost. They were out of reach until the cue turned up.

Conway and Pleydell-Pearce went further in 2000. Autobiographical memories are not lifted whole out of storage. They are built at the moment of asking, out of cues. No cue, nothing to build with.

Use all six. Each one fixes a different failure, and the protocol further down underperforms if you drop any of them.

• Answer to one named person, not to posterity. Decide before you start who sits on the other side of the microphone. Not "my family". One person, with a name, at a stated age: your daughter at thirty-four, your brother, a grandchild who does not exist yet. Pasupathi's 2001 review in Psychological Bulletin describes conversational recounting as co-constructed, with the listener shaping what the speaker produces. Alone in a room you have no listener, so you install a specific imagined one and let them do the shaping.

• Fix one date, never a period. "The eighties" gives you nowhere to stand. "The Tuesday we moved into Rosslyn Street" does. Levine and colleagues showed in 2002 that autobiographical recall separates into episodic detail, tied to one particular time and place, and general semantic knowledge tied to neither, and that the two are distinct enough to be scored apart. Episodic detail is the only kind worth recording.

• Ask in the third person, using your own name. "What did Chris do when the call came?" Kross and colleagues ran seven studies with 585 participants, published in the Journal of Personality and Social Psychology in February 2014, and found that using your own name or non-first-person pronouns during introspection enhances self-distancing, improves performance under social stress as rated by outside observers, and reduces rumination afterwards.

• Speak, do not write. Ruan and colleagues at Stanford measured speech input against a phone keyboard in 2016 and found speech about three times faster, 153 words per minute against 52 for English, with a lower corrected error rate. Voice memo. Phone face down.

• Stop mid-thought. End every session before the natural ending, ideally halfway through a sentence you already know how to finish. Zeigarnik reported in 1927 that unfinished tasks are held in mind differently from finished ones, and a meta-analysis by Wendsche, Weigelt and Syrek in Anxiety, Stress and Coping in January 2026 confirms across 17 between-person and 14 within-person studies that unfinished tasks generate more thinking about them afterwards.

• Never edit in the session where you record. Koriat and Goldsmith showed in 1996 that people trade quantity against accuracy by withholding: when you are trying to be right, you say less. Say the version you are unsure about, then say why you are unsure. Payne's 1987 review in Psychological Bulletin covers hypermnesia, the finding that repeated attempts at the same material keep turning up items the earlier attempts missed.

Three of those rules cost you something.

The named listener is the expensive one. In the same 2001 review, under what Pasupathi calls consistency, the version you tell influences what you subsequently remember. Tune hard toward your daughter at thirty-four and you have partly chosen your own memory. Do it anyway, because a version aimed at one real person beats an hour of material aimed at nobody, but the trade is real and pretending otherwise would be dishonest.

The fixed date is doing less than it appears to. A date works as a boundary rather than a cue, forcing you to reconstruct one specific set of circumstances, and the detail arrives with the reconstruction. That last step is our inference from the episodic and semantic split, not something Levine tested. The speech figure came from short messages typed on a phone, not from life stories, so treat the three-times multiple as an indication and nothing firmer. Speaking wins here for a different reason anyway. Writing runs at the speed of your editor, and your editor is the thing you are trying to get past.

One body of research goes straight at the problem, and it comes from police work.

In 2009, Fiona Gabbert, Lorraine Hope and Ronald Fisher published two studies in Law and Human Behavior on a tool they called the Self-Administered Interview. Police cannot get to every witness quickly enough, so the researchers built a printed booklet a witness could work through alone, carrying the retrieval instructions a trained interviewer would normally deliver out loud.

In the first study, people using the booklet reported more correct details than people simply asked to write down what they remembered, and performed at the same level as people given a Cognitive Interview by a trained interviewer. In the second, people who used the booklet immediately after witnessing a staged crime recalled more correct details a week later than people who had not.

It replicated in harder conditions. Gawrylowicz and colleagues found in 2014 that older adults who used the booklet reported more correct details, and carried some of that advantage into a later event where they had no tool at all. Dando, Gabbert and Hope found in 2020 that older witnesses who completed it recalled more correct information two days on, and confabulated less.

The booklet worked because somebody else wrote the questions in advance and the person answering simply followed them. Nobody had to invent a good question while also trying to remember. Those are two jobs, and doing both at once is what kills a self-interview at minute four. The Cognitive Interview the booklet was modelled on has its own long evidence base, reviewed meta-analytically by Memon, Meissner and Fraser in Psychology, Public Policy, and Law in November 2010. You do not need to read any of that literature. You need to accept one claim: a protocol beats improvisation, even when the protocol is a piece of paper and the interviewer has gone home.

Now the boundary, because I would rather draw it myself than let you find it later. The Self-Administered Interview was built for witnesses and tested on staged events seen minutes or days earlier, never on a house you left in 1979. The mechanism should carry across. The effect sizes will not.

A protocol also does nothing for accuracy. More material comes out, some of that material will be wrong, and some will be two different afternoons welded into one. Levine and colleagues found in 2002 that older adults drifted from specific scenes toward general facts, and that structured probing by an interviewer roughly doubled the number of details recalled while leaving that drift exactly where it was. More came out. The shape of what came out did not change.

Twenty minutes a session. Once a week, twice at most, never twice in a day. Voice memo rather than paper. The same imagined listener the whole way through. Every session opens by saying the fixed date out loud before anything else, and closes mid-sentence.

Sessions three, four and five sit where the material is densest. Munawar, Kuhn and Haque reviewed 68 quantitative studies in PLOS ONE in December 2018 and found that middle-aged and older people recall disproportionately from early life, though the peak moves depending on how you ask: roughly ten to thirty years of age for important events, roughly five to thirty when the prompt is a cue word, wider again in studies using life scripts. Those sessions are placed there for that reason and not out of sentiment.

Session 1. One address. A house you lived in before you were eighteen. Name it out loud, then walk from the front door to where you slept, saying everything you pass. What was on the walls of the room nobody used. Where in that house money got discussed, and in what tone.

Session 2. Your first paid work. Fix on the first day. What you were paid, exactly. What you did wrong in the first week and who told you. What you bought with the first money that was yours.

Session 3. The year you were seventeen. Fix on one specific month. Who you spent that month trying to impress. What you were listening to on the way somewhere. What you thought you were going to be, and how certain you were.

Session 4. The person you have not thought about in ten years. Fix on the last time you saw them. Not a eulogy. Their hands, their handwriting, what they always said, what they were wrong about.

Session 5. A decision that went the other way. Fix on the day you decided. Who told you not to. What you were actually afraid of, as against the reason you gave people. Whether you would do the same again, answered in the third person, using your own name.

Session 6. The kitchen. One meal that got made over and over. Who made it, what went in, what nobody wrote down. What you were allowed and not allowed to do at that table.

Session 7. Something you can do that most people cannot. Fix on the last time you did the thing. Explain it to someone who has never seen it done, in order, including the part you only know by feel and have never had to put into words. This is usually the most valuable of the eight, and the one people skip.

Session 8. Something you were certain about and are not now. Fix on when you changed. What broke the old position. What you would say to the version of yourself who still believed the old thing, using your name rather than "I".

Two closing rules. Do not listen back until all eight exist, because listening back is editing. And when the eight are done, run session one again from the same address on a different day. What surfaces the second time is the material the first pass could not reach, and it is usually the better material.

None of that needs an account, a card or a signup, and the eight sessions above are complete as they stand. If twenty formal minutes is more than you will realistically sit down for, the daily version of the same mechanism is journaling that asks you the questions instead. If you are reading this with fifty or sixty years behind you, the protocol rebuilt for turning sixty reorders the sessions and moves the anchor dates. If you want more questions after eight, StoryCorps publishes its Great Questions list free, Dan McAdams's Life Story Interview II is the academic standard, and our own set of life story questions is organised by area rather than by session, which makes it a good supply of anchor dates for a second run. For the file-handling side the protocol assumes, how to record your life story as usable audio covers the mechanics. And if the trouble is not method but the blank at the moment you press record, what you would even say to an AI built from you is the page to read first.

Run session one this week on the phone in your pocket. If the material comes out thin, you have learned something useful for nothing.

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

A protocol cannot ask about the thing you said in minute nine, and that single gap is the whole remaining problem.

The booklet matched a trained interviewer on a staged crime lasting a few minutes, and it could, because the researchers already knew what was in the event and wrote the questions to fit.

Nobody knows what is in your life. Not you, in the sense that matters here, because the accessible part is the part you have already told. A protocol written on Tuesday cannot contain a question about the sentence you produce on Thursday, and the sentence you produce on Thursday is where the material was.

Listen for the moment when you finally do play the recordings back. You say something slightly odd, in passing, at speed, and then carry straight on: "which is why I never learned to drive until I was thirty-one." Under a protocol, that sails past. In a real interview, that is where the interviewer stops everything and says hang on, why.

The second question is not a technique. The second question is the entire difference between a transcript and a life.

You interview yourself about your life story by working from a protocol written in advance rather than improvising questions. Answer to one named person, fix one date instead of a decade, ask in the third person, speak instead of writing, stop mid-sentence, and never edit on the day you record. What paper cannot do is hear the odd thing you said in passing and ask why.

Afterlife.ai™ is built around that one gap, so I will keep this narrow.

A guide called Idy runs the interview the way the protocol above runs, with fixed anchors and specific questions, and then does the thing paper cannot: reads what you actually said and asks about that, rather than moving to the next item on a list. Mention the driving in passing and Idy stops and asks why.

What comes out is stored as memories, yours to review, edit or delete at any point while you are alive. You record your voice and it is cloned, so your answers can be heard rather than read. You nominate the people who get access, and nothing reaches any of them until Executor Lock™, the seven-day evidence-verified release process, has completed. If the voice side is the part you are unsure about, whether voice cloning is safe to do at all deals with that question directly.

What the memories and the voice build together is what we call a "Persona": a version of you that answers from what you actually said, in your voice. Yours is not you, and not a summary of you either. Where you never covered something, Personas answer from what you did cover and never manufacture the missing detail. Any company in this category that blurs the line between Personas and the people they came from will deserve the backlash heading its way. For the technical argument about what a model built only from one person's own material can and cannot do, that case is set out separately.

The build, the voice recording, naming a Trusted Contact and setting up Executor Lock are all open on the free tier. Your first 50 memories cost nothing, no card is asked for at any stage, and the free build does not expire, so nothing you record in these eight sessions sits on a clock. Listening back in your cloned voice is metered on the free tier. What the paid plans add is room: more memories, more storage, more people you can name. Current numbers live on the plans page, because they move and a figure typed into an article goes stale.

Can I just use ChatGPT or another assistant to interview me?

You can, and the follow-up behaviour is genuinely better than a printed list. Two practical limits. Check the retention and training settings on whatever you use, because a life-story transcript is close to the most sensitive text you will ever produce. And decide where the transcript ends up, because a general assistant is not built to hold your material for thirty years or hand it to a named person later.

How long does eight sessions take?

About three hours of recording spread over four to eight weeks, plus a second pass at session one. The spacing is not padding. Payne's review of hypermnesia found that repeated attempts at the same material keep producing items that did not surface the first time, so the gaps between sessions are doing work for you.

Is it better to write or to speak?

Speak, for this. Writing runs at the pace of your internal editor, which is the thing suppressing the material. Speech input has been measured at roughly three times keyboard speed for short messages, and more importantly it strips out the sentence-polishing that turns remembering into composition. Transcribe later if you want text.

What if I get things wrong?

You will. Vividness and accuracy are separate, and a confident memory is not automatically a correct one. Say the uncertain version, then flag the uncertainty out loud in the recording: "I think that was 1994, could have been the year after." A hedged true memory beats a clean invented one, and anyone listening later can tell the difference.

Who owns what I record?

You do. Your recordings, transcripts and memories are yours, you can review, edit or delete memories while you are alive, and nobody sees any of your material unless you nominate them. Nothing you record is used to train a general model that serves other people. Ask every provider in this category the same question before you start, and be suspicious of a vague answer.

What happens if the company shuts down?

A fair question for every provider here, including us. Ask three things: what can I get back out, in what format, and what happens to that route if the business fails. A provider who cannot answer all three plainly has given you your answer. Keep your own copy of the raw voice memos regardless, because they cost nothing to store.

Every empirical claim above comes from a peer-reviewed primary source, listed below with a DOI or, where a paper has none, a permanent preprint identifier: the original Self-Administered Interview studies rather than write-ups of them, the Zeigarnik paper in the original German alongside the 2026 meta-analysis, and systematic reviews where a single study would have been thin. The two protocol references, McAdams and StoryCorps, are the published instruments themselves.

What is not settled: the Self-Administered Interview was validated on staged events witnessed minutes or days earlier, not on decades-old autobiographical memory, so the mechanism is well supported and the effect sizes are not transferable to your childhood. The six rules combine findings from separate literatures that have never been tested together as one protocol, and the eight-session structure is our construction rather than a validated instrument. The speech-versus-typing figure was measured on short phone messages. And the reminiscence bump does not sit in one fixed window, since the ages it covers move with the way the memories are prompted, so treat sessions three to five as aimed at a dense region rather than at an exact decade.

Found an error, or think we have read a study wrong? Tell us through the contact page and we will correct the page and say that we did.

1. Gabbert F, Hope L, Fisher RP. "Protecting eyewitness evidence: examining the efficacy of a self-administered interview tool." Law and Human Behavior, 33(4), 298-307, August 2009 (online 17 June 2008). doi.org/10.1007/s10979-008-9146-8

2. Gawrylowicz J, Memon A, Scoboria A, Hope L, Gabbert F. "Enhancing older adults' eyewitness memory for present and future events with the Self-Administered Interview." Psychology and Aging, 29(4), 885-890, December 2014. doi.org/10.1037/a0038048

3. Dando CJ, Gabbert F, Hope L. "Supporting older eyewitnesses' episodic memory: the self-administered interview and sketch reinstatement of context." Memory, 28(6), 712-723, July 2020. doi.org/10.1080/09658211.2020.1757718

4. Grossmann I, Kross E. "Exploring Solomon's paradox: self-distancing eliminates the self-other asymmetry in wise reasoning about close relationships in younger and older adults." Psychological Science, 25(8), 1571-1580, August 2014. doi.org/10.1177/0956797614535400

5. Nickerson RS. "How we know, and sometimes misjudge, what others know: Imputing one's own knowledge to others." Psychological Bulletin, 125(6), 737-759, November 1999. doi.org/10.1037/0033-2909.125.6.737

6. 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. doi.org/10.1016/S0022-5371%2866%2980048-8

7. Conway MA, Pleydell-Pearce CW. "The construction of autobiographical memories in the self-memory system." Psychological Review, 107(2), 261-288, 2000. doi.org/10.1037/0033-295X.107.2.261

8. 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. doi.org/10.1037/0882-7974.17.4.677

9. Kross E, Bruehlman-Senecal E, Park J, Burson A, Dougherty A, Shablack H, Bremner R, Moser J, Ayduk O. "Self-talk as a regulatory mechanism: How you do it matters." Journal of Personality and Social Psychology, 106(2), 304-324, February 2014. doi.org/10.1037/a0035173

10. Pasupathi M. "The social construction of the personal past and its implications for adult development." Psychological Bulletin, 127(5), 651-672, September 2001. doi.org/10.1037/0033-2909.127.5.651

11. Koriat A, Goldsmith M. "Monitoring and control processes in the strategic regulation of memory accuracy." Psychological Review, 103(3), 490-517, 1996. doi.org/10.1037/0033-295X.103.3.490

12. Payne DG. "Hypermnesia and reminiscence in recall: A historical and empirical review." Psychological Bulletin, 101(1), 5-27, January 1987. doi.org/10.1037/0033-2909.101.1.5

13. Ruan S, Wobbrock JO, Liou K, Ng A, Landay J. "Comparing Speech and Keyboard Text Entry for Short Messages in Two Languages on Touchscreen Phones." arXiv:1608.07323, 25 August 2016. arxiv.org/abs/1608.07323

14. Zeigarnik B. "Das Behalten erledigter und unerledigter Handlungen." Psychologische Forschung, 9, 1-85, 1927. doi.org/10.1007/BF02409755

15. Wendsche J, Weigelt O, Syrek CJ. "Unfinished work tasks and work-related thoughts during off-job time: meta-analysis of the Zeigarnik effect in a work-recovery context." Anxiety, Stress and Coping, 39, 385-407, published online 19 January 2026. doi.org/10.1080/10615806.2026.2616302

16. Munawar K, Kuhn SK, Haque S. "Understanding the reminiscence bump: A systematic review." PLOS ONE, 13(12): e0208595, 11 December 2018. journals.plos.org

17. 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. doi.org/10.1037/a0020518

18. McAdams DP. "The Life Story Interview II." Foley Center for the Study of Lives, Northwestern University, revised 2007. Archived protocol, checked 29 July 2026.

19. StoryCorps. "Great Questions." storycorps.org, checked 29 July 2026.

Last reviewed: 30 July 2026. Next review due 28 October 2026.

• If your notes already exist and you want to know what changes when notes can be questioned back: what a second brain that answers back actually does.

• Session seven expanded into a full method: how to capture your expertise before you retire.

• The same argument for domestic and practical knowledge nobody writes down: the things only you know how to do.

• If you cannot decide who your one named listener should be: questions your child will want to ask you at thirty.

• If part of what you want on record is a language nobody else in the family still speaks: how to keep your family language alive.

• If you have the family tree and want the voices behind the dates: the next step after Ancestry for a family historian.

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