Nine thousand peopledescribed the same door.
Between 1998 and 2026, strangers with nothing in common wrote down what happened to them at the edge of death. This is what the whole archive says when you read all of it at once — and where the standard account of it turns out to be wrong.
There is a great deal of data here, and several of the figures reward hovering, tracing and clicking. It is all readable on a phone, but it opens up properly on a desktop.
Every account analysed here belongs to three archives
- nderf.orgNear Death Experience Research Foundation
- adcrf.orgAfter Death Communication Research Foundation
- oberf.orgOut of Body Experience Research Foundation
Each account is jointly copyrighted by its author and the foundation, and many contributors did not grant permission to publish their words elsewhere. No testimony is reproduced on this site. Every story here is a summary written by the analyst, and every one links to the original page, where the account can be read in the writer's own words.
Archive read through 2 August 2026 · most recent submission dated 22 June 2026 · 9,865 story links indexed, 9,764 accounts analysed here
Before Anything Else — What Each Number Rests On
Not every finding here counts the same accounts.
Some questions can be answered by machine across all 9,764 accounts. Others require someone to read a narrative from beginning to end and record what happened in what order. That second kind is slow, so it runs on a random sample. Every chart below says which it is.
Counts, word totals, dates, countries. Anything a machine can total without interpreting a sentence.
Which archive a finding draws on also varies. Anything built from the questionnaire — demographics, belief, aftereffects, recall — comes from the near-death archive alone, because only it has those fields mapped and only it carries country and age. Anything built from reading narratives covers all three. Every chart says which above it.
A smaller sample widens the error bar; it does not bias the result. The 598 accounts were drawn at random from the 1,497, which were drawn at random from all 9,764. Reading 598 narratives carefully beats pattern-matching 9,764 badly — a lesson this project learned the hard way, and which Part Two describes.
Finding One — The Question Shapes The Answer
There are two archives here, and they disagree.
Everyone who submits an account does two things. They write their story in their own words. Then they answer about fifty questions.
Nobody had compared the two. When you do, the picture of what a near-death experience contains depends entirely on which half you read — and the gap is not small.
How to read this — each row is one element measured twice. The amber dot is the share who answer yes when the questionnaire asks about it directly. The blue dot is the share who describe it themselves, unprompted, in their own narrative. The line between them is not a range: it is simply the distance between the two measurements. The longer the line, the more the two methods disagree. Hover any row for detail.
Both measurements come from the near-death archive, so the two dots on a row describe the same population. Prompted rates use the questionnaire (n ≈ 5,400 per item); volunteered rates come from 500 of those narratives read individually. Two independent readers agreed on 96% of these judgements. Reading all 1,500 narratives across the three archives would move some rows sharply — “met the dead” runs at 33.2% pooled against 16.8% here — because the after-death archive is about meeting the dead. Holding the archive constant is what makes the two dots comparable.
Read the top of that chart and the near-death experience is a journey: time dissolves, everything is understood, a border is reached. Read the bottom and it is something much simpler — you leave your body and you see a light.
Three quarters of people agree that time stopped when a form asks them. Fewer than one in five mention it when telling their own story.
So which number is true?
Both. They measure different things, and the honest answer is that the gap itself is the result.
When the element is named, people recognise it in what happened to them. That is real information — recognition is not nothing. But a question supplies the word, the concept and the expectation all at once.
The element was salient enough to reach for unprompted, in the person's own vocabulary. This is the stronger evidence that it was central to the experience rather than merely acknowledged afterwards.
Either the element is genuinely present but peripheral — true, yet not what the experience was about — or the question is prompting agreement that would not have arisen on its own. This data cannot separate those two.
The element survives both tests. People raise it themselves and confirm it when asked. Only two elements do this: leaving the body, and the light.
The practical consequence is that the familiar description of this experience — tunnel, life review, border, timelessness — is largely built from prompted answers. Read only what people volunteer, and a smaller, stranger core remains.
Reports floating above a bed and recognising the body below as her own — and states plainly that there was no tunnel, no light and no one else there. Only stillness.
Finding Two — The Shape Of The Story
The beginning and end are fixed. The middle is a coin flip.
Six hundred accounts were coded event by event, in the order each person told it. Some pairs of events never once appeared in the wrong order.
Percentage of accounts containing both events in which the stated order holds. n shown per pair.
What the events mean
The most likely path, assembled from the transitions
No single account contains all of this, and the middle steps are close to unordered. But following the highest-probability transition at each stage produces one canonical route. Solid connectors are near-deterministic; dashed ones are near coin flips.
Assembled by chaining the most frequent next-event at each stage across 598 accounts. It is a modal path, not a template — treat the dashed segments as genuinely unordered rather than as a sequence we are unsure about.
The frame is rigid. Crisis comes first, return comes last, and leaving the body always precedes looking down at it — 119 times out of 119, never once reversed.
But the famous middle sequence — tunnel, then light, then beings — barely holds together. Whether the light comes before arriving somewhere is a 51% call. That is a coin flip. The canonical ordering people recite is, in the middle, mostly something we added afterwards.
How to read this — the dots are the events; the line is only the thread joining them. Follow any single line and you are reading one person's account in the order they told it: it starts at the left with whatever they described first, touches a dot on each row for each event they reported, and finishes at the right with whatever they described last. Left to right is position in their telling, not clock time — a line reaching halfway across is halfway through that account, however long it was. Rows are the sixteen coded events, stacked by where each one typically falls, so a line running straight down the dashed diagonal would be the textbook sequence. Hover a row to light up every account that reported that event — each lit line is a whole account, so they begin on whichever row that person started with, which is why they do not all start in the same place. Hovering a dot picks out that single account and names it; clicking opens the account itself on nderf.org, where the person wrote it. Every line here is someone who sat down and typed it out.
Rows are ordered by where each event typically falls, so the diagonal is the canonical sequence rather than an assumption imposed on the data. These 200 accounts take 185 different orderings between them, and 177 of those orderings belong to exactly one account. The commonest single path is shared by seven. Only 39% get from start to finish without doubling back. The shape is real — the threads crowd the diagonal and thin out away from it — but almost nobody walks it cleanly.
Near-death accounts only. The after-death archive was coded the same way, but its accounts are about a visit rather than a passage, so the events and their order are not the same thing and pooling them would produce a shape belonging to neither.
Reports a dental anaesthetic going badly wrong, then a passage, a light, scenes from his own past, and a boundary he understood he could choose not to cross.
Finding Three — Who Is Waiting
Mostly, it is family. Rarely, it is God.
First, how often anyone is there at all.
An unidentified presence — someone felt but never named — appears more often than Jesus, God and angels combined. Add deceased parents and grandparents and the picture is domestic, not theological.
What the literature emphasises: wordless, mind-to-mind contact. It is the detail most often singled out as marking these encounters as unlike ordinary life.
What the accounts describe: mostly talking. Of encounters where a mode of communication is described at all, ordinary speech outnumbers telepathy by more than two to one — 34.9% spoken against 14.8% wordless. In a further 39.2%, no exchange is described in either direction; the being is simply present.
Reports lying at the roadside after a collision and a clear voice asking whether she was ready to go. She answered no — she had two children to say goodbye to.
Finding Four — The Words People Reach For
The vocabulary is narrower than the experience.
Every narrative was scanned for the words that cluster around the three most recognisable elements. What people describe varies enormously. How they describe it barely varies at all.
Counts of content words within about fifty words either side of each term, across the 5,749 near-death narratives. Each occurrence of the anchor word is counted, so a narrative using it repeatedly contributes more than once — these are counts of proximity, not of accounts.
The light is white and it is bright — those two words dominate everything else. The third most common neighbour is not a visual word at all: it is love, which appears near “light” 3,417 times, more often than “bright”.
The tunnel is almost never described alone. “Light” sits near it 3,493 times, and the next most common neighbour is “end”. It is written about as a passage toward something rather than as a place in itself — which is consistent with how rarely it appears at all, in one account in ten.
How to read this — the bar charts above count what sits near light and near tunnel. This asks a different question: which words keep company with each other. Two words are pulled together when they turn up in the same sentence more often than their separate frequencies would predict, so common words do not simply drift to the middle. Position is the finding; nothing here was arranged by hand. Hover any word to see what it travels with.
Four groups fall out of the arithmetic rather than being imposed on it, and they hold when the window is changed from a sentence to a fixed span of words. The strength is very unevenly spread: of the 422 word pairs that occur together often enough to measure, ten carry most of the structure and the median pair is weak. Faint lines are drawn faint for that reason — a diagram that gave every pair the same weight would look like a web and would say much less than it appeared to.
Sharing a sentence is not the same as meaning the same thing, and none of this shows what anyone meant. It shows which words arrive together when people reach for language, which is a fact about the telling rather than about the experience.
There is a cliff in that list. God, angels, Jesus and Christ account for almost everything; Buddha, the Virgin Mary, Allah, Krishna and Muhammad together account for 117 mentions against God's 5,357.
Read carefully, this is a fact about who writes in before it is a fact about what anyone saw. The corpus is 61% American and overwhelmingly Anglophone, so the vocabulary available to describe an encounter is overwhelmingly Christian. It says little about whether the underlying experiences differ — and the cross-cultural comparison earlier on this page suggests the sensory elements do not.
A limit worth naming: word proximity is not meaning. “Dark” appearing near “tunnel” 723 times does not establish that the tunnel was dark, only that both words occur close together. These counts describe the language of the corpus, not its content.
Reports ten days in a coma after a crash, and an account she returns to repeatedly as impossible to convey in the words available to her.
Finding Five — The Setting
It happens in the bedroom.
What most people picture: a tunnel, a garden, a city of light.
What the accounts actually describe: a bedroom.
Below is every setting mentioned across 598 accounts, ranked. Ordinary rooms come first by a wide margin — the places the person's body already was. The famous imagery is real, but it sits far down the list.
The tunnel — the single most recognisable image attached to this subject — appears in one account in ten. The garden appears in one in thirty. A bedroom appears in one in four.
Whatever else is happening, most of it is reported as happening exactly where the person's body already was.
Reports hearing herself tell the doctor she felt unwell mid-procedure, then finding herself at the ceiling of the operating room looking down. The entire account is set in that room.
Finding Six — The Paradox
The people who say it cannot be described, describe the most.
"There are no words for it" reads like an apology for a thin account. It is the opposite.
Accounts were sorted into three groups by how often the experiencer states that the experience cannot be put into words. Each bar is the average count of distinct events those accounts go on to describe — separation, light, an encounter, a message, and so on. If “beyond words” meant having little to say, the bars would fall. They rise.
598 accounts read in full. Events counted from a fixed 16-item vocabulary, so the scale is comparable across groups.
People who insist language fails them report 54% more structural content than people who never mention it. Ineffability is not an absence of material. It is a marker of having more to say.
It predicts life change better than any dramatic content does — better than the light, better than the tunnel. Saying "I can't explain it" forecasts transformation at 75.0% against 57.9%. Reaching a border, the most cinematic moment available, predicts it at exactly zero difference.
Reports waking at the top of the bedroom and seeing herself and her partner asleep below — then describing at length an experience she maintains cannot be described.
Finding Seven — What Actually Changes A Life
The drama predicts nothing. The speechlessness predicts everything.
Every experiencer is asked, plainly: did your values and beliefs change because of this? Two thirds say yes. The question is what separates them from the third who say no.
It is not the spectacular content. Compare the most cinematic moment in the whole corpus with the least cinematic thing a person can say.
Two elements, each splitting the corpus in two. For each, the amber bar is the share of people who did report that element and later said their values changed; the blue bar is the same share among people who did not. A difference between the pair means the element predicts life change. Identical bars mean it predicts nothing at all.
n = 4,593 for ineffability, 4,803 for the border, among those answering both items.
Reaching a border — the threshold, the point of no return, the single most dramatic beat available — predicts life change at 66.7% against 66.7%. Identical to the decimal. It makes no difference whatsoever.
Saying “I cannot put it into words” predicts it at 75.0% against 57.9% — a seventeen-point gap.
Which inverts the intuition. The part that would make a film — the tunnel, the barrier, the turning back — leaves no measurable mark on how people afterwards describe their own values. The thing that does is the admission that the experience exceeded language. Whatever changes a life here is not the spectacle. It is the encounter with something the person could not fit into words.
Reports an experience she credits with redirecting the whole course of her life. On the standard depth scale it scores 2 out of 32 — among the lowest in the archive.
Finding Eight — Something Changed
The life review did not fade. It fell off a cliff.
Your whole life flashing before you is the most cinematically famous element of all. It held steady for thirty years — then, in experiences from the 2010s, it halved.
Every dated account, grouped by the decade the experience happened in. No filtering by how long anyone waited before writing — see the check below for why that matters.
n shown per decade. Accounts without a stated experience date are excluded.
Flat at 57–61% from the 1980s through the 2000s, then 31.8% for experiences in the 2010s. Not a slow erosion — a step change in a single decade.
The check that nearly produced the opposite story
People write in at wildly different distances from the event. Splitting by that distance is the standard way to test whether a trend is really about memory rather than about the era — but the windows are not available in every decade, and choosing among them can manufacture a trend that is not there.
Reading only the middle two columns would produce a smooth decline from about 70% to about 27% — and it would be wrong. The 1980s have no under-ten-year column, because the archive did not exist yet, while for the 2000s that column holds the largest group and runs the other way at 70.0%. Any trend drawn from a subset of these columns is an artefact of which ones happen to be populated.
Comparing like with like — accounts written within ten years of the event, the only window available for both recent decades — the fall is real and sharp: 70.0% for the 2000s against 29.1% for the 2010s.
Why? Here is what it is not.
Each obvious explanation was tested and each failed. What remains is genuinely open.
That last point is the shape of it. What fell away is the elaborated narrative — the review, the vision of the future, the encounter, the total knowledge. What held is the sensory and emotional core — the light, the strangeness of time, the failure of language, the changed life.
One further change is visible and may or may not be the cause: the archive's own editors classified 70% of 2000s accounts as definite near-death experiences, against 55% for the 2010s, with “probable” and “possible” rising correspondingly. Whether that reflects a shift in who submits, in how the archive triages, or in something about the experiences themselves, this data cannot say. It is the single most interesting unresolved question in the corpus.
Reports scenes from his own life returning to him during surgery, along with music and a reluctance to come back. An account of the element that later grew scarce.
Finding Nine — Does It Depend On Who You Are
The sensations are universal. The story around them is not.
Split the same twelve elements by gender, then by region, and a clean line appears — one nobody set out to look for.
What the body seems to do is reported at the same rate by everyone. What it is taken to mean is where the groups separate.
Region groups Western Europe, North America, Australia, New Zealand and South Africa against everywhere else. The comparison group is small (roughly 700 accounts) and internally very mixed, so read it as a signal rather than a settled fact.
“Became more liberal” counts those who reported a non-liberal stance before and a liberal one after.
Where women and men differ
On the structural elements they are indistinguishable. Leaving the body: 56.8% against 57.4% — the one element men report slightly more often. The life review is identical: 55.5% against 55.5%. Entering another world differs by 1.4 points.
The gaps open only on interpretation and aftermath. Women report abilities afterwards at 46.2% against 37.3% — the largest demographic gap anywhere in the corpus. They report the experience as beyond words more often (53.3% against 46.8%), and reaching a border more often (69.1% against 63.3%). Same reported events; different account of what they meant and what followed.
Where cultures differ
The elements usually called culturally Western turn out not to be. The tunnel is reported more often outside the West — 36.4% against 29.7% — and so is the light, 64.4% against 59.8%. Time distortion, entering another realm and ineffability show no meaningful difference at all.
What does differ is relational and narrative: meeting the dead (45.7% vs 37.3%), the life review (56.5% vs 50.4%), and glimpsing the future (44.6% vs 38.5%) are all more common in Western accounts. The raw sensory core travels; the biography wrapped around it does not.
Read this one with the most caution of anything on this page. These archives are English-language and American-hosted: 61% of accounts that name a country name the United States, and the comparison group here is roughly 630 accounts from everywhere else combined — Mexico, India, Brazil, the Philippines and some sixty other countries pooled into a single bucket because no one of them is large enough to stand alone. That bucket is not “non-Western culture”. It is “people who found an American website and wrote to it in English”. The differences above clear significance, but what separates the groups may be who reaches the archive rather than what happens at the threshold.
Reports being aware throughout an operation under general anaesthetic — feeling the surgery proceed, and reaching a limit she could not pass.
Finding Ten — What Travels
The tunnel is less American than you would guess.
Prompted rates. n ≈ 2,600–3,100 United States, 1,600–1,900 elsewhere. All three differences are statistically robust.
The tunnel, the most Western-cliché image in the whole subject, is reported more often outside the United States. Meeting the dead and reviewing your life run the other way.
And people who already knew what a near-death experience was supposed to look like report fewer tunnels than people who had never heard of one — 29.2% against 37.6%, holding within every decade. Whatever expectation is doing here, it is not simply painting the picture people expect.
Reports collapsing at home after a distressing day, and a brief experience recorded in an archive where accounts from Iran number in the dozens rather than the thousands.
Finding Eleven — What Happens To Belief
Most people keep their religion. Almost nobody keeps their certainty.
Experiencers state their religious position twice — as it was before, and as it is now. The two answers turn out to measure different things, and only one of them moves.
The item records two separate facts: a stance — conservative, moderate or liberal — and an affiliation, the tradition itself. Separated, they behave nothing alike.
Each row is where people started. The three figures are where they ended up. Read across a row to see how a group dispersed. Movement is not symmetrical: liberal is the hardest position to leave and the easiest to arrive at.
Net effect: liberal rises from 35.1% to 42.1% of the corpus; conservative falls slightly, from 20.5% to 19.0%. 73.4% report no change at all.
Affiliation is stickier than stance: 62.2% name the same tradition afterwards. But the retention rates are not evenly spread, and the lowest one is the one worth noticing.
Of experiencers who described themselves as atheist beforehand, 36.9% still did afterwards — the lowest retention of any group, well below Catholics (57.8%) or Protestants (58.0%). The position least likely to survive the experience is the one that held there was nothing to survive it.
A caution on this one: prior affiliation is self-reported afterwards, in the same sitting as the current answer. Someone recalling their earlier position through the lens of what happened may not recall it as they would have stated it at the time. That applies to every before-and-after item here.
Reports arriving at the experience as an atheist and leaving it unable to hold that position — settling afterwards on uncertainty rather than on any tradition.
Finding Twelve — Belief Before The Experience
Prior belief shifts the odds. It does not decide them.
Everyone states their religion as it was before. Set that against whether a religious figure is named anywhere in the account, and the effect is real — but so is what happens to the people who had no belief at all.
Affiliated beforehand: 42.3%. Unaffiliated: 30.7%. An eleven-point gap, and statistically unambiguous at this sample size.
So belief predicts the rate without gating it. Nearly three in ten prior atheists reach for this vocabulary anyway, against roughly four in ten Catholics. The difference between a lifelong believer and a self-described atheist is fourteen points — smaller than most people would guess, in either direction.
The figure named barely varies. God leads for every group including atheists and the unaffiliated; Jesus and angels follow in the same order throughout. What prior belief changes is how often the vocabulary appears, not which vocabulary is available.
Worth stating: prior religion is reported after the fact, in the same sitting as the account itself. Someone whose beliefs changed may not describe their earlier position as they would have described it at the time.
Finding Thirteen — What Forty Years Does To It
Memory takes less than you would expect, and takes it selectively.
Accounts arrive anywhere from months to half a century after the event. If these were ordinary memories decaying, everything should thin with the gap. Almost nothing does.
Bands: under 10 years (n=1,433), 10–25 (907), 25–40 (869), over 40 (715). The two extremes are shown; the middle bands sit between them.
The light does not fade: 65.8% at under ten years, 68.3% at over forty. Nor does the sense that words fail, nor reported life change. Across four decades of delay these barely move.
What does thin is the interpretive layer. Sudden total understanding falls from a peak of 73.8% to 60.6%. Meeting the dead declines steadily, 48.4% to 38.4%. And past forty years the writing itself contracts — narratives shorten by around 13% and carry fewer distinct elements.
This is the same shape as the decade finding, arrived at from a completely different direction. Whether you sort by when the experience happened or by how long someone waited to write, the sensory core holds and the elaborated narrative erodes. Two independent cuts of the data, one pattern.
Finding Fourteen — What "Abilities Afterwards" Actually Means
Half of them say something changed. Almost none of them tested it.
Around half of experiencers answer yes to being asked whether they have psychic or other unusual capacities they lacked before. It is the largest aftereffect in the corpus after changed values — and the least examined. So here is what the 2,159 people who said yes and then explained actually describe.
Categories overlap — 15% name more than one. Half the answers fit none of them; the median explanation is 26 words, often too brief to classify.
The two commonest are quieter than the word “psychic” suggests. Knowing things without being told (18.9%) and sensing or seeing the dead (17.9%) lead. Foreknowledge of events follows at 9.3%, healing at 7.1%, and picking up other people's feelings or illness at 6.3%.
Further down are the more cinematic claims: hearing guidance (3.7%), seeing auras (2.5%), leaving the body at will (2.4%), and interfering with electrical devices (1.5%). The last of these is a fixture of popular accounts and is among the rarest things anyone actually reports.
Is any of it tested?
So the honest answer is no. Nothing in this archive constitutes a test. The questionnaire asks people whether they have such abilities; it does not ask them to demonstrate one, and no mechanism here could evaluate a demonstration if offered. These are self-reports, collected once, at a distance of years.
What the corpus does show is that 6.9% hedge or actively doubt their own claim in the same breath as making it — attributing it to coincidence, or saying they are unsure whether it counts. That is a small share, but it is not zero, and it is the only internal check the data contains.
Two limits on this section. Half the answers matched none of my categories, so the breakdown describes the classifiable half rather than the whole. And the categories were written after reading a sample, which means they fit this corpus by construction — a different reader would draw different lines. Treat the ranking as sturdier than the exact percentages.
Finding Fifteen — The Part That Gets Left Out
Not all of it is light.
Popular accounts of this subject are overwhelmingly benign — peace, love, reunion. The archive is not.
Frightening experiences are harder to submit and harder to publish, so any archive of volunteered accounts will under-represent them. Even so, they are plainly present, and leaving them out would misdescribe the corpus.
Peace shown alongside for scale. An account can contain both — many do.
That is the closest this data comes to identifying a distressing experience: not merely frightening at the outset, which is common, but frightening without the resolution that usually follows.
Explicitly infernal language is rarer but not marginal. 16.4% of narratives use it somewhere — “hell” appears in 620 accounts, “screaming” in 583, “evil” in 308, with the devil, Satan, demons and the abyss trailing well behind. Out-of-body and related experiences carry by far the most fear (39.0%), roughly double the near-death rate.
Frightening accounts are easy to leave out of a summary — they are less quotable, less shareable, and less like the story most people expect. Omitting them would give a rosier picture than the archive supports, which is why they have a section of their own.
Reports leaving through the top of his head, descending rather than rising, and a frightening encounter far below. Nothing in this account resolves into peace.
Finding Sixteen — Things That Could Have Been Checked
More than one in four, where the experience is about someone else.
A recurring shape in these accounts: the experiencer learns something they say they had no way of knowing, and someone else later tells them it was right.
Every such claim was recorded — what was asserted, how specific it was, and whether anyone is said to have confirmed it. Nothing here is adjudicated. The question is what people claim and in what detail, not whether it happened. The after-death communication archive has now been read in full: 1,588 accounts, every one coded claim by claim.
“Specific” excludes vague impressions. “Confirmed” means the account states that another person verified it.
Across the after-death archive alone, 2,348 claims were recorded. 920 assert third-party confirmation and 581 of those are specific rather than impressionistic. They appear in 440 of 1,588 accounts — 27.7%, with a margin of just ±2.2 points.
The concentration is the finding. After-death accounts carry these at 27.7% against 12.5% for near-death and 9.5% for out-of-body experiences. The claims cluster where the experience is about someone else rather than about the experiencer.
What is actually being claimed
The largest category is now one that did not exist in the first pass. Learning a specific fact the experiencer says they had no way of knowing accounts for 214 of the 581 confirmed claims — more than any other kind. In the earlier coding these were disappearing into a generic “other” bucket; naming the category pulled them out.
Of 542 such claims in total, 259 state that another person later confirmed the fact. And the sharpest version remains the smallest: 77 accounts describe meeting someone the experiencer says they did not know had died, 66 of them asserting the death was later confirmed. Across the full archive, 14.4% of all encounters with the dead are with someone not known to have died — almost exactly the 14.6% the small sample predicted.
Two limits, unchanged by the larger corpus. A third of confirmed claims still fall into “other”, so the category breakdown remains partly provisional even after adding a code for the commonest pattern. And every one of these is a recollection written down later, in which the sequence of learning and confirming is reported by the same person. Nothing in a corpus of volunteered accounts can establish what anyone knew and when. What it can establish is that the pattern is not rare, not confined to one culture, and not going away.
Reports passing outside the building during surgery and watching a helicopter land, a baby being unloaded and carried in — details he could not have seen from the operating table.
If You Remember Only This
Fourteen things the archive says.
Every one of these is a count, not an interpretation. None of them says anything about what caused these experiences.
-
The question shapes the answer
Ask directly about time distortion and 76.6% agree. Let people simply tell their story and 17.6% mention it. The familiar description of a near-death experience may be, in part, a description of the questionnaire that collected it.
-
The frame is rigid, the middle is improvised
Crisis first, return last — holding in 98–100% of accounts containing both. But whether the light comes before or after arriving somewhere is 51%. A coin flip.
-
It happens in the bedroom
Bedroom 25.8%, hospital room 19.9%, tunnel 10.4%, garden 3.2%. Most of this is reported as happening exactly where the person's body already was.
-
Mostly it is family, rarely God
81.1% meet someone. An unidentified presence appears more often than Jesus, God and angels combined, and ordinary speech outnumbers wordless communication more than two to one.
-
Speechlessness predicts a changed life; drama predicts nothing
Saying it cannot be described forecasts changed values at 75.0% against 57.9%. Reaching a border — the most cinematic beat available — predicts it at 66.7% against 66.7%. Identical.
-
The sensations travel; the story does not
Tunnel, light, time distortion and ineffability show no meaningful East–West difference. Meeting the dead, the life review and glimpsing the future are all markedly more Western.
-
Men and women report the same events, a different aftermath
Indistinguishable on leaving the body and the life review. Then a 12.7-point gap on abilities reported afterwards — the largest demographic difference in the corpus.
-
Belief moves in one direction only
73.4% keep the religious stance they arrived with. But of those who move, the traffic runs one way — liberal is the hardest position to leave and the easiest to reach. The lowest retention of any group is atheist, at 36.9%.
-
What fades is the story, not the sensation
Across four decades of delay before writing, the light holds at 65.8% then 68.3%, and so do ineffability and reported life change. Sudden total understanding falls 73.8% to 60.6%, meeting the dead 48.4% to 38.4%. Sort by decade or by delay and the same layer erodes.
-
Prior belief shifts the odds, it does not set them
The affiliated name a religious figure in 42.3% of accounts, the unaffiliated in 30.7%. But 28.4% of prior atheists name one too — most often God, the same figure that leads for every group.
-
Half report new abilities. Almost none were tested
The commonest are quiet ones: knowing things unprompted 18.9%, sensing the dead 17.9%. Electrical interference, a fixture of popular accounts, is among the rarest at 1.5%. Just 3.6% mention verification of any kind.
-
One account in six is frightening and stays that way
15.9% describe fear or terror with no peace and no love anywhere in them. Out-of-body accounts carry the most at 39.0%, roughly double the near-death rate. The benign version of this subject is a selection effect.
-
The claims cluster where the experience is about someone else
27.7% of after-death accounts assert something specific that another person later confirmed — against 12.5% for near-death and 9.5% for out-of-body. Read in full across all 1,588.
-
The life review fell off a cliff
Steady at 57–61% from the 1980s through the 2000s, then 31.8% for experiences in the 2010s. Comparing only accounts written within ten years, the drop is 70.0% to 29.1% in a single decade.
And What It Might Mean
What follows from this, and what does not.
Nine thousand people sat down and wrote out the most significant thing that ever happened to them, for strangers, without being asked to and without anything in return. Whatever else this archive is, it is that first.
What follows is what emerges when all of it is read at once. Counting cannot settle what these experiences were — that was never in its power. What it can do is show where the accounts agree with each other, and where the familiar version of this story does not match what the people who lived it actually wrote.
If you are trying to make sense of your own life
If you are thinking about someone you have lost
Where the counting stops
Every limit here is a limit of arithmetic, not of the accounts. Percentages describe a corpus, not a person: no figure on this page tells any individual what their own experience was, and a rare element is not a less real one. The people who wrote these did not submit data — they described something that changed their lives, and the parts that mattered most to them are frequently the parts that no coding scheme can hold.
Two things follow. Because everyone here chose to write to an archive about near-death experience, none of these rates estimate how common any of this is in the wider world; they describe who wrote in. And where the accounts contradict the popular picture — the ordinary rooms, the plain conversation, the middle of the sequence that turns out to have no fixed order — the thing being corrected is the popular picture, not the people. On every point where the familiar story and the accounts disagree, this page sides with the accounts.
What no method reaches is the thing most of them were actually trying to say. The strongest statistical signal in the entire corpus is that the people who insist their experience cannot be put into words are the ones it changed most. That is as close as counting gets to admitting what it is missing.
Why I Built This
Two fears, and what reading nine thousand accounts did to one of them.
I started reading these for reasons that have nothing to do with statistics, and it seems right to say so plainly rather than let the charts imply I came to this neutrally.
I have two fears. One is losing the people I love. The other is being the one who is lost — becoming, someday, an absence in someone else's life.
Reading these accounts has quieted the first. Not settled it, and not by proving anything. Something about sitting with thousands of people who describe the same threshold, in their own words, from every country and every decade, has made that fear smaller than it was.
The second one I cannot read my way out of. But I hope that what calmed me might reach the people I love — my child especially. That when I am gone, what they carry is not only that I am gone for good, but some part of the hope these strangers were willing to write down.
Where I actually stand
I am agnostic, in the specific sense that I believe there is something beyond us — and that I cannot be argued into a position on what it is, or on whether it has any interest in this life.
Reading these accounts has not resolved that. What it has done is make me hold the first half more firmly than before: that there is a beyond, and that what these thousands of people describe finding there is, overwhelmingly, connection and love.
But the reason I keep going is not to shore that up. It is the opposite. I would rather test the belief than protect it — approach it as a hypothesis, look at what the accounts actually say instead of what I would like them to say, and get close enough to something like a real conclusion that it could turn out I was wrong.
That is the motivation, and it is worth knowing when you read the numbers. It is also why every finding here is reported as it came out, including the ones that cut against a comforting reading — the frightening accounts, the elements that turn out rarer than the story suggests, the abilities almost nobody has tested, and the plain fact that none of this establishes what any of it means.
What I have come to believe from this research
Three things settled in me over the course of it. None of them is a finding. Everything above this section is what the accounts say and can be checked against the data. What follows is only what reading them did to one person, and it should be weighed accordingly.
There is something beyond this, and I do not know what it is. I want to be precise about how little that claims — not a place, not a mechanism, not a promise about who is waiting. Only that the idea of everything simply stopping here has come to seem to me the least likely account on offer. I thought that before this research, which is exactly why I will not present this research as having proved it.
People are good. This is the one I did not expect to change. The internet had me fairly well convinced otherwise. But the scene I kept meeting was this: someone arrives in front of a person they had lost, carrying something they never got to say — an apology, usually, or some version of one — and is told, in one form or another, that it was already alright, and to go back and make something of the time they have left. That is not universal, and this page does not pretend it is; the frightening accounts are documented a few sections above and I have not looked away from them. But forgiveness is the ordinary case in what I read, not the rare one. I have come to think that our capacity to understand each other, to feel with each other, and to forgive is the native condition rather than the achievement.
Love may be the only real thing. This is the hardest to say without sounding like a greeting card, so let me say it plainly. Ask people what the light was, or the being, or the place, and what comes back is very often not a description at all. It is love — at an intensity they say has no equivalent here and no words able to carry it. It is what people reach for after every other word has failed them, which is the part I cannot get past.
What I have come to think is that everything else we feel is a fragment of it. Jealousy is a fear about the absence of being loved. Grief is love with nowhere left to go. Put a white light through a prism and it comes apart into colours, and we spend our lives naming the colours and mistaking them for separate things — when what we are looking at each time is one light, split by whatever it happened to pass through.
I hold all three loosely, and none of them follows from the numbers. A corpus of volunteered accounts can tell you what people report; it cannot tell you what is true beyond the reporting. If you finish this page believing something different, that is an entirely reasonable place to land — and I would genuinely like to hear it.
Before You Go
What did you make of this?
This is an open analysis of an archive that keeps growing, and the questions worth asking next are not obvious from the inside.
Three things are genuinely useful to hear: what struck you, what you did not believe, and what you now want to know that this page does not answer. Several sections here exist because someone asked a question the data could answer and nobody had thought to ask.
One button, one message. It opens your own mail client with a few prompts already in it — answer any of them, ignore the rest, and say at the bottom whether you would like to hear when the analysis is updated. Nothing is collected by this page, there is no tracking on it, and no third party sees what you write. Replies are read personally.
Questions already open
Part Two
Methodology
Everything above is a claim about tens of thousands of pages. Here is how those pages were counted, what was checked, and the five occasions the numbers were confidently wrong before they were right.
What Nearly Went Wrong
Five times the data lied convincingly.
Every number above survived a specific attempt to break it. These are the attempts that succeeded, and what they would have produced if nobody had checked.
Answers with elaboration were being dropped
Many people write "Yes" and then keep going. Counting only the bare word discarded every one of them.
Saw a light: read as 9.7% → actually 62.6%Six questions were never yes/no at all
They offer descriptive alternatives instead, so the word "yes" appears in no valid answer.
Time distortion: read as 0.0% → actually 76.6%Pattern-matching missed most of what it looked for
Automated text matching found only 40% of what human readers found, and its accuracy varied wildly by element — so the ranking between elements was arbitrary.
Retired and replaced by 1,500 read-through judgementsAn age pattern was really a waiting-time pattern
Reports of psychic ability appeared to fall with age at the time of the experience. They actually rose with how long the person had to notice.
Apparent 34-point age gradient → mostly recall intervalPooling three archives invented a pattern
Combining collections with different base rates made encounters with the dead appear to repel every other element. Within a single collection, they are among the strongest pairings.
Simpson's paradox, caught by splitting the corporaNone of these threw an error. Each produced a confident, plausible, publishable number. All five were caught the same way — by a figure being implausible rather than by anything failing.
Still provisional: the claim-type taxonomy, where a third of cases fall into an "other" bucket; the message themes, which are keyword matches never validated against blind reading; and the life-review measure, whose two affirmative options behave so differently that merging them was probably a mistake. These are flagged rather than fixed.
Everything here is descriptive. It reports what tens of thousands of pages say and how often they say it. It makes no claim about what produced the experiences.