Chapter 18

Memory, Confabulation, and Narrative Reliability

A man named Jacques woke up after a serious medical event and told his nurse’s aide something strange. While the aide had been out of the room, Jacques said, the head nurse — Danièle — had just been on the phone. With his biological mother. They had been talking about him. He had no relationship with that woman. He didn’t want them talking about him.

The aide froze. Then she ran to the nurses’ station, where Danièle was hanging up the phone. With Jacques’s biological mother. They had been talking about Jacques. Jacques recited the conversation back, almost word for word. He had been unconscious in another room. He should not have heard a thing. He even told them about the surgeon walking back to the operating theater — with tape on the arms of his glasses. The surgeon, when called, had tape on his glasses.¹

Now imagine standing in that hallway. Marsh would say: don’t trust the memory. Memory is fallible. Stories get edited. Witnesses make mistakes. Patients fill in the blanks. People remember what they want to remember.

I take that seriously. Memory really is fallible. So is observer testimony. But there is a problem with using the fallibility of memory to dismiss every NDE account. If memory were the issue, the details would be wrong. In Jacques’s case, the details were right.

This chapter is about memory. It is about why NDE accounts cannot simply be waved away as confabulation — the brain’s habit of filling gaps with invented material. And it is about why the case Marsh leans on hardest, the famous A. J. Ayer case, actually backfires when you read it carefully.

A. The Critic’s Argument

In chapter 2 of his book, Marsh raises one of his most far-reaching objections. NDE memories, he says, are not trustworthy. The accounts we read in books like Raymond Moody’s Life After Life or Pim van Lommel’s Lancet paper have been filtered, polished, and retold. They may bear little resemblance to what actually happened — if anything happened at all.²

Marsh names several mechanisms.

First, editing. Every story changes when it gets retold. The teller leaves out the dull parts. They sharpen the dramatic ones. Over time the rough edges round off. The story becomes more coherent than the experience ever was.

Second, stylization. NDE accounts get shaped by familiar patterns. The tunnel. The light. The being of love. After hearing other people’s stories, an experiencer may unconsciously fit her own into the same shape. As Marsh puts it, every account will become stylized, stereotyped, and edited in its subsequent tellings and re-tellings.³

Third, observer error and false recall. Marsh leans on the police-investigation literature. Sober, sincere witnesses to ordinary events get details wrong all the time. Memory is reconstructive. The brain fills in missing pieces from expectations and prior knowledge. NDEs happen during medical crises, often under heavy drugs, in profound stress. If memory fails routinely under everyday conditions, why would it suddenly become reliable here?⁴

Fourth, the philosopher’s case. Marsh anchors his memory objection in a single high-profile NDE account: the case of British philosopher A. J. Ayer (Marsh, pp. 28–30). Ayer was a hard-nosed positivist — the founder of British logical positivism, the man who quite literally wrote the book on what counts as a meaningful proposition. In 1988 he choked on a sliver of smoked salmon, suffered a four-minute cardiac arrest, and reported a vivid NDE. Ayer first wrote that the experience had “weakened” his confidence that death is the end. Then he recanted. He insisted his brain had never actually died. He blamed the experience on a malfunctioning physical system.⁵

Marsh treats this as the master argument. Even Ayer — the most evidence-driven philosopher of his generation — recognized that a vivid experience cannot be trusted as evidence of anything beyond the brain. If Ayer would not trust his own NDE memory, why should anyone trust theirs?

Fischer and Mitchell-Yellin push the philosophical edge further. Vividness does not equal truth. People remember dreams vividly. People remember childhood things that never happened. Subjective certainty proves nothing on its own.⁶

Marsh adds one more case to seal the deal. He describes a woman with hysterical paralysis who, during her medical crisis, claimed she had crossed a river, seen her father’s face, heard beautiful music, and was surrounded by angels. But her vital signs had been normal throughout. The crash team confirmed she was never actually in danger. Marsh calls this a fabricated NDE.⁷ His point is sharp. If even one NDE turns out to be confabulation, why trust the rest?

That is the argument we have to answer.

B. Identifying Weaknesses

Three problems. Each one cuts deep.

First, Marsh treats the possibility of memory error as if it were proof of memory error. Memory is fallible. Granted. But “memory can fail” is not the same thing as “this memory has failed.” When a witness reports the make and model of a car, and a security camera confirms it, the camera does not suddenly become unreliable because witnesses elsewhere have been wrong. The same logic applies to NDEs. The relevant question is not can memory go wrong. It is did this memory go wrong. Marsh almost never asks the second question.

Second, the “stylization over time” hypothesis makes a clear empirical prediction — and the prediction has been tested. If NDE accounts get reshaped by retelling, they should drift over time. They should become more elaborate, more polished, more conformist. They should slide toward the cultural template. Marsh writes as if this is obvious. Bruce Greyson, the longtime NDE researcher at the University of Virginia, decided to actually test it. His 2007 study, “Consistency of Near-Death Experience Accounts over Two Decades: Are Reports Embellished over Time?” gathered original NDE narratives and then re-interviewed the same experiencers many years later. The result? No measurable embellishment. Accounts remained remarkably stable in core content, sometimes for thirty years or more.⁸ Marsh’s hypothesis predicted drift. The data shows no drift. The hypothesis fails the test it was given.

Third — and this is the largest problem — Marsh’s memory critique runs parallel to the strongest evidence for NDEs without ever meeting it. He argues that NDE memories might be wrong. Then he turns the page. He does not stop to ask: what would tell us whether they are wrong? The answer is independent verification. Veridical NDEs — cases where experiencers report things they could not have known by ordinary means, and those things turn out to be correct — give us an objective check that pure memory cannot. We treat veridical perception fully in Chapter 4. Janice Holden’s analysis of the published NDE literature found that 92% of out-of-body veridical perceptions were entirely accurate, with only 8% containing any error.⁹ If NDE memories were the product of editing, retelling, and confabulation, the accuracy rate would not be 92%. It would be near zero.

A confabulator does not look at a doctor he has never seen and accurately describe his eccentric red bow tie. A confabulator does not name a surgical instrument no one in the room ever told him about. A confabulator does not see his upper dentures placed by a particular nurse in the upper drawer of a yellow crash cart in the hallway and tell that nurse, days later, exactly where she put them.¹⁰

The memory hypothesis carries a prediction. The data falsifies the prediction. That is not a small problem. That is the kind of finding an honest scientist reports as a refutation. Marsh does not.

C. The Pro-NDE Response

So what should we say about NDE memory? I want to make four points. Each on its own dents Marsh’s argument. Together, they refute it.

1. Many cases are reported immediately upon waking, with no time for stylization.

Marsh’s editing-and-retelling hypothesis depends on time. Stories drift in the telling. Memories accommodate to expectations. But what happens when there is no telling? When the experiencer wakes up and immediately describes the experience — before she has spoken to another NDEr, before she has read a book about it, before she has had time to edit anything?

Take Jacques again. On regaining consciousness he immediately told his nurse’s aide three remarkable things: that nurse Danièle was on the phone with his biological mother (a phone call no one in the room had witnessed); that the surgeon was walking back to the operating room with tape on the arms of his glasses (confirmed when the surgeon was called in); and that he had hovered above his own body during the procedure. The witnesses confirmed every detail in real time.¹¹ There was no months-long shaping of the narrative. There was no slow drift toward a stereotype. There was a man waking up and describing — out loud, to skeptical caregivers — events that had happened in another room while he was unconscious.

Or take the dentures case from van Lommel’s Lancet paper. A nurse removed an unconscious patient’s upper dentures and placed them in the drawer of a crash cart in the hallway. A week later, when the nurse re-entered the patient’s room, the patient looked at her and said, in effect, oh, that nurse knows where my dentures are. He then described the cart, the drawer, and the procedure — all from the perspective of someone watching from above.¹² We treat this case fully in Chapter 4. What matters here is the timing. The nurse’s testimony came not from the patient’s polished retelling but from her own jaw-dropped surprise.

These are not the kinds of cases the editing-and-retelling hypothesis can absorb.

2. NDE memories are unusually stable over time. Greyson’s study tested it.

Bruce Greyson did exactly what a careful scientist should do. He took the editing hypothesis at face value and asked: if NDE accounts get embellished over time, then the same person’s account, taken twenty years apart, should look measurably different.

It did not.

When Greyson and his collaborators compared accounts taken decades apart from the same experiencers, the core content was strikingly stable. The same elements, the same emotional contours, the same details. No measurable accretion of dramatic flourishes. No drift toward a culturally expected template.¹³ Long’s data tells a parallel story. NDErs interviewed an average of twenty-two years after their experiences described the same encounters with the same content, and they held the same convictions about what they had seen.¹⁴

Memories of imagined experiences fade. Memories of fabricated experiences shift, drift, accumulate detail or lose it. NDE memories do not. They behave like memories of significant real events, not like memories of dreams or hallucinations.

Notice what this finding does to Marsh’s framing. He treats “memory is unreliable” as a general truth that applies to NDE accounts by default. Greyson reverses the burden. He shows that NDE memories, as a class, behave more reliably than ordinary autobiographical memories. They are not the suspicious outliers Marsh assumes. They are the unusually solid ones. And no theory that depends on memory drift can survive that result, because there is no drift to work with.

3. Brain imaging confirms it. The Thonnard study.

A 2013 study by Marie Thonnard and colleagues at the University of Liège went a step further. They used a tool called the Memory Characteristics Questionnaire to compare NDE memories with three other kinds of memory: memories of real events surrounding a coma; memories of imagined events; and memories of other emotionally significant events.

Their finding was striking. NDE memories did not look like memories of imagined events. They looked more like memories of real events than even ordinary memories of significant real events did. They scored higher on sensory detail, self-referential content, and emotional vividness.¹⁵

We treat the vividness question more fully in Chapter 21, and I want to be careful here. Vividness alone is not proof of veridicality. Fischer and Mitchell-Yellin are right about that much. But Thonnard’s study is not measuring vividness in isolation. It is measuring a recognized fingerprint of real-event memory. NDE memories carry that fingerprint. Confabulated memories do not.

4. The A. J. Ayer case actually undercuts Marsh’s argument.

This is the case Marsh leans on hardest. Look closely.

Ayer’s NDE happened in 1988. He published “What I Saw When I Was Dead” shortly afterward, describing the experience and admitting it had weakened his lifelong conviction that death is final. Then he recanted, in a follow-up piece, insisting his brain had not actually died. Marsh treats this recantation as the deciding blow.

There are two things wrong with that reading.

First, Ayer’s recantation was not a memory revision. He never said the experience had not happened. He never said he had remembered it wrong. He said his philosophical interpretation of it had been too generous. The change was not in the data. It was in the framework brought to the data — driven, as anyone familiar with Ayer’s career can see, by his deep prior commitment to materialism. That is exactly the kind of post-hoc reasoning Marsh accuses NDE researchers of, but he never names it when one of his own witnesses does it.

Second, and more striking. Ayer’s attending physician, Dr. Jeremy George, has stated that Ayer privately confided to him after the recantation that he had seen a Divine Being and would have to revise his various books and opinions. Ben Rogers, in his careful biography A. J. Ayer — A Life, treats this report seriously.¹⁶ If George’s account is accurate, Ayer’s public recantation was driven not by honest memory revision but by concern for his philosophical legacy. The private Ayer was less certain than the public one.

Marsh cites the public recantation. He does not engage the private one.

Set the Ayer case alongside the larger pattern, and a clear picture emerges. NDE memories are stable over decades. They show the neural fingerprint of real-event memory. They are corroborated, often immediately, by independent witnesses. They get the verifiable details right with stunning accuracy. None of these features fits a confabulation. All of them fit a real experience honestly recalled.

D. Counter-Objections

Two objections deserve a serious response.

Objection 1: Even immediate reports could be confabulated. The brain “fills in” the gap as soon as consciousness returns. The patient pieces a story together in the moment of waking. Rapid timing does not rule confabulation out.

This sounds plausible until you ask what the prediction is. A brain piecing fragments together at the moment of awakening should produce confused, jumbled, or random reports. It should not produce coherent narratives that contain accurate, third-party-verifiable details about events the patient could not have witnessed. Jacques did not wake up rambling. He woke up correcting his caregivers about a phone call he could not have heard. The dentures patient did not wake up describing a generic room. He pointed to a specific drawer in a specific cart in a specific hallway. The rapid-confabulation idea has the same problem as the slow-confabulation idea. Both predict the wrong outputs. NDE memories are too organized, too detailed, and too specifically right for any confabulation hypothesis to absorb.¹⁷

Objection 2: Witnesses might unconsciously confirm what the patient says, especially when emotions run high. The doctor wants the patient to be right. The nurse wants the story to be remarkable. Confirmation bias creeps in.

A real concern. We treat researcher bias fully in Chapter 19. Two things to note here. First, the strongest cases involve facts the witnesses did not know until the patient told them — the dentures cart, the surgeon’s tape, the contents of a phone call in another room. Witnesses cannot unconsciously confirm what they do not yet know. Second, prospective hospital studies (van Lommel, Parnia, Sabom) collect details immediately and document them in writing, with multiple independent observers.¹⁸ That is precisely the design that makes confirmation bias hardest to sustain.

The memory objection is not crazy. It is just not enough. NDEs are not vague impressions floating in the haze of recovering brains. They are stable, specific, externally verified, and neurally distinct from imagined events. The leap from “memory is fallible” to “every NDE is confabulation” is a leap the data does not justify. The cases — when you actually look at them — refute it.

Marsh has done his job by pressing the question. We honor him by answering it carefully. The answer, as the evidence stands today, is clear. Memory cannot dissolve veridical NDEs. The dentures stay where the patient said they were. The phone call really did happen. The bow tie was red. And Jacques really did know what Danièle was saying — from another room, while his eyes were closed and his consciousness was somewhere else entirely.