The Experience Elicited by Hallucinogens Presents the Highest Similarity to Dreaming within a Large Database of Psychoactive Substance Reports

Sanz and colleagues frame the study around a longstanding observation that certain drug-induced states, notably those produced by classic serotonergic psychedelics, have a ‘‘dreamlike’’ character. Earlier neurophysiological and anecdotal work suggested overlaps between dreaming—particularly REM sleep dreams—and experiences induced by substances such as LSD, psilocybin and DMT, but a rigorous, large-scale, phenomenological comparison of first-person reports was lacking. The study therefore set out to quantify the semantic similarity between large corpora of subjective reports of psychoactive-substance experiences and reports of dreams with differing levels of lucidity. The principal aim was to test whether reports associated with serotonergic psychedelics, and more broadly other hallucinogen classes (dissociatives and deliriants), are more similar to dream reports than reports associated with non-hallucinogenic drug classes, and to explore whether similarity depends on dream lucidity.

The analysis used two large text corpora. Subjective reports of drug experiences were sourced from Erowid's reviewed collection (the Erowid corpus); only single-substance reports and substances with at least 10 reports were included, yielding 165 substances. Dream reports came from Dreamjournal.net (the Dreamjournal corpus), which includes user-rated lucidity; the authors separated dreams rated as low lucidity (score 1) and high lucidity (score 5). The dream corpora comprised 2,914,498 and 716,189 words for low and high lucidity reports, respectively. The authors note that neither database guarantees chemical purity, precise dosing, nor, for the dream reports, that narratives derive from REM sleep. Text preprocessing was performed in Python using NLTK. Documents were tokenised, lowercased, lemmatised and short tokens (<3 characters) removed. To avoid confounding by mentions of substance names or administration routes, a manually compiled list of 12,465 substance- and route-related words was removed from the Erowid corpus so that analysis focussed on experiential language rather than drug identity or logistics. Latent Semantic Analysis (LSA) formed the basis of the comparison. Term frequencies were converted to a tf–idf matrix (terms-by-documents, where each document was either a substance aggregate from Erowid or the collection of high/low lucidity dreams). Terms present in more than 95% or fewer than 5% of documents were excluded. Singular value decomposition (SVD) reduced dimensionality; the authors retained D = 20 largest singular values for the main analyses and assessed stability across other D values. Semantic similarity between each substance's reports and the dream-report sets was computed as the Pearson linear correlation coefficient between the corresponding columns of the reduced term-by-document matrix. Group-wise comparisons used multi-factor ANOVA (drug category and dream lucidity as factors) and post-hoc Wilcoxon signed-rank tests. Stability of rankings across parameter choices was assessed with Spearman correlation.

Across 165 substances, reports associated with hallucinogens showed the greatest semantic similarity to dream reports. For high-lucidity dreams, LSD ranked highest, followed by Lophophora williamsii (peyote, mescaline-containing) and plants of the Datura genus (deliriant tropane alkaloids); the highest-ranked dissociative was PCP. Hallucinogens made up almost 80% of the top-20 substances most similar to high-lucidity dreams. By contrast, the bottom-20 substances (least similar) comprised mainly depressants/sedatives (barbiturates, benzodiazepines), opioids, stimulants and antidepressant/antipsychotics. When similarity was computed against low-lucidity dream reports, deliriants featured more prominently: Datura and Brugmansia genera ranked first and third respectively, and dimenhydrinate (a deliriant at high doses) was among the top-ranked substances for low-lucidity similarity but absent from the high-lucidity top ranks. Melatonin, Silene undulata and Calea zacatechichi showed the largest increases in similarity when comparing high versus low lucidity dream comparisons; alprazolam, sertraline and clonazepam had the largest movement toward less ‘‘lucid’’ similarity. Category-level analysis ranked dissociative psychedelics, serotonergic psychedelics and deliriants highest on average, followed by entactogens, depressants/sedatives, antipsychotic/antidepressants and stimulants. A multi-factor ANOVA showed a significant main effect of drug category on similarity to dreams (F = 31.34, p < 0.0001), while the interaction between drug category and dream lucidity was non-significant (F = 0.02, p = 0.88). Post-hoc Wilcoxon tests indicated significant differences between hallucinogen categories and depressants/sedatives, antipsychotic/antidepressants and stimulants, but not among hallucinogen subclasses themselves. Qualitative inspection of word-frequency patterns (presented as word clouds) revealed that both dream reports and hallucinogen reports frequently included terms related to perception (see, visual, face, colour), emotion (fear), setting/location (outside, inside, street) and relatives (mom, dad, brother). Reports of substances with low similarity to dreams tended to include terms tied to medical use, addiction or preparation logistics rather than perceptual or affective content. The substance rankings were highly stable to the number of retained SVD components: Spearman correlations between rankings across parameter choices exceeded 0.995.

Sanz and colleagues interpret their findings as phenomenological support for the long-standing hypothesis that hallucinogen-elicited experiences resemble dreaming. They organise the discussion around three domains: sensory perception, self-awareness and metacognition. On sensory perception, the researchers note that REM dreams and certain drugs both produce vivid multimodal imagery, but that the nature of visual content may differ by drug class and dose. Serotonergic psychedelics typically induce ‘‘simple’’ visual phenomena (colour, patterns, trails), consistent with effects in early visual cortical areas, whereas deliriants and high-dose dissociatives can produce complex, realistic hallucinations that users often perceive as real—paralleling low-lucidity dreams. The authors link experimentally observed changes (for example, occipital metabolism and altered evoked potentials) to these perceptual effects, and note that dosage and specific pharmacology likely modulate imagery complexity. Regarding neurophysiology, the paper highlights serotonergic 5-HT2A receptor activation as a key mechanism in classical psychedelics, and discusses evidence for altered occipital and medial temporal lobe (MTL) activity together with suppression of posterior cingulate cortex (PCC) function. The authors propose that reduced PCC-mediated inhibition and altered MTL activity could provide a common basis for the dreamlike quality seen in both REM sleep and psychedelic states. They also describe alpha-band power reductions reported after LSD, psilocybin and sub‑anesthetic ketamine and consider how these changes may reflect loss of top‑down inhibition. For deliriants, the pharmacology is anticholinergic, which the investigators note is paradoxical given that dreaming is associated with elevated acetylcholine; electrophysiological data for anticholinergics do not show the same alpha suppression pattern, suggesting a different mechanistic route. On self-awareness, the paper emphasises overlaps in altered bodily self-location, ego‑dissolution and out‑of‑body phenomena. Dissociative psychedelics (e.g., ketamine, PCP) and serotonergic psychedelics both produce disruptions to self-boundaries, and about 7% of freely reported Erowid narratives across several psychedelics explicitly described ego-dissolution, while questionnaire-based studies report higher incidences. Neuroimaging links LSD-induced ego-dissolution to increased global connectivity in fronto-parietal and insular networks, including the temporo-parietal junction, a region implicated in multisensory self‑integration. With respect to metacognition and dream lucidity, the authors discuss how non-lucid dreams and deliriant states share impaired metacognitive function—reduced insight and inability to identify the experience as a dream or drug‑induced state—potentially connected to dorsolateral and frontopolar prefrontal deactivation during REM sleep. Lucid dreaming, by contrast, reflects preserved metacognitive capacity; this may explain why some oneirogen plants (Calea zacatechichi, Silene undulata) increased their relative similarity to high-lucidity dreams, though the authors caution that Erowid reports about these plants often focus on preparation and expectations rather than experiential detail. Finally, the researchers acknowledge methodological limitations inherent to analysing large, uncontrolled online corpora. Erowid reports may contain inaccuracies about substance identity, purity and dosage, and Dreamjournal entries do not guarantee REM sleep origin. The free-report format reduces experimental control and allows expectation effects; still, the authors argue that the large sample sizes mitigate random noise and that the approach is useful for generating population-level phenomenological comparisons and hypotheses for targeted experimental work.

The study developed and applied an LSA-based method to compare the phenomenology of altered states using large public corpora of subjective reports. Applying this comparative-phenomenology approach, Sanz and colleagues found that reports associated with hallucinogens—broadly defined to include serotonergic psychedelics, dissociatives and deliriants—show the greatest semantic similarity to dream reports, with certain deliriants particularly similar to low-lucidity dreams. The authors propose that shared alterations in perception, self-awareness and metacognitive function may underlie this similarity and recommend future controlled studies to probe these phenomenological features and their neurophysiological bases.