Snap Shot
- The question: How accurately do people remember what was said in a conversation, including who said it, in what words, and in what order, and what does that mean for testimony about threats, consent, confessions, and admissions?
- What the research shows: People generally retain the meaning of what was said far better than the exact words, and memory for the gist of a conversation is often fairly good. Exact wording is often lost quickly, but remarks that carry social weight or are unexpected, such as an insult, a joke, or sexual language, are retained better than neutral content. Errors cluster in understanding, in who said what, in whether something was said or only thought, in sequence, and in reports by interviewers of how they questioned someone; contemporaneous notes and, especially, recordings preserve far more.
- What it does not show: The research does not show that any particular witness misremembered a particular statement, that a confident report of exact words is false, or that a remembered admission or threat was never made. Most studies use laboratory conversations with low stakes.
- The law: A party’s own statements offered against that party are not hearsay (Fed. R. Evid., Mil. R. Evid., and Ariz. R. Evid. 801(d)(2)), so their accuracy is tested through cross-examination, recordings, and corroboration; in courts-martial, admissions also require corroboration (Mil. R. Evid. 304(c)). Expert testimony is governed by Rule 702, Houser, Chapple, and Nordstrom.
- Issues for each side: Prosecutors can rely on the relative durability of gist and on any recording or contemporaneous note, but should not present a remembered paraphrase as the defendant’s exact words; defense counsel can show where wording, source, or sequence matters and is fragile, but should not argue that memory for what was said is generally unreliable, which the research does not support.
Much of what witnesses report in court is something they heard rather than saw: a threat, a refusal, an apology, an admission, an order, or a warning. Remembered statements are offered to prove intent, consent, knowledge, and guilt, and many conversations are never recorded, so the only evidence of what was said is what the participants, or someone who overheard them, later remember.
Memory for conversation follows the general rules of memory, but it has features of its own. Because spoken language must be interpreted before it can be stored, a listener can remember only what the listener understood, and understanding depends on context, shared knowledge, and the speaker’s clarity as well as the listener’s attention. Errors of this kind are usually sincere. A witness who reports the gist of a conversation in good faith may still have the words, the speaker, or the order wrong, and a witness whose account is vague on wording may still be accurate about what mattered.
This paper summarizes research on gist and verbatim memory for what was said, understanding, memory for who said what, threats, consent, confessions and admissions, and recorded versus recalled conversations, followed by the limits of the evidence and the governing federal, military, and Arizona law. The general theory of gist and verbatim memory and its link to false memory are covered in the companion paper on gist memory.
Gist and Verbatim Memory for What Was Said
A classic experiment showed how quickly exact wording fades. Listeners heard passages and were tested on whether a sentence had appeared exactly as heard. When the test came almost immediately, they noticed changes in grammatical form, such as a switch from active to passive voice, as well as changes in meaning; after only a short stretch of further passage, they were much poorer at noticing changes of form but still detected changes of meaning well (Sachs, 1967). Reviews of discourse research confirm that surface wording is rarely retained for long and is replaced by a representation of what the speaker meant (Davis & Friedman, 2007).
Exact wording is not always lost. In a study of a natural conversation among colleagues, remarks that carried information about the speaker’s attitudes or relationship with the listener, such as jokes, put-downs, and personal evaluations, were remembered word for word much better than neutral content, and studies of dialogue have reported similar effects (Keenan et al., 1977; Bates et al., 1978). Unexpected wording, such as profanity, sexual language, or a rude remark, draws attention and is kept more precisely than conventional phrasing, which listeners process with little thought (Davis & Friedman, 2007).
The distinction matters because legal questions often turn on exact words. Davis and Friedman (2007) contrast an angry vow to kill someone, which is a threat, with an angry remark that the same person has no right to live, which a factfinder might treat as opinion. A listener may take both as meaning the same thing, store that shared meaning, and later report either version, and the chance of that substitution rises if the person was in fact harmed. The same research also points to a strength. Memory for the central meaning of an exchange is generally much more durable than memory for its wording, so a witness who cannot reproduce exact words may still accurately report that a threat was made or that a request was refused.
Understanding Comes First
A listener cannot remember a meaning the listener never grasped. Comprehension can fail at hearing the sounds, tracking complex sentences, identifying whom and what the speaker is talking about, recognizing what the speaker intends, and reading what a remark implies about the relationship (Davis & Friedman, 2007). It depends on common ground, the knowledge and context speaker and listener share, and speakers often leave out referents, change topics abruptly, or speak indirectly. Remarks that seem irrelevant to the question asked are understood and remembered poorly; in one study, participants asked to recall an evasive reply often recorded only that the speaker had not answered the question (Davis & Holtgraves, 1984).
Hints, polite requests, and sarcasm are particularly open to misreading, and interpretation is shaped by expectations and motives. Remarks heard in a particular setting are later remembered as more typical of that setting in style (Brewer & Hay, 1984), and a speaker believed at the time of listening to hold high status is later credited with having spoken more forcefully, an effect that did not appear when status was learned only afterward (Holtgraves et al., 1989). Knowledge of how events turned out can also reshape memory: in a laboratory study described by Davis and Friedman (2007), participants who learned that a man in a recorded argument had been charged with killing his girlfriend were more likely both to remember his threats correctly and to falsely remember threats and violence that had not occurred.
Overhearers face added difficulty because they lack the common ground the participants share. In a study of 60 participants recalling 60 conversations after a brief delay, overhearers recalled significantly less than speakers and addressees, and what they recalled was less accurately attributed to the person who said it (Brown-Schmidt et al., 2025). Participants also tend to remember their own contributions better than their partner’s (Diachek & Brown-Schmidt, 2024; Davis & Friedman, 2007).
Who Said What: Source and Reality Monitoring
Remembering the content of a statement is different from remembering where it came from. Source monitoring is the term for how people decide where a remembered item came from, such as a particular speaker, conversation, or time (Johnson et al., 1993). These judgments tend to succeed for distinctive exchanges that received full attention and to fail when many similar conversations run together, as with a supervisor’s or investigator’s routine interviews. Davis and Friedman (2007) list ten kinds of conversational source error, among them who said something, to whom, in which conversation, in what order, who else was present, and whether it came from a conversation at all.
In “who said what” experiments, listeners tend to confuse statements by people of the same gender, race, or other social category, especially when older or distracted (Davis & Friedman, 2007). In a study reviewed in the same chapter, memory for who asked a question was poor and fell to chance after a week, while memory for who answered was better. A 2023 series of laboratory experiments reported a related pattern: items named in answers to questions were later recognized better than items named in the questions, and people seeking information remembered the answers better than people giving information remembered the questions they had been asked (Zormpa et al., 2023). If the pattern extends to interviews and interrogations, which has not been tested directly, answers may be remembered better than the questions that prompted them.
Reality monitoring, the distinction between what was said and what was only thought, planned, or imagined, raises a separate issue. Participants made more false reports of having made statements when they were shown pictures that made the topic feel familiar, an effect that grew over time (Buehler & Ross, 1993). In studies in which participants planned to say a phrase but were prevented from saying it, they often reported with confidence that they had said it (Parks, 1997). A witness may therefore sincerely recall having given a warning, refused, or disclosed something that was only intended. Memory for order and for which conversation an exchange occurred in is generally weaker than memory for content, and negated statements can lose their negation in memory over time (Davis & Friedman, 2007).
Threats, Consent, Confessions, and Admissions
Threats illustrate the gap between gist and wording. A listener who was frightened may retain the threatening meaning well and the exact words less well, and an incomplete or ambiguous remark may later be remembered as its expected ending, especially if the speaker was angry and harm followed (Davis & Friedman, 2007). Emotionally arousing exchanges are often remembered well in their core content, while peripheral details such as when or where something was said, or who else was present, are more prone to error (Davis & Friedman, 2007).
Consent and refusal are communicated partly through indirect speech and nonverbal cues, which are among the forms of communication most open to misunderstanding. Davis and Friedman (2007) observe that misperceptions of what another person’s remarks imply about the relationship are often at issue in sexual harassment and sexual assault cases, and that the gist later remembered and reported may be a conclusion (for example, that the other person wanted to proceed) rather than what was said. The same principle applies to every party: a complainant’s memory of having refused and an accused person’s memory of having perceived agreement are both memories of meaning, and the research does not indicate which is more likely to be accurate in a given case, nor that refusals are generally ambiguous. The Counterintuitive Testimony hub on this site addresses consent communication and misperception directly.
Confessions and admissions raise two memory questions: what the suspect said, and how the questioner elicited it. Research on interrogation found that implied promises and threats conveyed by pragmatic implication were understood much like explicit ones (Kassin & McNall, 1991). When 16 experienced officers interrogated innocent mock suspects in covertly recorded sessions, their written reports often omitted or understated confrontation, maximization, leniency, and false evidence, and lay readers of a report rather than a transcript saw less pressure and more often judged the suspect guilty; the authors noted the brief, low-stakes task (Kassin et al., 2017). Studies of adults recounting their own interviews with children show a similar pattern. Mothers remembered the gist of what their preschool children said much better than the wording, had difficulty remembering whether statements were spontaneous or prompted, and often accepted altered transcripts as accurate (Bruck et al., 1999), and experienced interviewers recalled much of the substance of what children told them yet often could not recall their own leading questions (Warren & Woodall, 1999). This research concerns how a statement was obtained; it does not show that a reported admission was not made.
Recorded Versus Recalled Conversations
Notes taken during or soon after a conversation help, but they are incomplete. When 20 investigative interviews with children were compared with audio recordings, the investigators’ “verbatim” notes omitted 57% of the interviewers’ own utterances and 25% of the incident-relevant details the children provided, and they attributed many details to more open-ended questions than had actually been asked (Lamb et al., 2000). A 2023 study of conversational memoranda found that participants who wrote a summary shortly after a conversation recalled more of it a week later than their partners who did not, but the content each recalled was similarly and largely accurate; only 4.7% of details were recalled by both partners after a week (Brown-Schmidt et al., 2023). Note-taking thus appears to increase how much is remembered rather than how accurately it is remembered, and two honest participants may remember largely different parts of the same exchange.
A recording preserves words, sequence, and tone, but it may not capture gestures, earlier conversations that give a remark its meaning, or what happened before or after recording. Differences between a recording and a witness’s memory are an occasion to examine what the witness understood, not necessarily evidence of dishonesty.
Pretext calls and post-incident messages, common in sexual assault cases, preserve content but still require interpretation in context; the Counterintuitive Testimony hub addresses what can fairly be inferred from apologies, non-denials, and denials in such communications.
Limits of the Evidence
Research on memory for conversation is thinner than research on memory for faces and events; Davis and Friedman (2007) described it as largely neglected by witness researchers, and many source errors they catalogued had not been studied directly. Much of the work uses students, scripted dialogues, or brief laboratory conversations. Recent studies use real conversations (Brown-Schmidt et al., 2023, 2025; Diachek & Brown-Schmidt, 2024), but the stakes are low and the delays are days rather than months. Group findings describe tendencies and conditions; they cannot establish whether a particular witness remembered a particular statement correctly.
Questions to Ask About This Research
When Sachs (1967) or later work is cited to show that witnesses do not retain exact words, counsel should ask what exactly was counted: those studies measured recognition of changes in sentence form, not whether a listener could report that a threat or refusal occurred. When Lamb et al. (2000) or Kassin et al. (2017) is cited to challenge an investigator’s account of an interview, counsel should ask whether the conclusion follows for this interview: the studies show that notes and reports omit questions and tactics, not that a particular reported statement was not made, and whether a recording exists usually answers the question better than any study. When Brown-Schmidt et al. (2023) is cited to show that partners remember largely different parts of a conversation, counsel should ask who was studied and what the stakes were, since participants were students in low-stakes conversations recalling them after a week. These questions are explained in “Reading Research Claims: A Guide for Attorneys” on this site.
Memory for Oral Communications and the Law
Federal and Military Standards
A lay witness’s account of a conversation must rest on what the witness personally heard (Fed. R. Evid. 602). An out-of-court statement offered to prove the truth of what it asserts is hearsay and is inadmissible unless an exclusion or exception applies (Fed. R. Evid. 801(c), 802); exceptions include statements of the declarant’s then-existing state of mind (Fed. R. Evid. 803(3)). A party’s own statements offered against that party are not hearsay at all: under Fed. R. Evid. 801(d)(2), an opposing party’s statement, including one the party adopted or one made by the party’s agent or co-conspirator, is excluded from the definition of hearsay, and Mil. R. Evid. 801(d)(2) and Ariz. R. Evid. 801(d)(2) are to the same effect. A witness’s memory of what a defendant said is therefore usually admissible, and its accuracy is tested by cross-examination, recordings, and other evidence. Testimonial statements of a nontestifying declarant also implicate the Confrontation Clause (Crawford v. Washington, 2004). Expert testimony on memory for conversation is governed by Rule 702, as amended December 1, 2023, Daubert v. Merrell Dow Pharmaceuticals, Inc. (1993), and Kumho Tire Co. v. Carmichael (1999), which extended the gatekeeping obligation to all expert testimony, including technical or other specialized knowledge.
In courts-martial, an admission or confession of the accused may be considered on guilt only if independent evidence tending to establish its trustworthiness has been admitted, although that evidence need raise only an inference of truth and corroboration is not required for statements made before or at the time of the act (Mil. R. Evid. 304(c)). Expert testimony is assessed under Military Rule of Evidence 702 and the six factors of United States v. Houser (1993); through Military Rule of Evidence 1102, the 2023 federal amendment became part of the military rule in 2025 (Joint Service Committee on Military Justice, 2025), and practitioners should confirm current text.
Arizona
Arizona Rule of Evidence 801(d)(2) parallels the federal rule on statements of an opposing party. Arizona Rule of Evidence 702 adopted the Daubert standard effective January 1, 2012, and was amended effective January 1, 2024, to track the 2023 federal amendment. State v. Chapple (1983), Arizona’s leading case on eyewitness expert testimony, held exclusion of a qualified memory expert an abuse of discretion on its facts; admission otherwise remains within the trial court’s discretion, and an expert may explain general factors affecting memory but may not opine on the accuracy of a particular witness (State v. Nordstrom, 2001). More broadly, experts may not tell jurors whether a particular witness is truthful (State v. Lindsey, 1986, 149 Ariz. at 475). No Arizona appellate decision addressing expert testimony specifically on memory for conversation was located.
Special Issues in Legal Practice
The principles described above apply with equal force to every party in a legal proceeding. The science operates the same way regardless of who is asking the questions, and the research summarized here does not by itself favor either side. What differs between the parties is not the science but the vantage point — the particular places where these issues tend to surface, and the particular errors each party is best positioned to guard against. The two summaries below identify those recurring issues. They are offered for orientation, not as litigation strategy; in every case the controlling question is what the evidence in that case will, and will not, support, and the expert’s role is to inform the trier of fact rather than to advocate for either side.
Special Issues for Prosecutors
The research supports several propositions prosecutors can legitimately present. Memory for the meaning of what was said is generally far more durable than memory for wording, so a witness who cannot repeat exact words may still accurately report that a threat, refusal, or admission was made. Statements that carry emotional or social weight, such as insults and sexual remarks, tend to be remembered more precisely than routine talk, and the content people recall from real conversations is largely accurate even when it is incomplete. Prosecutors can strengthen this evidence by preserving recordings, recording interviews and interrogations in full, and documenting when, and to whom, a witness first reported a statement.
The research also counsels caution. A remembered paraphrase should not be presented as the defendant’s exact words when the words matter, as they often do for threats and for the difference between an opinion and a statement of intent. An investigator’s account of an unrecorded interview is likely to omit questions and tactics, and research suggests that factfinders who see only such a report may perceive less pressure than occurred. The fact that a party’s statement is admissible does not establish that it was accurately remembered, and in courts-martial an admission must also be corroborated.
Special Issues for Defense
The research supports several propositions defense counsel can legitimately present. Exact wording is often lost quickly, except for remarks that carry social weight or are unexpected; understanding depends on context and common ground; and sincere witnesses can confuse who said something, in which conversation, in what order, and whether it was said or only thought. Overhearers recall less and attribute statements less accurately than participants, and interviewers often fail to remember or record their own questions and the tactics they used. Where a case turns on precise words, sequence, or how a statement was elicited, expert testimony on these principles may assist the factfinder in evaluating what was said.
The research also counsels caution about the limits that apply to the defense. Memory for gist is often good, and the research does not support an argument that remembered statements are generally unreliable or that a reported admission was probably not made. An expert cannot tell the factfinder that a particular witness misremembered a statement. The same principles apply to the accused’s own account of a conversation, including a memory of having perceived consent or of having said something exculpatory, and to defense witnesses who report what a complainant said.
This paper is for educational purposes only and is not legal advice. The author is a psychologist, not an attorney; summaries of law are general and must be verified against current authority or with a licensed attorney.
References
Bates, E., Masling, M., & Kintsch, W. (1978). Recognition memory for aspects of dialogue. Journal of Experimental Psychology: Human Learning and Memory, 4(3), 187–197. https://doi.org/10.1037/0278-7393.4.3.187
Brewer, W. F., & Hay, A. E. (1984). Reconstructive recall of linguistic style. Journal of Verbal Learning and Verbal Behavior, 23(2), 237–249. https://doi.org/10.1016/S0022-5371(84)90159-2
Brown-Schmidt, S., Jaeger, C. B., Evans, M. J., & Benjamin, A. S. (2023). MEMCONS: How contemporaneous note-taking shapes memory for conversation. Cognitive Science, 47(4), Article e13271. https://doi.org/10.1111/cogs.13271
Brown-Schmidt, S., Jaeger, C. B., Lord, K., & Benjamin, A. S. (2025). Remembering conversation in group settings. Memory & Cognition, 53(4), 1037–1054. https://doi.org/10.3758/s13421-024-01630-8
Bruck, M., Ceci, S. J., & Francoeur, E. (1999). The accuracy of mothers’ memories of conversations with their preschool children. Journal of Experimental Psychology: Applied, 5(1), 89–106. https://doi.org/10.1037/1076-898X.5.1.89
Buehler, R., & Ross, M. (1993). How do individuals remember their past statements? Journal of Personality and Social Psychology, 64(4), 538–551. https://doi.org/10.1037/0022-3514.64.4.538
Davis, D., & Friedman, R. D. (2007). Memory for conversation: The orphan child of witness memory researchers. In M. P. Toglia, J. D. Read, D. F. Ross, & R. C. L. Lindsay (Eds.), The handbook of eyewitness psychology: Vol. 1. Memory for events (pp. 3–52). Lawrence Erlbaum Associates.
Davis, D., & Holtgraves, T. (1984). Perceptions of unresponsive others: Attributions, attraction, understandability, and memory of their utterances. Journal of Experimental Social Psychology, 20(5), 383–408. https://doi.org/10.1016/0022-1031(84)90034-9
Diachek, E., & Brown-Schmidt, S. (2024). Linguistic features of spontaneous speech predict conversational recall. Psychonomic Bulletin & Review, 31(4), 1638–1649. https://doi.org/10.3758/s13423-023-02440-w
Holtgraves, T., Srull, T. K., & Socall, D. (1989). Conversation memory: The effects of speaker status on memory for the assertiveness of conversation remarks. Journal of Personality and Social Psychology, 56(2), 149–160. https://doi.org/10.1037/0022-3514.56.2.149
Johnson, M. K., Hashtroudi, S., & Lindsay, D. S. (1993). Source monitoring. Psychological Bulletin, 114(1), 3–28. https://doi.org/10.1037/0033-2909.114.1.3
Joint Service Committee on Military Justice. (2025). Amendments to the Military Rules of Evidence since 1984 (G. E. Maggs, Ed.; 2025 ed.). https://jsc.defense.gov/Portals/99/MRE%20Amendments%202025.pdf
Kassin, S. M., Kukucka, J., Lawson, V. Z., & DeCarlo, J. (2017). Police reports of mock suspect interrogations: A test of accuracy and perception. Law and Human Behavior, 41(3), 230–243. https://doi.org/10.1037/lhb0000225
Kassin, S. M., & McNall, K. (1991). Police interrogations and confessions: Communicating promises and threats by pragmatic implication. Law and Human Behavior, 15(3), 233–251. https://doi.org/10.1007/BF01061711
Keenan, J. M., MacWhinney, B., & Mayhew, D. (1977). Pragmatics in memory: A study of natural conversation. Journal of Verbal Learning and Verbal Behavior, 16(5), 549–560. https://doi.org/10.1016/S0022-5371(77)80018-2
Lamb, M. E., Orbach, Y., Sternberg, K. J., Hershkowitz, I., & Horowitz, D. (2000). Accuracy of investigators’ verbatim notes of their forensic interviews with alleged child abuse victims. Law and Human Behavior, 24(6), 699–708. https://doi.org/10.1023/A:1005556404636
Parks, T. E. (1997). False memories of having said the unsaid: Some new demonstrations. Applied Cognitive Psychology, 11(6), 485–494. https://doi.org/10.1002/(SICI)1099-0720(199712)11:6<485::AID-ACP479>3.0.CO;2-Y
Sachs, J. S. (1967). Recognition memory for syntactic and semantic aspects of connected discourse. Perception & Psychophysics, 2(9), 437–442. https://doi.org/10.3758/BF03208784
Warren, A. R., & Woodall, C. E. (1999). The reliability of hearsay testimony: How well do interviewers recall their interviews with children? Psychology, Public Policy, and Law, 5(2), 355–371. https://doi.org/10.1037/1076-8971.5.2.355
Zormpa, E., Meyer, A. S., & Brehm, L. E. (2023). In conversation, answers are remembered better than the questions themselves. Journal of Experimental Psychology: Learning, Memory, and Cognition, 49(12), 1971–1988. https://doi.org/10.1037/xlm0001292
Legal Authorities
Crawford v. Washington, 541 U.S. 36 (2004).
Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993).
Kumho Tire Co. v. Carmichael, 526 U.S. 137 (1999).
State v. Chapple, 135 Ariz. 281, 660 P.2d 1208 (1983).
State v. Lindsey, 149 Ariz. 472, 720 P.2d 73 (1986).
State v. Nordstrom, 200 Ariz. 229, 25 P.3d 717 (2001).
United States v. Houser, 36 M.J. 392 (C.M.A. 1993).
Fed. R. Evid. 602.
Fed. R. Evid. 702.
Fed. R. Evid. 801.
Fed. R. Evid. 802.
Fed. R. Evid. 803(3).
Ariz. R. Evid. 702.
Ariz. R. Evid. 801(d)(2).
Mil. R. Evid. 304(c).
Mil. R. Evid. 702.
Mil. R. Evid. 801(d)(2).
Mil. R. Evid. 1102.
Related topic papers: Gist Memory · Earwitness Memory · Source Monitoring · External Sources of Memory Distortion · Reconstruction and Recall · Memory Expert Testimony
Related topics from Arizona Forensic Services: Counterintuitive Testimony · False Confessions · Courts-Martial