The Generation Effect

Snap Shot

• The question: Is information a person produces in his or her own mind remembered differently from information the person only reads, hears, or sees, and what does that mean for interviews and testimony?
• What the research shows: Material people generate is generally remembered better than the same material presented to them, an advantage of almost half a standard deviation across studies. Witnesses who are pressed to answer questions about events they did not see can later come to remember their own answers as part of the event, especially after confirming feedback and longer delays.
• What it does not show: The advantage is for remembering the content, not for remembering where it came from; generation can leave memory for source and context unchanged or worse. Forced confabulation effects come mostly from laboratory studies of filmed events, are smaller than those for voluntary guesses, and cannot show that any particular detail in a real account was generated rather than observed.
• The law: Rule 702 and Houser govern this testimony in federal and military courts, and Ariz. R. Evid. 702, Chapple, and Nordstrom in Arizona; an expert may explain how question formats can introduce self-generated content but may not opine on whether a witness is truthful (Lindsey).
• Issues for each side: Prosecutors can show that a witness’s own free narrative is a strength and that interviews which allowed “I don’t know” protect accuracy; defense counsel can examine whether a disputed detail first appeared as a guess, inference, or speculative answer the witness was asked to produce, without treating every later detail as confabulated.

The generation effect is the finding that people remember information better when they produce it themselves than when they simply receive it. A person who completes the word pair “hot – c___” by supplying “cold” will, on average, remember “cold” better than a person who read the completed pair. It has been replicated in hundreds of experiments and illustrates the principle that memory depends on what a person does with information, not only on what the person was shown.

The effect reaches court in two ways. The first concerns the ordinary benefit of active recall: a witness who retrieves and describes an event in his or her own words is engaging in the kind of effortful processing that tends to strengthen memory. The second concerns a less obvious consequence. Witnesses generate not only memories but also inferences, guesses, and answers to questions they cannot actually answer. Because self-produced content is memorable, and because memory for where content came from is a separate process, a detail the witness once inferred or guessed can later be remembered, and reported in good faith, as something the witness observed.

This paper reviews the classic research, the two large meta-analyses, the explanations offered for the effect, and its boundary conditions. It then turns to source monitoring, forced confabulation, and interview questions that ask witnesses to generate content, and closes with the limits of the evidence, the law, and issues for each side.

The Classic Finding and the Meta-Analyses

Slamecka and Graf (1978) gave the effect its name. Across five experiments, participants either read word pairs or generated the second word of each pair from a cue and a rule, such as producing an associate, a synonym, an opposite, or a rhyme from the first word and an initial letter. Generated words were remembered better than read words on both recognition and recall tests, and the advantage held across the different generation rules. Because the generated and read words were the same words, the difference could be attributed to the act of producing them rather than to the material itself.

Two meta-analyses have since summarized the literature. Bertsch et al. (2007) combined 445 effect sizes from 86 studies and found an average generation advantage of .40, a benefit of almost half a standard deviation, with substantial variation depending on how the studies were designed. McCurdy et al. (2020) conducted a larger meta-analysis and theoretical review of 126 articles reporting 310 experiments and 1,653 estimates. They examined memory for the generated items themselves and, separately, memory for context, meaning details associated with the item such as where or how it was presented. They tested seven prominent theories against the data and also found that generation constraint, meaning how tightly the task limits what a person can produce, significantly changed the size of the effect.

Why Generating Helps

No single explanation has accounted for every finding, and the current view is that several processes contribute. One family of accounts holds that generating requires a person to search existing knowledge and to work out how a cue and a target relate, so that generated items receive both more item-specific processing and more processing of relationships. McDaniel et al. (1988) proposed a three-factor account along these lines, noting that generation can strengthen processing of the individual item, of the cue–target relationship, and in some designs of the list as a whole. Another account emphasizes mental effort. In four recent experiments using pupil dilation as an independent measure of effort, generating information drew more effort than reading it, and differences in effort partly accounted for the memory advantage when the same participants both generated and read (Grudzien & Unsworth, 2026).

A third line of research shows that the advantage depends on how well people process the material they read. Generating beat reading only when the comparison group’s reading was shallow, as when participants merely said each word aloud; when that group formed a mental image of each word, the two conditions produced similar memory (Begg et al., 1991). Generation is thus one reliable route to the active, meaningful processing that produces durable memory, which links it to the levels-of-processing and self-reference effects described elsewhere in this hub.

Boundary Conditions

The effect has well-documented limits. It depends on existing knowledge. In four experiments using pronounceable but meaningless letter strings, producing them conferred no memory benefit, and in two more the benefit was much smaller for unfamiliar sentences than for familiar sayings (Lutz et al., 2003). The design of the study also matters. Generation effects are often smaller or absent when one group of participants generates and a different group reads, rather than the same participants doing both, and in one series of experiments intentional learning instructions eliminated the effect in free recall (McDaniel et al., 1988).

The limit with the greatest forensic significance concerns memory for source and context. In a mock conversation, three on-screen faces either asked participants questions about various topics or showed them statements about those topics. Answering the questions improved memory for the topics but made participants worse at recalling which face had raised each one, a negative generation effect for source memory (Jurica & Shimamura, 1999). A 12-experiment series reached a similar conclusion: producing a word reliably made the word itself easier to remember, gave no benefit for remembering surrounding details such as location or background color, and made one detail, the ink color of the produced word, harder to remember (Mulligan, 2004). Generation, in short, makes content more memorable without making its circumstances, which people rely on to separate what they saw from what they thought, any easier to remember.

Generated Versus Received Information

The source monitoring framework explains how people decide where a memory came from. Memories do not carry labels identifying their origins; people attribute a memory to a source by evaluating its characteristics, such as perceptual detail, contextual information, and traces of the mental operations that produced it, and they err more often when sources are similar or when encoding was impoverished (Johnson et al., 1993; Lane & Houston, 2021, pp. 61–66). The earlier reality monitoring account addressed the specific task of distinguishing memories of perceived events from memories of events that were thought, imagined, or inferred (Johnson & Raye, 1981), as the Source Monitoring paper in this hub describes.

The generation research bears directly on this task. A witness who sees a person reach into a jacket may generate the thought that the person was reaching for a weapon. The generation effect predicts that such self-produced inferences will be well remembered, and the source and context findings suggest that the witness’s memory for having inferred them, rather than having seen or heard them, may be no better and possibly worse. Research on imagination inflation shows a related pattern: imagining childhood events increased confidence that they had occurred (Garry et al., 1996), and repeatedly imagining simple actions led participants to remember performing actions they had only heard described or had never encountered (Goff & Roediger, 1998). Research on memory for language points the same way. When people later recalled sentences, a substantial share of what they reported was an inference the sentence invited rather than what it stated (Brewer, 1977), and in a 1975 laboratory study of listeners to mock courtroom testimony, many remembered implied facts as if they had been asserted, even after a warning against doing so (Harris et al., 1975). These studies concern inferences drawn from words heard or read in the laboratory, not inferences about a witnessed event, and inferences are often correct.

Forced Confabulation

The most direct forensic test of self-generated content comes from research on forced confabulation, in which witnesses are required to answer questions about events that did not occur. In an early study, first graders, third and fourth graders, and college students watched a movie clip and were pressed to answer questions about events that plainly had not happened in it. Although the evidence indicated that participants would not have answered without being pushed, one week later participants in every age group showed false memories for details they had knowingly made up, and the children were more susceptible than the adults (Ackil & Zaragoza, 1998). In face-to-face interviews with adults, Zaragoza et al. (2001) found that forced confabulations produced false memories a week later and that confirmatory feedback from the interviewer, such as telling the participant that an answer was correct, increased those false memories, increased confidence in them, and increased the likelihood that participants would freely report the confabulated events one to two months later. When participants were forced to fabricate entire fictitious events, they strongly resisted doing so and showed limited false memory after one week, but after eight weeks they freely reported their fabrications nearly half the time, including some they had correctly rejected at one week (Chrobak & Zaragoza, 2008).

Other findings temper these results. When some participants could answer “don’t know” and others could not, forced guesses were less likely than volunteered answers to be remembered later, although confidence in a repeated answer to an unanswerable question rose over time in both conditions; pressing witnesses to answer, especially repeatedly, increased errors without increasing correct recall (Pezdek et al., 2007). A cognitive interview conducted before the forced fabrication reduced later false memory for the fabricated events, while the same interview conducted afterward did not undo it (Memon et al., 2010; Lane & Houston, 2021, pp. 130–131). A registered-report meta-analysis confirmed that forced confabulation can produce false memories, that confirmatory feedback appears to strengthen the effect, and that longer delays are needed before false memories for whole events emerge (Riesthuis et al., 2023). The same meta-analysis cautioned that voluntary confabulation produced more false memories than forced confabulation, that the effect was observed mainly when the same participants contributed to both conditions, and that part of the effect may arise because unanswerable questions themselves carry misinformation.

Interview Questions That Ask Witnesses to Generate Content

These findings identify question formats that ask a witness to generate content rather than to retrieve it: requests to speculate or give a best guess, questions that presuppose a detail the witness did not report, repeated questions after the witness has said he or she does not know, invitations to reconstruct what someone “must have” said or done, and feedback that confirms an answer as correct. None guarantees error, and a generated answer may be accurate; the research indicates only that such answers may later be reported without the qualification that accompanied them.

The generation research also supports the opposite side of the interviewing question. A free narrative in the witness’s own words is itself an act of retrieval, and early, open-ended recall tends to reduce later forgetting without reducing accuracy (Lane & Houston, 2021, pp. 129–130). Interview approaches that encourage complete reporting while permitting “I don’t know,” discouraging guessing, and avoiding feedback on the content of answers are consistent with both lines of research. The forced confabulation studies concern witnesses; they did not test suspects, and readers interested in suggestion during suspect interviews are referred to the False Confessions resources.

Limits of the Evidence

Most generation research tests university students’ memory for word lists, word pairs, or arithmetic problems over minutes to days. Effect sizes vary widely with design, constraint, materials, and the type of test, so the meta-analytic average does not describe any particular situation. The forced confabulation studies come closer to forensic conditions, but they use filmed events, low stakes, and participants who know they are in a study, and the meta-analysis found the effect concentrated in within-subject designs. No study can determine whether a particular detail in a particular witness’s account began as an inference or a guess. That question turns on evidence of what the witness was asked, what the witness said, and when.

Questions to Ask About This Research

When the figure of .40 from Bertsch et al. (2007) is cited, counsel can ask what was compared (generated versus read items in laboratory tasks) and whether it bears on memory for source, which the research suggests it does not. When forced confabulation research is offered, counsel can ask how large the effect is and under what conditions it appeared, since it was smaller than for voluntary guesses and concentrated in within-subject designs (Riesthuis et al., 2023), and whole-event false memories emerged mainly after eight weeks (Chrobak & Zaragoza, 2008). Counsel can ask whether the conclusion follows for this witness, which depends on whether the record shows the witness was actually pressed to answer, given feedback, or asked to speculate, and whether the disputed detail first appeared in that setting. Counsel can also ask whether the interview allowed “I don’t know” (Pezdek et al., 2007). These questions are explained in “Reading Research Claims: A Guide for Attorneys” on this site.

The Generation Effect and the Law

Federal and Military Standards

Expert testimony about memory and interviewing is governed by Federal Rule of Evidence 702, as amended December 1, 2023, under which the proponent must show the court that it is more likely than not that the rule’s requirements are met, including that the opinion reflects a reliable application of reliable principles and methods to the facts. Daubert v. Merrell Dow Pharmaceuticals, Inc. (1993) assigned the trial judge a gatekeeping role, and Kumho Tire Co. v. Carmichael (1999) extended that obligation to all expert testimony, including testimony based on technical or other specialized knowledge. In courts-martial, Military Rule of Evidence 702 tracks the federal rule, and the 2023 amendment became part of the military rule in 2025 through Military Rule of Evidence 1102 (Joint Service Committee on Military Justice, 2025); practitioners should confirm the current text. Military judges evaluate proffered expert testimony under the six factors of United States v. Houser (1993).

Arizona

Arizona Rule of Evidence 702 adopted the federal formulation and the Daubert standard effective January 1, 2012, and was amended effective January 1, 2024, to conform to the 2023 federal amendment (Ariz. Sup. Ct. Order R-23-0004). State v. Chapple (1983) is Arizona’s leading case on expert testimony concerning the reliability of memory and eyewitness accounts; excluding the expert was an abuse of discretion on its facts, while admission generally remains within the trial court’s discretion, and State v. Nordstrom (2001) approved limiting such an expert to general factors without opinions on the accuracy of a particular witness. Experts may not opine on whether a particular witness is telling the truth (State v. Lindsey, 1986, 149 Ariz. at 475), so testimony on this topic should describe how generated content can enter an account rather than declare that a detail was confabulated.

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. Active retrieval in a witness’s own words tends to strengthen memory, and an early, open-ended account is among the most valuable forms of evidence a witness can provide. Forced confabulation effects depend on specific conditions, including pressure to answer questions the witness cannot answer, confirming feedback, and long delays; where the record shows an interview that permitted “I don’t know,” avoided speculation, and gave no feedback on content, those conditions are absent. The research also shows that people resist fabricating events and that forced answers are, on average, less likely to be retained than volunteered ones. Prosecutors can protect a witness’s account by ensuring interviews are recorded and witnesses are not asked to guess.

The research also counsels caution. A detail that first appears as an answer to a speculative, presumptive, or repeated question, particularly one followed by confirmation, carries a recognized risk of being self-generated, and the witness’s later confidence in it does not resolve that risk. Inferences a witness drew at the time, such as about intent, identity, or what an object was, can be remembered as observations, so testimony should distinguish what the witness saw from what the witness concluded. Treating growing certainty about a detail first offered as a guess as corroboration goes beyond the research.

Special Issues for Defense

The research supports several propositions defense counsel can legitimately present. People remember content they produce better than its source, and witnesses who were pressed to answer questions about events they did not see have later reported their own answers as memories, with more confidence when an interviewer confirmed them. Removing the option to say “I don’t know” and repeating questions increased errors in laboratory studies without improving correct recall. Expert testimony on these principles, combined with a record of when and how a disputed detail first appeared, can legitimately help the factfinder evaluate it.

The research also counsels caution about the limits that apply to the defense. Forced confabulation effects are smaller than voluntary-guess effects, emerge mainly under specific laboratory conditions, and have not been shown to apply to every question asked in an interview. A detail added in a later account is not by that fact a confabulation; new details in later interviews are often accurate (Lane & Houston, 2021, p. 129), and inferences are often correct. An expert cannot tell the factfinder that a particular detail was generated rather than observed. The same principles apply to the accused, whose account may include inferences or answers he or she was pressed to produce during questioning.

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

Ackil, J. K., & Zaragoza, M. S. (1998). Memorial consequences of forced confabulation: Age differences in susceptibility to false memories. Developmental Psychology, 34(6), 1358–1372. https://doi.org/10.1037/0012-1649.34.6.1358

Begg, I., Vinski, E., Frankovich, L., & Holgate, B. (1991). Generating makes words memorable, but so does effective reading. Memory & Cognition, 19(5), 487–497. https://doi.org/10.3758/BF03199571

Bertsch, S., Pesta, B. J., Wiscott, R., & McDaniel, M. A. (2007). The generation effect: A meta-analytic review. Memory & Cognition, 35(2), 201–210. https://doi.org/10.3758/BF03193441

Brewer, W. F. (1977). Memory for the pragmatic implications of sentences. Memory & Cognition, 5(6), 673–678. https://doi.org/10.3758/BF03197414

Chrobak, Q. M., & Zaragoza, M. S. (2008). Inventing stories: Forcing witnesses to fabricate entire fictitious events leads to freely reported false memories. Psychonomic Bulletin & Review, 15(6), 1190–1195. https://doi.org/10.3758/PBR.15.6.1190

Garry, M., Manning, C. G., Loftus, E. F., & Sherman, S. J. (1996). Imagination inflation: Imagining a childhood event inflates confidence that it occurred. Psychonomic Bulletin & Review, 3(2), 208–214. https://doi.org/10.3758/BF03212420

Goff, L. M., & Roediger, H. L., III. (1998). Imagination inflation for action events: Repeated imaginings lead to illusory recollections. Memory & Cognition, 26(1), 20–33. https://doi.org/10.3758/BF03211367

Grudzien, A. M., & Unsworth, N. (2026). Investigating a mental effort explanation of the generation effect using pupillometry. Memory & Cognition, 54(3), 980–996. https://doi.org/10.3758/s13421-025-01791-0

Harris, R. J., Teske, R. R., & Ginns, M. J. (1975). Memory for pragmatic implications from courtroom testimony. Bulletin of the Psychonomic Society, 6(5), 494–496. https://doi.org/10.3758/BF03337547

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

Johnson, M. K., & Raye, C. L. (1981). Reality monitoring. Psychological Review, 88(1), 67–85. https://doi.org/10.1037/0033-295X.88.1.67

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

Jurica, P. J., & Shimamura, A. P. (1999). Monitoring item and source information: Evidence for a negative generation effect in source memory. Memory & Cognition, 27(4), 648–656. https://doi.org/10.3758/BF03211558

Lane, S. M., & Houston, K. A. (2021). Understanding eyewitness memory: Theory and applications. New York University Press.

Lutz, J., Briggs, A., & Cain, K. (2003). An examination of the value of the generation effect for learning new material. The Journal of General Psychology, 130(2), 171–188. https://doi.org/10.1080/00221300309601283

McCurdy, M. P., Viechtbauer, W., Sklenar, A. M., Frankenstein, A. N., & Leshikar, E. D. (2020). Theories of the generation effect and the impact of generation constraint: A meta-analytic review. Psychonomic Bulletin & Review, 27(6), 1139–1165. https://doi.org/10.3758/s13423-020-01762-3

McDaniel, M. A., Waddill, P. J., & Einstein, G. O. (1988). A contextual account of the generation effect: A three-factor theory. Journal of Memory and Language, 27(5), 521–536. https://doi.org/10.1016/0749-596X(88)90023-X

Memon, A., Zaragoza, M., Clifford, B. R., & Kidd, L. (2010). Inoculation or antidote? The effects of cognitive interview timing on false memory for forcibly fabricated events. Law and Human Behavior, 34(2), 105–117. https://doi.org/10.1007/s10979-008-9172-6

Mulligan, N. W. (2004). Generation and memory for contextual detail. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30(4), 838–855. https://doi.org/10.1037/0278-7393.30.4.838

Pezdek, K., Sperry, K., & Owens, S. M. (2007). Interviewing witnesses: The effect of forced confabulation on event memory. Law and Human Behavior, 31(5), 463–478. https://doi.org/10.1007/s10979-006-9081-5

Riesthuis, P., Otgaar, H., Bogaard, G., & Mangiulli, I. (2023). Factors affecting the forced confabulation effect: A meta-analysis of laboratory studies. Memory, 31(5), 635–651. https://doi.org/10.1080/09658211.2023.2185931

Slamecka, N. J., & Graf, P. (1978). The generation effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4(6), 592–604. https://doi.org/10.1037/0278-7393.4.6.592

Zaragoza, M. S., Payment, K. E., Ackil, J. K., Drivdahl, S. B., & Beck, M. (2001). Interviewing witnesses: Forced confabulation and confirmatory feedback increase false memories. Psychological Science, 12(6), 473–477. https://doi.org/10.1111/1467-9280.00388

Legal Authorities

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. 702.

Mil. R. Evid. 702.

Mil. R. Evid. 1102.

Ariz. R. Evid. 702.

Ariz. Sup. Ct. Order R-23-0004 (amending Ariz. R. Evid. 702, effective Jan. 1, 2024).

Related topic papers: Source Monitoring · Imagination Inflation · The Misinformation Effect · The Levels-of-Processing Effect · The Self-Reference Effect · Improving Eyewitness Memory

Related topics from Arizona Forensic Services: False Confessions: Suggestibility