The evidence boundary: These sources support related concepts and specific findings. They do not establish synthetic imprinting as a single scientific effect. This is a selected reading guide, not a systematic review.
Memory can change after an event
Misleading information received after an event can affect later memory reports. Studies with fabricated childhood photographs also show why the procedures used to prompt recollection matter. (Loftus and Palmer, 1974; Wade et al., 2002)
How to interpret it
These studies help frame questions about generated images, but their results should not be generalized to every AI image or every viewer.
A question to investigate
Does a generated format change the outcome compared with the same information in ordinary text or images?
Loftus and Palmer (1974)| Research question | Can question wording affect reports of a filmed collision? |
|---|
| Participants | 45 students in Experiment 1; 150 students in Experiment 2. |
|---|
| Design and comparison | Participants watched collision films. Questions used different verbs; Experiment 2 also included a group without a speed question. |
|---|
| Measured findings | Speed estimates varied with wording. In Experiment 2, the “smashed” group more often reported glass that the film did not show. |
|---|
| Timing and follow-up | Speed estimates were collected after viewing. Experiment 2 tested the glass report one week later. |
|---|
| Limits on interpretation | Student samples, short films, and prompted reports limit generalization. This study did not test AI media or effects beyond the measured interval. |
|---|
Published study · Read the full paper (PDF). Experiments I–II, pages 586–588.
Wade and colleagues (2002)| Research question | Can a fabricated personal photograph, together with retrieval prompts, elicit false childhood recollections? |
|---|
| Participants | 20 participants, with family informants helping supply photographs and check the invented event. |
|---|
| Design and comparison | A doctored balloon-ride photograph appeared alongside three genuine photographs. Interviews used repeated prompts and guided imagery. |
|---|
| Measured findings | By the final interview, 10 participants reported partial or clear false recollections of the invented event. |
|---|
| Timing and follow-up | Three interviews across 7–16 days, followed by debriefing. This was repeated exposure during the study, not a test of persistence after correction. |
|---|
| Limits on interpretation | The small sample and combined procedure cannot isolate a photograph-only effect or estimate how often ordinary viewing changes memory. |
|---|
Published study · Read the full paper (PDF). Method and Results, pages 598–600.
- Reconstruction of Automobile Destruction: An Example of the Interaction Between Language and Memory — Elizabeth F. Loftus, John C. Palmer (1974). Journal of Verbal Learning and Verbal Behavior, 13(5), 585–589.
Peer-reviewed research. A foundational experiment showing that wording after an event can affect later reports and recollection.
- A Picture Is Worth a Thousand Lies: Using False Photographs to Create False Childhood Memories — Kimberley A. Wade, Maryanne Garry, J. Don Read, D. Stephen Lindsay (2002). Psychonomic Bulletin & Review, 9, 597–603.
Peer-reviewed research. Examines whether a fabricated childhood photograph, paired with imagery instructions, can contribute to false reports of an event.
Remembering where an experience came from
Source-monitoring research examines how people infer the origin of remembered information. Work by Bonnail and colleagues studies source confusion between physical and virtual experiences. (Johnson et al., 1993; Bonnail et al., 2024)
How to interpret it
A source error can occur even when someone remembers the content. Confusing a source is different from forgetting an event entirely.
A question to investigate
Which source cues help people remember later whether an experience was virtual, generated, imagined, or physical?
Bonnail and colleagues (2024)| Research question | Do people confuse the physical and virtual origins of remembered objects? |
|---|
| Participants | 29 participants at the first memory test; 28 at the second. |
|---|
| Design and comparison | Participants encountered objects across 20 tasks, half in a physical environment and half in VR. Source-identification tests and interviews examined their judgments. |
|---|
| Measured findings | Source errors occurred. Participants described using remembered perceptual cues and assumptions about what VR could render to identify an object’s origin. |
|---|
| Timing and follow-up | Source-identification tests took place one and two weeks after the initial session. |
|---|
| Limits on interpretation | A small study of specified environments and objects does not establish universal confusion, indefinite persistence, or personality change. |
|---|
Published study · Read the full paper (PDF). Method, section 3 and Figure 2; results and discussion.
- Source Monitoring — Marcia K. Johnson, Shahin Hashtroudi, D. Stephen Lindsay (1993). Psychological Bulletin, 114(1), 3–28.
Peer-reviewed review. Develops the source-monitoring framework for understanding how people attribute memories and information to origins.
- Was It Real or Virtual? Confirming the Occurrence and Explaining Causes of Memory Source Confusion Between Reality and Virtual Reality — Elise Bonnail, Julian Frommel, Eric Lecolinet, Samuel Huron, Jan Gugenheimer (2024). Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems.
Peer-reviewed research. Investigates errors in recalling whether objects were encountered in physical or virtual environments.
Deepfakes, uncertainty, and false memories
Experiments examine several outcomes, including uncertainty, trust, and memory. Murphy and colleagues found differences between formats in reported memory and some political measures. Follow-up after debriefing did not detect significant political differences, but small persistent effects could not be ruled out. (Vaccari and Chadwick, 2020; Murphy et al., 2025)
How to interpret it
An immediate response and a persistent effect are different claims. Results from one political example or sample do not settle what all synthetic media will do.
A question to investigate
When do changes persist, and which corrections or explanations help them fade?
Murphy and colleagues (2025)| Research question | How do homemade political deepfakes compare with text and cloned audio? |
|---|
| Participants | 443 Irish residents recruited initially; 300 Prolific participants completed the follow-up. |
|---|
| Design and comparison | An online experiment compared formats of fabricated political stories and measured reported memory, opinions, voting intentions, suspicion, and detection. |
|---|
| Measured findings | Reported false memory was higher for the paid deepfake than for text. Political outcomes varied by story and measure; formats were not equivalent across all outcomes. |
|---|
| Timing and follow-up | Invitations went out one week after debriefing; the survey remained open five days. Previously presented and new false stories both had low memory-report rates. Political comparisons were not statistically significant. |
|---|
| Limits on interpretation | One national context, selected stories, and incomplete follow-up limit generalization. Nonsignificant political differences do not rule out small persistent effects; the authors discuss that possibility. |
|---|
Published study · Read the full paper (PDF). Sections 2–6 and Discussion 7. The paper has inconsistent post-exclusion totals; recruitment and follow-up counts are used here.
- Deepfakes and Disinformation: Exploring the Impact of Synthetic Political Video on Deception, Uncertainty, and Trust in News — Cristian Vaccari, Andrew Chadwick (2020). Social Media + Society, 6(1).
Peer-reviewed research. An experiment examining deception, uncertainty, and trust after exposure to synthetic political video.
- An Average Joe, a Laptop, and a Dream: Assessing the Potency of Homemade Political Deepfakes — Gillian Murphy, Didier Ching, Eoghan Meehan, John Twomey, Aaron Bolger, Conor Linehan (2025). Applied Cognitive Psychology.
Peer-reviewed research. Compares political misinformation formats and follows up after debriefing. Memory and political outcomes varied; nonsignificant follow-up political differences do not rule out small persistent effects.
Realistic faces and perceived trust
Nightingale and Farid tested judgments of a specific set of real and AI-generated faces, including authenticity and perceived trustworthiness. (Nightingale and Farid, 2022)
How to interpret it
How trustworthy a face looks does not tell us whether a person or message is reliable. The study is about judgments in a defined task.
A question to investigate
How do context, source labels, and familiarity affect confidence as well as accuracy?
- AI-Synthesized Faces Are Indistinguishable from Real Faces and More Trustworthy — Sophie J. Nightingale, Hany Farid (2022). Proceedings of the National Academy of Sciences, 119(8).
Peer-reviewed research. Tests participants’ ability to distinguish a set of AI-synthesized faces from real faces and evaluates perceived trustworthiness.
Human responses to characters and environments
Research on anthropomorphism, parasocial interaction, and presence offers ways to study humanlike attribution, perceived connection, and the feeling of being there. (Epley et al., 2007; Horton and Wohl, 1956; Lombard and Ditton, 1997)
How to interpret it
An AI character that replies to a user differs from a one-way broadcast persona. Existing concepts need to be applied with attention to those differences.
A question to investigate
What changes when a generated character remembers context, responds personally, or is encountered repeatedly?
- On Seeing Human: A Three-Factor Theory of Anthropomorphism — Nicholas Epley, Adam Waytz, John T. Cacioppo (2007). Psychological Review, 114(4), 864–886.
Peer-reviewed theory. Proposes a framework for why people attribute humanlike qualities, intentions, or mental states to nonhuman agents.
- Mass Communication and Para-Social Interaction: Observations on Intimacy at a Distance — Donald Horton, R. Richard Wohl (1956). Psychiatry, 19(3), 215–229.
Peer-reviewed theory. Introduces parasocial interaction as a way to describe perceived intimacy with media performers.
- At the Heart of It All: The Concept of Presence — Matthew Lombard, Theresa Ditton (1997). Journal of Computer-Mediated Communication, 3(2).
Peer-reviewed theory. Synthesizes ways researchers use presence to describe the experience of a medium becoming less noticeable or feeling immediate.
Media literacy, corrections, and provenance
Media-literacy guidance encourages analysis and evaluation of messages. Research reviews examine correction of misinformation, while provenance standards describe records of media creation and editing. (National Association for Media Literacy Education, n.d.; Lewandowsky et al., 2012; Coalition for Content Provenance and Authenticity, n.d.)
How to interpret it
A provenance record can describe origin without proving a claim true. A lesson or label should be evaluated for whether people understand and use it.
A question to investigate
Which approaches improve informed judgment without encouraging indiscriminate distrust?
- What Is Media Literacy? — National Association for Media Literacy Education (n.d.). NAMLE.
Reference material. A professional definition and explanation of the skills involved in accessing, analyzing, evaluating, creating, and acting with media.
- Misinformation and Its Correction: Continued Influence and Successful Debiasing — Stephan Lewandowsky, Ullrich K. H. Ecker, Colleen M. Seifert, Norbert Schwarz, John Cook (2012). Psychological Science in the Public Interest, 13(3), 106–131.
Peer-reviewed review. Reviews why misinformation can continue to influence reasoning and how corrections can be designed more effectively.
- C2PA and Content Credentials Explainer, version 2.2 — Coalition for Content Provenance and Authenticity (n.d.). C2PA.
Reference material. Explains signed records of media history and their limits. Provenance information does not determine whether a claim is true.
Find full reference details in the bibliography, or suggest research we should consider.
